01 Jmri - 2017 PDF

ri
/jm
m
co
VOLUME 45 | NUMBER 1
.
ing
JANUARY 2017
E
CM earn
l
a lth
he
JOURNAL OF MAGNETIC RESONANCE IMAGING
li ey
w
ww.
w
JMRI
VOLUME 45 | NUMBER 1 | JANUARY 2017 | PAGES 1– 312
STEAM-DTI APPLIED TO BREAST TISSUE. AXIAL AND RADIAL DIFFUSIVITY PARAMETRIC MAPS OF A HEALTHY FGT
VOLUNTEER, AGE: 28 YEARS OLD; AT THE SHORTEST AND THE LONGEST DIFFUSION TIMES (tD). PARAMETRIC MAPS
ARE SHOWN BY DIRECT OVERLAY ON CORRESPONDING SLICES OF T1-WEIGHTED VIBE ANATOMICAL IMAGES FOR
ONE SLICE. FROM THE ARTICLE BY TERUEL ET AL. (PP 84-93)
EDITOR-IN-CHIEF
Mark E. Schweitzer
JMRI JOURNAL OF MAGNETIC RESONANCE IMAGING
Officers of the Society ISMRM Advertising Policy

President Garry E. Gold, MD According to the advertising policy of the International Society for Magnetic Resonance in
Past President James G. Pipe, PhD Medicine, the information describing radiologic equipment, products, and services offered by
Vice President Daniel K. Sodickson, MD, PhD vendors in their advertisements should be ethical, educational, professional, and of value in fur-
thering the radiologic evaluation and treatment of patients and furthering and education of the
Vice President-Elect Pia C. Maly Sundgren, MD, PhD radiologists and allied scientists. The guidelines below should be followed for advertising in
Past President Peter Jezzard, PhD JMRI.
Secretary Margaret A. Hall-Craggs, MD n All advertising must be directed to radiologic diagnosis, therapy, research, communica-
Treasurer Clare Tempany-Afdhal, MD tion, and/or to the professional and managerial roles of radiologists and allied scien-
Program Chair Scott B. Reeder, MD, PhD tists. Advertising material should be consistent with the highest standards of
professionalism.
n The ISMRM reserves the right to reject copy which, in its opinion, is not in good taste, is unprofes-
2016 Corporate Members sional, or is inappropriate for a radiology publication, or offers products or services not in the best
interest of the patient or the Society.
Gold Corporate Members
n Acceptance of advertising does not mean endorsement by the ISMRM of the products
GE Healthcare or services advertised, the advertisers, or the claims made.
Philips Healthcare For media kits and questions
Siemens about advertising, contact:
Stephen Donohue, Advertising Sales Executive
Silver Corporate Member
Tel.: 781-388-8511; Fax: 781-388-8265
Toshiba
E-mail: sdonohue@wiley.com
Bronze Corporate Members

JMRI is indexed in:
Bracco Biotechnology & Bioengineering Abstracts (ProQuest)/CSA Biological Sciences
Bruker Database (ProQuest)/Current Contents: Clinical Medicine (Thomson Reuters)/Embase
Hitachi Medical Systems (Elsevier)/EMCare (Elsevier)/HEED: Health Economic Evaluations Database
United Imaging Healthcare (UIH) (Wiley-Blackwell)/IBIDS: International Bibliographic Information on Dietary Supplements
(NIH)/Index Medicus/MEDLINE (NLM)/INSPEC (IET)/
Journal Citation Reports/Science Edition (Thomson Reuters)/MEDLINE (NLM)/RECAL
Associate Corporate Members Legacy Database (RECAL)/Science Citation Index (Thomson Reuters)/
Nova Medical, Inc. Science Citation Index Expanded (Thomson Reuters)/SCOPUS (Elsevier)/SIIC Databases
(Sociedad Iberoamericana de Informacion Cientifica)/Web of Science
ZMT Zurich MedTech AG (Thomson Reuters)
Disclaimer The Publisher, ISMRM, and Editors cannot be held responsible for
errors or any consequences arising from the use of information contained in
this journal; the views and opinions expressed do not necessarily reflect those
of the Publisher, ISMRM, and Editors, neither does the publication of advertise-
ments constitute any endorsement by the Publisher, ISMRM, and Editors of
the products advertised.
Journal of Magnetic Resonance Imaging, (Print ISSN 1053-1807; Online ISSN 1522-2586), is published monthly on behalf of the International Society of Magnetic Resonance
in Medicine by Wiley Periodicals, Inc., a Wiley Company, 111 River St., Hoboken, NJ 07030-5774. Periodical Postage Paid at Hoboken, NJ and additional offices. Postmaster:
Send all address changes to Journal of Magnetic Resonance Imaging, Journal Customer Services, John Wiley & Sons Inc., 350 Main St., Malden, MA 02148-5020. Back issues:
Single issues from current and prior year volumes are available at the current single issue price from cs-journals@wiley.com. Earlier issues may be obtained from Periodicals Service
Company, 11 Main Street, Germantown, NY 12526, USA. Tel: +1 518 537 4700, Fax: +1 518 537 5899, Email: psc@periodicals.com. Delivery Terms and Legal Title: Where the
subscription price includes print issues and delivery is to the recipient’s address, delivery terms are Delivered at Place (DAP); the recipient is responsible for paying any import duty
or taxes. Title to all issues transfers FOB our shipping point, freight prepaid. We will endeavour to fulfil claims for missing or damaged copies within six months of publication,
within our reasonable discretion and subject to availability. Send subscription inquiries in care of: John Wiley & Sons, Inc., Attn: Journals Admin Dept UK, 111 River Street, Hoboken,
NJ 07030. Advertising inquiries: should be addressed to: Stephen donohue, advertising sales executive, Tel.: 781-388-8511; Fax: 781-388-8265 E-mail: sdonohue@wiley.com. For advertising
inquiries outside the United States, contact Jackie Sibley, Non-Subscription Sales manager, c/o John Wiley & Sons, Ltd., Baffins Lane, Chichester, West Sussex PO19 1UD, United King-
dom; 44 (0) 1234 770351, fax, 44 (0) 1234 770432. Information for subscribers: JMRI is published in 12 issues per year. Institutional subscription prices for 2017 are: Print & Online:
US$3650 (US, Canada, Mexico), e 2519 (Europe), £1992 (UK), US$3899 (Rest of the World). Prices are exclusive of tax. Asia-Pacific GST, Canadian GST and European VAT will be applied
at the appropriate rates. For more information on current tax rates, please go to www.wileyonlinelibrary.com/tax-vat. The price includes online access to the current and all online back files
to January 1st 2012, where available. For other pricing options, including access information and terms and conditions, please visit www.wileyonlinelibrary.com/access. Cancellations: Sub-
scription cancellations will not be accepted after the first issue has been mailed. Production questions: may be directed to jmri@wiley.com; journal submissions: may be sent to http://mc.manu-
scriptcentral.com/jmri. Journal of Magnetic Resonance Imaging accepts articles for Open Access publication. Please visit http:// olabout.wiley.com/WileyCDA/Section/id-406241.html for further
information about OnlineOpen. Journal Customer Services: For ordering information, claims and any enquiry concerning your journal subscription please go to www.wileycustomerhelp. com/
ask or contact your nearest office. Americas: Email: cs-journals@wiley.com; Tel: +1 781 388 8598 or +1 800 835 6770 (toll free in the USA & Canada). Europe, Middle East and Africa: Email:
cs-journals@wiley.com; Tel: +44 (0) 1865 778315. Asia Pacific: Email: cs-journals@wiley.com; Tel: +65 6511 8000. Japan: For Japanese speaking support, Email: cs-japan@
wiley.com; Tel: +65 6511 8010 or Tel (toll-free): 005 316 50 480. Visit our Online Customer Get-Help available in 6 languages at www.wileycustomerhelp.com. Commer-
cial Reprints: E-mail: corporatesaleseurope@wiley.com; corporatesalesusa@wiley.com; or corporatesalesaustralia@wiley.com. Reprint Rep at Wiley: Emily Kaine
(ekaine@wiley.com); Author Reprints (50--- 500 copies): order online at: http://www.sheridanreprints.com/orderForm.html and E-mail: cjones@tsp.sheridan.com.
Change of Address: Please forward the subscriptions address listed above 6 weeks prior to move; please enclose a copy of the mailing label or cite your subscrip-
tion reference number in order to expedite handling. Missing copies will be supplied when losses have been sustained in transit and where reserve stock permits.
Send claims in care of John Wiley & Sons, Inc., Attn: Journals Admin Dept UK, 111 River Street, Hoboken, NJ 07030. Wiley’s Corporate Citizenship initiative seeks
to address the environmental, social, economic, and ethical challenges faced in our business and which are important to our diverse stakeholder groups. Since
launching the initiative, we have focused on sharing our content with those in need, enhancing community philanthropy, reducing our carbon impact, creating
global guidelines and best practices for paper use, establishing a vendor code of ethics, and engaging our colleagues and other stakeholders in our efforts. Follow
our progress at www. wiley.com/go/citizenship. Access to this journal is available free online within institutions in the developing world through the HINARI initiative
with the WHO. For information, visit www.healthinternetwork.org. Printed in the USA by Cenveo Publisher Services. Copyright and Photocopying V C 2016 Wiley
Periodicals, Inc. All rights reserved. No part of this publication may be reproduced, stored or transmitted in any form or by any means without the prior permission
in writing from the copyright holder. Authorization to photocopy items for internal and personal use is granted by the copyright holder for libraries and other users
registered with their local Reproduction Rights Organisation (RRO), e.g. Copyright Clearance Center (CCC), 222 Rosewood Drive, Danvers, MA 01923, USA
(www.copyright.com), provided the appropriate fee is paid directly to the RRO. This consent does not extend to other kinds of copying such as copying for general
distribution, for advertising or promotional purposes, for creating new collective works or for resale. Special requests should be addressed to:permissionsuk@wiley.
com. View this journal online at wileyonlinelibrary.com/journal/JMRI.
An official Journal of the International Society for Magnetic Resonance in Medicine
Editor-in-Chief Mark E. Schweitzer, MD

Stony Brook, NY, USA
Editors Emeritus C. Leon Partain, MD, PhD

Nashville, TN, USA
Gary D. Fullerton, PhD
Denver, CO, USA
Deputy Editors
Abdomen Imaging Evidence-Based Practice Physics

Elizabeth Hecht, MD Matt McInnes, MD Pelin Aksit Ciris, PhD
New York, NY, USA Ottawa, Ontario, Canada Antalya, Turkey & Boston, MA, USA
Gary Glover, PhD
Body Imaging
Stanford, CA, USA
Hersh Chandarana, MD Musculosketal Imaging
New York, NY, USA Bruno Madore, PhD
Garry E. Gold, MD Boston, MA, USA
Shahid M. Hussain, MD, PhD Stanford, CA, USA
Stony Brook, NY, USA Eric Sigmund, PhD
Ravinder Regatte, MD New York, NY, USA
Tamotsu Kamishima, MD, PhD New York, NY, USA
Sapporo, Japan Michael L. Wood, PhD, MBA
Toronto, Ontario, Canada
Daniel Moses, MD
Neuroimaging
Sydney, New South Wales, Safety
Australia William G. Bradley, MD, PhD
San Diego, CA, USA Tim Leiner, MD, PhD
Val M. Runge, MD Utrecht, Netherlands
Bern, Switzerland Kevin Chan, PhD
Pittsburgh, PA, USA Statistical Editors
Breast Imaging Marco Essig, MD James Babb, PhD
Linda Moy, MD Winnipeg, Manitoba, Canada New York, NY, USA
New York, NY, USA Feng Feng, MD Franco Momoli, PhD
Katja Pinker-Domenig, MD Beijing, China Ottawa, Ontario, Canada
Vienna, Austria
Systems Engineering
Cardiovascular Imaging New Developments and Maxim Zaitsev, PhD
Hildo J. Lamb, MD, PhD Future Direction Freiburg, Germany
Leiden, Netherlands Martin Prince, MD, PhD
New York, NY, USA Thoracic Imaging
Ruth Lim, MD
Melbourne, Victoria, Australia Edwin van Beek, MD, PhD
Edinburgh, Scotland, UK
Reza Nezafat, PhD
Boston, MA, USA Oncologic Imaging
Women’s Imaging
Paul E. Sijens, PhD
Gautham Reddy, MD, MPH Caroline Reinhold, MD
Groningen, Netherlands
Seattle, WA, USA Montreal, Quebec, Canada
Book Reviews
Diego R. Martin, MD, PhD
Tucson, AZ, USA
Continuing Medical Education
Mustafa R. Bashir, MD
Durham, NC, USA
Social Media
Jaron Chong, MD
New Haven, CT, USA
Associate Editors
Ronald Arildsen, MD Robert Edelman, MD Michael E. Moseley, PhD Neil M. Rofsky, MD

Nashville, TN, USA Evanston, IL, USA Stanford, CA, USA Dallas, TX, USA
N. Cem Balci, MD Diego Jaramillo, MD, MPH Ponnada A. Narayana, PhD Maythem Saeed, DVM, PhD
St. Louis, MO, USA Philadelphia, PA, USA Houston, TX, USA San Francisco, CA, USA
Bobby Kalb, MD Hajime Sakuma, MD
Alberto Bizzi, MD David G. Norris, PhD
Tucson, AZ, USA Nijmegen, Netherlands Tsu, Mie, Japan
Milano, Italy
Angelo Konstas, MD, PhD Richard Semelka, MD
David A. Bluemke, MD, PhD Stony Brook, NY, USA Tomohisa Okada, MD, PhD
Kyoto, Japan Chapel Hill, NC, USA
Baltimore, MD, USA
Harald Kramer, MD Frank G. Shellock, PhD
Jeff W. M. Bulte, PhD Munich, Germany Catherine M. Owens, MD Los Angeles, CA, USA
Baltimore, MD, USA Walter Kucharczyk, MD London, England, UK
Daniel K. Sodickson, MD, PhD
Geoffrey D. Clarke, PhD Toronto, Ontario, Canada Dennis Parker, PhD New York, NY, USA
San Antonio, TX, USA Denis Le Bihan, MD, PhD Salt Lake City, UT, USA Clare Tempany, MD
Orsay, France Boston, MA, USA
John V. Crues, MD Cynthia Paschal, PhD
Los Angeles, CA, USA Martin O. Leach, PhD Nashville, TN, USA Mark A. van Buchem, MD, PhD
Surrey, England, UK Leiden, Netherlands
Albert de Roos, MD Kim Butts Pauly, PhD
Robert Lenkinski, PhD Steve C.R. Williams, PhD
Leiden, Netherlands Stanford, CA, USA
Dallas, TX, USA London, England, UK
Andre Duerinckx, MD, PhD David R. Pickens, PhD
Debiao Li, PhD Ian R. Young, PhD
Dallas, TX, USA Nashville, TN, USA
Los Angeles, CA, USA Hertfordshire, England, UK
Jeffrey L. Duerk, PhD Christine H. Lorenz, PhD Stephen J. Riederer, PhD William T. C. Yuh, MD
Cleveland, OH, USA Frederick, MD, USA Rochester, MN, USA Columbus, OH, USA
Consultants
Stanley Baum, MD Tom Brady, MD John C. Gore, PhD Roderic Pettigrew, MD, PhD
Philadelphia, PA, USA Boston, MA, USA Nashville, TN, USA Bethesda, MD, USA
Matthew A. Bernstein, PhD R. Nick Bryan, MD, PhD Herbert Y. Kressel, MD Felix W. Wehrli, PhD
Rochester, MN, USA Philadelphia, PA, USA Boston, MA, USA Philadelphia, PA, USA
Managing Editors
Sara Welliver
Review Article
11 Practical Applications of Balanced Steady-State Free-Precession (bSSFP) Imaging

in the Abdomen and Pelvis
Nicola Schieda, Inga Isupov, Andrew Chung, Niamh Coffey, and Leonard Avruch
21 Whole-Body PET/MRI for Colorectal Cancer Staging: Is It the Way Forward?

Dong Ho Lee and Jeong Min Lee
Commentary
36 Guidelines for Documentation and Consent for Nonclinical, Nonresearch MRI in

Human Subjects
Scott B. Reeder, Vera Kimbrell, Titti Owman, Michael Steckner, and Fernando Calamante,
on behalf of the ISMRM Safety Committee
Original Research
Musculoskeletal 42 MRI and the Distribution of Bone Marrow Fat in Hip Osteoarthritis
Jennifer. S. Gregory, Rebecca J. Barr, Victor Varela, Trevor S. Ahearn, Jennifer Lee Gardiner,
Fiona J. Gilbert, Thomas W. Redpath, James D. Hutchison, and Richard M. Aspden
51 Quantifying Metal-Induced Susceptibility Artifacts of the Instrumented Spine at

1.5T Using Fast-Spin Echo and 3D-Multispectral MRI
S. Sivaram Kaushik, Robin Karr, Matthew Runquist, Cathy Marszalkowski, Abhishiek Sharma,
Scott D. Rand, Dennis Maiman, and Kevin M. Koch
59 MRI of the Knees in Asymptomatic Adolescent Soccer Players:

A Case Control Study
Simone B. Matiotti, Ricardo B. Soder, Rafaela G.Becker, Francisco S. Santos,
and Matteo Baldisserotto
66 Comparison of Chemical Shift-Encoded Water Fat MRI and MR Spectroscopy in

Quantification of Marrow Fat in Postmenopausal Females
Guanwu Li, Zheng Xu, Hao Gu, Xuefeng Li, Wei Yuan, Shixin Chang, Jingzheng Fan,
Horea Calimente, and Jiani Hu
Breast 74 Background Parenchymal Enhancement Over Exam Time in Patients With

and Without Breast Cancer
Amy Melsaether, Akshat C. Pujara, Kristin Elias, Kristine Pysarenko, Anjali Gudi,
Katerina Dodelzon, James S. Babb, Yiming Gao, and Linda Moy
84 Stimulated Echo Diffusion Tensor Imaging (STEAM-DTI) with Varying Diffusion

Times as a Probe of Breast Tissue
Jose R. Teruel, Gene Y. Cho, Melanie Moccaldi RT, Pål E. Goa, Tone F. Bathen,
Thorsten Feiweier, Sungheon G. Kim, Linda Moy, and Eric E. Sigmund
94 Dynamic Contrast-Enhanced and Diffusion-Weighted MRI of Estrogen

Receptor-Positive Invasive Breast Cancers: Associations Between Quantitative
MR Parameters and Ki-67 Proliferation Status
Jong Ki Shin and Jin You Kim
Pelvis 103 Haralick Textural Features on T2-Weighted MRI Are Associated With
Biochemical Recurrence Following Radiotherapy for Peripheral Zone
Prostate Cancer
Kh
emara Gnep, Aur eline Fargeas, Ricardo E. Guti
errez-Carvajal, Fr
ed
eric Commandeur,
Romain Mathieu, Juan D. Ospina, Yan Rolland, Tanguy Rohou, S ebastien Vincendeau,
Mathieu Hatt, Oscar Acosta, and Renaud de Crevoisier
118 Incremental Value of MRI for Preoperative Penile Cancer Staging

Fabiano Rubi~ao Lucchesi, Rodoldo Borges Reis, Eliney Ferreira Faria, Roberto Dias Machado,
Rodrigo Ribeiro Rossini, Leonardo D. Borregales, Gyl Eanes Barros Silva,
and Valdair Francisco Muglia
125 Optimal High b-Value for Diffusion Weighted MRI in Diagnosing High Risk
Prostate Cancers in the Peripheral Zone
Harsh K. Agarwal, Francesca V. Mertan, Sandeep Sankineni, Marcelino Bernardo,
Julien Senegas, Jochen Keupp, Dagane Daar, Maria Merino, Bradford J. Wood,
Peter A. Pinto, Peter L. Choyke, and Baris Turkbey
VOLUME 45, NUMBER 1, JANUARY 2017

Cardiac 132 Endogenous Assessment of Diffuse Myocardial Fibrosis in Patients With
T1q-Mapping
Joep W.M. van Oorschot, Fatih G€ uçl€
u, Sanne de Jong, Steven A.J. Chamuleau,
Peter R. Luijten, Tim Leiner, and Jaco J.M. Zwanenburg
139 MRI-Based Computational Hemodynamics in Patients With Aortic Coarctation

Using the Lattice Boltzmann Methods: Clinical Validation Study
Hanieh Mirzaee, Thomas Henn, Mathias J. Krause, Leonid Goubergrits, Christian Schumann,
Mathias Neugebauer, Titus Kuehne, Tobias Preusser, and Anja Hennemuth
147 Effects of Cortisol on the Heart: Characterization of Myocardial Involvement in

Cushing’s Disease by Longitudinal Cardiac MRI T1 Mapping
Charles Roux, Nadjia Kachenoura, Zainab Raissuni, Elie Mousseaux, Jacques Young,
Martin J. Graves, Christel Jublanc, Philippe Cluzel, Philippe Chanson, Peter Kamenick
y,
and Alban Redheuil
Neuro 157 Betel Quid Chewing Alters Functional Connectivity in Frontal and Default
Networks: A Resting-State fMRI Study
Xiaojun Huang, Zhening Liu, Tumbwene E. Mwansisya, Weidan Pu, Li Zhou, Chang Liu,
Xudong Chen, Robert Rohrbaugh, Carla Marienfeld, Zhimin Xue, and Haihong Liu
167 Benign and Malignant Orbital Lymphoproliferative Disorders: Differentiating

Using Multiparametric MRI at 3.0T
Xiao-Quan Xu, Hao Hu, Hu Liu, Jiang-Fen Wu, Peng Cao, Hai-Bin Shi, and Fei-Yun Wu
177 Intrinsic Brain Network Abnormalities in Codeine-Containing Cough

Syrup-Dependent Male Individuals Revealed in Resting-State fMRI
Ying-wei Qiu, Huan-Huan Su, Xiao-fei Lv, Xiao-fen Ma, Gui-hua Jiang, and Jun-zhang Tian
187 Reproducibility Measurement of Glutathione, GABA, and Glutamate: Towards

In Vivo Neurochemical Profiling of Multiple Sclerosis With MR Spectroscopy at 7T
Hetty Prinsen, Robin A. de Graaf, Graeme F. Mason, Daniel Pelletier, and Christoph Juchem
199 Differential Diagnosis of Mitochondrial Encephalopathy With Lactic Acidosis and

Stroke-Like Episodes (MELAS) and Ischemic Stroke Using 3D Pseudocontinuous
Arterial Spin Labeling
Rui Li, Hua-feng Xiao, Jin-hao Lyu, Danny J.J. Wang, Lin Ma, and Xin Lou
Technical Development
Neuro 207 Magnetic Susceptibility-Induced Echo-Time Shifts: Is There a Bias in Age-Related

fMRI Studies?
Giang-Chau Ngo, Chelsea N. Wong, Steve Guo, Thomas Paine, Arthur F. Kramer,
and Bradley P. Sutton
Original Research
Vascular 215 Quantification of Common Carotid Artery and Descending Aorta Vessel Wall
Thickness From MR Vessel Wall Imaging Using a Fully Automated
Processing Pipeline
Shan Gao, Ronald van ’t Klooster, Anne Brandts, Stijntje D. Roes, Reza Alizadeh Dehnavi,
Albert de Roos, Jos J.M. Westenberg, and Rob J. van der Geest
229 Thrombus-Mimicking Artifacts in Two-Point Dixon MRI: Prevalence, Appearance,

and Severity
Tilman Schubert, Peter Bannas, Sonja Kinner, Samir Sharma, James H. Holmes,
Mahdi Salmani Rahimi, Frank R. Korosec, and Scott B. Reeder
237 Optimization of Two-Compartment-Exchange-Model Analysis for Dynamic

Contrast-Enhanced MRI Incorporating Bolus Arrival Time
Guy Nadav, Gilad Liberman, Moran Artzi, Nahum Kiryati, and Dafna Ben Bashat
Abdomen 250 Application of Texture Analysis on Parametric T1 and T2 Maps for Detection of
Hepatic Fibrosis
HeiShun Yu, Anne-Sophie Touret, Baojun Li, Michael O’Brien, Muhammad M. Qureshi,
Jorge A. Soto, Hernan Jara, and Stephan W. Anderson
260 Intravoxel Incoherent Motion Diffusion-Weighted Imaging of the Pancreas:

Characterization of Benign and Malignant Pancreatic Pathologies
Bohyun Kim, Seung Soo Lee, Yu Sub Sung, Hyunhee Cheong, Jae Ho Byun,
Hyoung Jung Kim, and Jin Hee Kim
270 Evaluation of Antiangiogenic and Antiproliferative Effects of Sorafenib by
Sequential Histology and Intravoxel Incoherent Motion Diffusion-Weighted
Imaging in an Orthotopic Hepatocellular Carcinoma Xenograft Model
Shuo-hui Yang, Jiang Lin, Fang Lu, Zhi-hong Han, Cai-xia Fu, Peng Lv, Hao Liu,
and Dong-mei Gao
281 Gadoxetic Acid-Enhanced MRI for the Characterization of Hepatocellular

Carcinoma: A Systematic Review and Meta-Analysis
Joanna K. Duncan, Ning Ma, Thomas D. Vreugdenburg, Alun L. Cameron, and Guy Maddern
291 Acute Kidney Damage Induced by Low- and Iso-osmolar Contrast Media in Rats:
Comparison Study With Physiologic MRI and Histologic-Gene Examination
Chen-Jiang Wu, Mei-Ling Bao, Qing Wang, Xiao-Ning Wang, Xi-Sheng Liu, Hai-Bin Shi,
and Yu-Dong Zhang
303 Persistent T2*-Hypointensity of the Liver Parenchyma After Irradiation to the

SPIO-Accumulated Liver: An Imaging Marker for Responses to Radiotherapy in
Hepatic Malignancies
Toshihiro Furuta, Masayuki Yamaguchi, Manabu Minami, Kuni Ohtomo, and Hirofumi Fujii
Volume 45, Number 1 was mailed the week of December 19, 2016
JMRI INSTRUCTIONS FOR AUTHORS
The Journal of Magnetic Resonance Imaging (JMRI) is owned by the Inter- Number all manuscript pages in the upper right corner, beginning with
national Society for Magnetic Resonance in Medicine (ISMRM) and pub- the abbreviated title page. Do this even though the system also supplies
lished by Wiley Subscription Services, Inc., a Wiley Company. JMRI is page numbers in the PDF view of the submission after it is uploaded.
devoted to the timely publication of basic and clinical research, educa- Double-Blinded Review: To maintain anonymity during the review
tional and review articles, and other information related to the diagnostic process, reviewers receive manuscripts with no authors’ names listed
applications of magnetic resonance. to ensure unbiased, double-blinded review.
All manuscripts must be submitted electronically. For complete instruc- Instead of providing the authors’ names and affiliations in the main
tions on how to do so, go to http://mc.manuscriptcentral.com/jmri and fol- manuscript file, please do so on a separate title page file. Do not
low the instructions for creating an account and submitting a manuscript. give names of authors or their affiliations in the main manuscript
text, on the figures or tables, or in the acknowledgments and disclo-
sures. In the body of the text make a conscious effort to use phras-
General correspondence concerning JMRI may be ing that does not reveal the identity of the authors. For example,
directed to instead of using phrases such as ‘‘In our study. . .’’ followed by a ref-
erence number identifying your group, use phrases such as ‘‘In a pre-
Mark E. Schweitzer, MD, Editor-in-Chief, JMRI vious study. . .’’. After review, the original wording can be restored.
Stony Brook Department of Radiology When submitting your manuscript, (1) upload the full title page as
HSC L4 Room 120 the first file (designate it as ‘‘Title Page’’) and (2) upload the manu-
Stony Brook, NY 11794-8460 script (i.e., all elements except the full title page) as the next file(s)
Phone: (+1) (631) 444-7955 (designate them as Main Document or Image [as appropriate]). NOTE:
Fax: (+1) (631) 444-7538 If you have uploaded the files correctly, you WILL NOT be able to
view the full title page in the PDF and HTML proofs of your manu-
jmrieditor@ismrm.org
script. However, the editor and editorial office can easily view the
author information during the pre-review check and reviewer selection
stages.
n TYPES OF MANUSCRIPTS
General guidelines for manuscript length and number of references and
illustrations are as follows:
Original Research: up to 20 manuscript pages (5,000 words), 40 refer-
Full Title Page
ences, 10 figures and tables. Title: Do not use registered/trademark terms in the manuscript title.
Technical Development: up to 10 manuscript pages (2,500 words), 20 Use their generic equivalents instead. Do not use all but the most familiar
references, 5 figures and tables. abbreviations. Avoid the use of such terms as CMR, MRE, as they may
have more than one meaning.
Case reports: up to 4 manuscript pages (1,000 words), 3 figures and 15
Author Names and Degrees: Provide first names, middle initials, and
references, no abstract.
highest academic/technical degree(s) of all authors (e.g., MD, PhD, MS,
Reviews: up to 30 manuscript pages (7,500 words), 100 references, 15 BS, RN, RT). Please do not list honorary degrees, or fellowships.
figures and tables. Author Affiliations: Please list departmental/institutional affiliations for
Editorial/commentary: most of these are solicited. Five pages, 15 refer- authors. Please tie each author to the appropriate department(s)/institution(s)
ences and only a rare figure. using consecutive, superscripted Arabic numbers.
Letters to the Editor: up to 500 words, 10 references, one figure/table. Corresponding Author Info: Provide the name, street address, tele-
phone and fax numbers, and the e-mail address of the corresponding
author.
Reporting Standards Acknowledgments: Use this section to acknowledge individuals who
Authors of systematic reviews and meta-analyses are advised to strictly have provided personal assistance. Please give full names of all individuals
adhere to the PRISMA statement (Preferred Reporting items for Systematic mentioned in this section.
Reviews and Meta-Analyses), www.prisma-statement. org. Grant Support: List grant support, if any, on the full title page.
Authors of diagnostic accuracy studies are advised to strictly adhere to Running Title: At the bottom of the page, please include a short run-
the STARD statement (STAndards for the Reporting of Diagnostic accuracy ning title of 40 characters/spaces or less.
studies) http://www.stard-statement.org/
Authors of clinical trials are advised to strictly adhere to the CONSORT
statement (CONsolidated Standards of Reporting Trials). http://www.con-
sort-statement.org/
Terminology
Avoid the use of subjective terms like high resolution, fast, high field. State
the resolution or field strength at the beginning of your paper then call it
NOTE ‘‘high’’ later in the paper.
The instructions that follow address preparation of INITIAL manuscript sub- Define each used abbreviations within each section of the manuscript
missions and resubmissions intended for review and possible publication in (abstract, body, and figures) or provide an abbreviation key. Avoid abbre-
JMRI. Instructions for preparation and submission of REVISED manuscripts viations in the title.
intended for possible publication in JMRI are included in decision notifica-
tion letters that are sent to authors AFTER peer review. Authors preparing
REVISED manuscripts should refer to the instructions and checklist pro- Abstract/Keywords
vided in their decision notification letter.
Abstract: An abstract of no more than 250 words must be submitted
with each manuscript. Include the abstract in the text file AND in the
appropriate field of the website. Abstracts submitted with Original
n MANUSCRIPT ORGANIZATION Research papers and Technical Developments must be structured as fol-
Original writings will be accepted with the understanding that they are lows: Purpose (state the reason for the study; use an infinitive phrase
contributed solely to JMRI. Please ensure that all pages of the manuscript rather than a complete sentence [e.g., To evaluate. . .]), Materials and
(including the reference section) are double-spaced and in 12-point type, Methods (briefly describe how the study was performed), Results (briefly
with at least 2.5 cm margins. Acceptable typefaces include Arial, Times describe the most important results), results must include numerical data in
New Roman, and Bookman Old Style. all situations. Avoid the use of descriptive adjectives. Conclusion (briefly
Each manuscript component should begin on a new page in this order: state what is concluded from the results). Do not conjecture about what
Full Title Page with acknowledgments (upload as a SEPARATE file when this might be used for in the future. The conclusion should be the mirror
submitting manuscript) image of the purpose. Abstracts submitted with Case reports, and Review
Title/Abstract/Keywords Articles follow an unstructured format and should state in a single para-
Text graph what was done, what was found, and what was concluded. Other
References types of papers should not have an abstract.
Tables (each on a separate page) Keywords: Every manuscript must include a list of three to six key
Figure Legends words below the abstract. These words will assist indexers who will then
Figures classify the paper under specific subject headings for future indexes.
INSTRUCTIONS FOR AUTHORS
CONTINUED
Text in the text, in parentheses, other material (manuscripts submitted, unpub-
lished data, personal communications, and the like) as in the following
Authors must use Microsoft Word for preparation of text files. The text of example: (Smith J, unpublished data). If the owner of the unpublished data
observational and experimental studies is arranged in sections with the head- or personal communication is not an author of the manuscript under
ings Introduction, Materials and Methods, Results, and Discussion. Long review, a signed statement is required verifying the accuracy of the attrib-
papers may need subheadings in both the Materials and Methods and the uted information and agreement to its publication. Anything which is not
Results sections to clarify their content. The discussion section should be common knowledge to the presumed reader must be references. If you
entirely in narrative form with no subheadings. Begin last paragraph of Discus- state that something is controversial, authors disagree, or the literature is
sion with ‘‘In conclusion.’’ unclear, 2-3 references should be provide for that statement. Most other
Introduction: Set the stage for why you did what you did. Never state statements rarely require more than one reference. Always reference the
‘‘we are the fist’’, ‘‘not previously done’’ etc. The introduction should end original source, and JMRI prefers references of non-review type papers.
with a clearly stated purpose of the study. This should be similar to the The abbreviations used for periodicals cited in the references should fol-
purpose stated in your abstract. Do not provide an overview of your meth- low the style of Index Medicus. For journal articles, list surnames and ini-
ods anywhere in your introduction. Give only strictly pertinent references, tials of all authors when six or less, such as:
and do not review the subject extensively.
Materials and Methods: Must include a statement about IRB/REB ap-
1. Geurts JJG, Barkhof F, Castelijns JA, Uitdehaag BMJ, Polman CH, Pouwels
proval and informed consent or the waiver thereof at the beginning of this
PJW. Quantitative H-MRS of healthy human cortex, hippocampus, and thala-
section. For animal studies a statement about animal care committee ap-
mus; Metabolite concentrations, quantification precision, and reproducibility.
proval should be provided. Describe your selection of the observational
J Magn Reson Imaging 2004;20:366--371.
or experimental subjects (patients or experimental animals, including con-
trols) clearly. Describe in detail your inclusion and exclusion criteria and
When seven or more authors are listed, only the first three names need to
the numbers of patients excluded for each reason. Identify the method,
be identified, followed by ‘‘et al,’’ such as:
instrumentation (manufacturer’s name and address in parentheses), and
procedures in sufficient detail to allow other workers to reproduce the
results. For subjective observations, always provide the exact years of ex- 2. Schwickert HC, Roberts TP, Shames DM, et al. Quantification of liver
perience of the observer, and if a co-author their initials. Provide assess- blood volume: comparison of ultra short TI inversion recovery echo pla-
ments of variability in subjective assessments and measurements. Do not nar imaging (ULSTIR-EPI), with dynamic 3D-gradient recalled echo imag-
include the name of the institution where the work was performed, either ing. Magn Reson Med 1995;34:845-852.
within the text or at the top of each manuscript page. Give references to
established methods, including statistical methods that have been pub- For online only articles, or articles ‘‘in press’’, the DOI (Digital Object Iden-
lished but are not well known; describe new or substantially modified tifier) provides a persistent link to the article:
methods,. Include numbers of obser - vations . Statistical tests used should
be a subsection at the end of your methods section, and you should set 3. Madelin G and Regatte RR. Biomedical applications of sodium MRI in
the level of your type I and type II errors when appropriate.. If there are vivo. 2013; in press. doi: 10.1002/jmri.24172
co-authors we are employees of a commercial entity so discussion on how
potential conflicts of interest were minimized is usually needed. In the case of books, the authors of a chapter, title of the chapter, edi-
Results: Present the results in logical sequence in the text, tables, and tor(s), title of the book, edition, city, publisher, year, and specific pages
illustrations. Do not repeat in the text all the data in the tables and/or illus- must be provided. For the complete book, the number of pages is given
trations; summarize only important observations. Be progressively granular. instead, e.g., 512 p.
Do not repeat the statistical tests listed in the methods section. Always
provide precise p values when you say ‘‘significant’’ or ‘‘non-significant’’. If 4. Desnick R, Ioannou YA, Eng CM. a-Galactosidase A deficiency: Fabry
you conclude the negative-there is no association- provide confidence disease. In: Scriver CR, Beaudet AL, Sly WS, Valle D, editors. The meta-
intervals or a power analysis. Avoid descriptive adjectives. bolic and molecular bases of inherited disease, 8th edition. New York:
Discussion: The portion of the paper does not repeat your findings but McGraw-Hill; 2001. p 3733--3774.
puts your findings into context. Often the first paragraph of the discussion
highlights the most important findings of your research. Emphasize the new 5. Feller W. An introduction to probability theory and its applications, 2nd
and important aspects of the study and conclusions that follow from them. edition. New York: John Wiley and Sons Inc; 1968. 48 p.
State when you results are similar to others and when they are different. If they
are different state why you think this occured. Do not repeat in detail the data
given in the Results section. Do not repeat numerical results in your discussion. Abstracts, editorials, and letters to the editor should be noted as such,
Do not reference figures or tables. Provide reasons for using your techniques, giving page number range or abstract number in parentheses at end of
and evaluate the limitations of these techniques. Link the conclusions with the reference.
goals of the study, but avoid unqualified statements and conclusions that are For meetings other than ISMRM, give the author list, presentation title,
not supported completely by the data. Do not add any new data that was not meeting title, sponsoring organization name, month, day(s), year of meet-
mentioned in your results section. Avoid claiming priority and alluding to work ing, location, page number range or abstract number, and if the proceed-
that has not been completed. Avoid self-agrandizement or hyperbole, few ings have been published, give the publisher name and location:
studies change how medicine is practiced. Be concerned about generalizabil-
ity, and do not claim that your findings might be applied to different popula- 6. Rodenburg JB, Smink J, Salverda P, et al. Slice tracking and navigator
tions. State new hypotheses when warranted, but clearly label them as such. gating provide good image quality without the need for breath-holding
Recommendations, when appropriate, may be included, but these should be in lung imaging. In: Proceedings of the 8th Annual Meeting of ISMRM,
modest, and well considered. The penultimate paragraph of your discussion Denver, 2000. (abstract 954).
should be a study limitations paragraph. This may include limitations of your
sequence or processing, but should be focused to a greater degree, on the Please be sure that all documentation is accurate.
limitations of your study design. The discussion should end with a conclusion.
This conclusion is based on you results, put into context of prior authors results
and your study limitations. Discussions should not provide an overview of a dis- Tables
ease, treatment or technique, but provide just enough information for the Tables should be numbered in Arabic numerals in the order of their
reader to put the findings into context. Discussions should be commensurate appearance in the text and should have a brief title. All abbreviations used
in length with complexity of your results, and in general should be brief. in the table should be explained in a footnote. Each table should be struc-
tured with three horizontal lines and no vertical lines or boxes.
References
Number references consecutively in the order in which they are first men- Figure Legends
tioned; place the numbers between parentheses on the line. Only literature A legend must be supplied for each illustration and should not duplicate
that is published or in press (with the name of the publication known) may text material. All figure legends should appear collectively on one or more
be numbered and listed; abstracts and letters to the editor may be cited, pages at the end of the text file, and each legend should also appear
but they must be less than 3 years old and identified as such. Refer to only below the corresponding figure.
CONTINUED
Figures Electronic Submission of Graphics
Figures should be limited to those required to show the essential features Images submitted must be in Tagged Image File Format (TIFF) or Encapsu-
described in the paper. Excessive illustrations encroach on the total space lated PostScript (EPS). They may be submitted from either a MS-DOS or
available in JMRI to the detriment of the publication. Images to be com- Macintosh system. PowerPoint figures are acceptable, but are not pre-
bined into one cut, such as anteroposterior and lateral views, should be ferred for best quality in print. If figures are in PowerPoint, Adobe Photo-
the same height to facilitate reproduction. Images should correspond in shop, or EPS, it is recommended to use Aldus Freehand or QuarkXPress.
appearance to the tonal relations of the original image (with the patient’s Figures prepared in Word, Excel, Microsoft Publisher, Lotus 123 and Corel
right to the observer’s left for coronal images; axial images should employ Draw are not acceptable. All images must be saved and submitted in final
the ‘‘view from below’’). size; all cropping and manipulation must be completed before the images
Figures should be numbered in the order in which they are men- are submitted to the publisher. Figures should not exceed 8-in (21.6-cm) in
tioned in the text. All figures should be labeled a, b, c, etc. in the height; figure widths should be 3-in (8.25-cm) for 1 column, 5-in. (13-cm)
for 1.5 columns, or 6-in (17.15-cm) for 2 columns.
upper left corner, but not within the figure frame. Symbols, letters, and
Rapid InspectorTM gives authors of scientific, technical and medical jour-
numbers must be legible after reduction, with the smallest data points
nals a resource for certifying their illustrations prior to submission. Utilizing
no less than 1.5- to 1.7-mm high. Images need to be cropped and all
this software will ensure that your graphics are suitable for print produc-
irrelevant parts removed. Figure parts should be limited to 10.
tion. To download this journal’s free Rapid Inspector software, please visit
For each accepted manuscript, the Journal provides print color free http;//rapidinspector.cadmus.com/wi/index.jsp.
of charge. Authors are encouraged to submit color illustrations that high- Minimum figure resolutions are as follows: 1200 dots per inch (dpi) for
light the text and convey essential scientific information. For best repro- line art; 600 dpi for color, and 300 dpi for black and white. To ensure
duction, bright, clear colors should be used. Please place your color good reproduction online and in print, the minimum black value on a half-
images against a white background wherever possible. Please contact the tone scan (highlight) should be 4% black. The maximum value (shadow)
production editor at jmri@wiley.com for further information. should be no higher than 95%. Avoid the use of fine lines (point and
It is extremely important that authors check the accuracy of any keys on below) for graphs and charts. Use only Adobe Type 1 fonts in creating
the images before the final versions are sent to the Editorial Office. For images, and limit the number of fonts used. Do not reletter images in Pho-
additional information regarding preparation of figures, see ‘‘Electronic toshop. If relettering must be done, import the image into either Freehand
Submission of Graphics’’ in these instructions. or QuarkXPress and reletter, then make an EPS file. Make sure all scanned
images are ‘‘clean.’’ Look for and clean up dust specks, scratches, tape
marks, and anything that is not part of the actual image. Files generated in
n OTHER MANUSCRIPT ELEMENTS Freehand should be saved in EPS format.
If QuarkXPress is used for image assembly, please follow these
Units & Abbreviations guidelines:
Radiation measurements and laboratory values should be given in the (a) After creating the image, adjust the page size in the ‘‘Document
International System of Units (SI) (resources: SI Units in Radiation Protec- Setup’’ window in the pull-down menus. (b) Modify the page size to be
tion and Measurements, NCRP Report no. 82 [August 1985]; ‘‘Now Read only slightly larger than the image size. If the image is 20 pica wide by 10
This: The SI Units Are Here,’’ JAMA 1986;255:2329--2339). Blood pressure pica deep, make the page size 22 pica wide by 12 pica deep and center
should still be reported in millimeters mercury. Abbreviations should be the image within the page. (c) Once this has been done, save the page as
spelled out when first used in the text, such as ‘‘cerebrospinal fluid (CSF).’’ an EPS file by doing a ‘‘Save As. . .’’ from the pull-down menu. (d) Each
Magnetic resonance terminology should conform to the Glossary of MR image must be submitted as a separate EPS file. Do not submit an entire
Terms published by the American College of Radiology. Laboratory slang, Xpress file saved in EPS format.
clinical jargon, and uncommon abbreviations should be avoided. Discus-
sion of previous literature and material presented must be restricted to the
significant findings. Submission of Supporting Information
Supporting Information may include (but is not limited to) video clips, large
Informed Consent sections of tabular data, program code, or electronic graphical files that are
Manuscripts reporting the results of experimental studies on human subjects otherwise not suitable for print media. When submitting material for consid-
must include a statement to the effect that informed consent was obtained after eration please follow the guidelines below.
the nature of the procedure(s) had been fully explained. Peer review: Supporting Information must be submitted at the time of
The maintenance and care of experimental animals to provide humane peer review.
treatment and to ensure reliable results are described in the National Insti- Acceptance and electronic submission: Once material has been
tutes of Health guidelines for use of laboratory animals. Authors are accepted by the editors, it must be submitted in electronic form. Please
advised to comply with these guidelines and to acknowledge such compli- note that Supporting Information is not copyedited or formatted in any
ance in their manuscripts. way by the Publisher.
Data file types: There are no restrictions on filetypes of the data that
Author Responsibility and Exclusive Submission you submit. Please keep in mind, however, that the more universal the file-
Authors must be certain that no manuscript on the same or similar material type the more accessible to the community. The use of popular compres-
has been or will be submitted to another journal by themselves or others sion protocols is highly encouraged. If your material is presented in
at their institution before their work appears in JMRI. The submission by PostScript format, the submission of an accompanying Adobe Acrobat
authors of similar material to advertising media must be indicated at the PDF file is encouraged, in that event both formats will be offered to the
time JMRI receives the manuscript. journal’s readers. In cases where PDF files do not accompany a PostScript
file, only the Postscript will be offered.
n ADDITIONAL GUIDELINES Publication Access: Supporting Information for published articles can
be accessed through the online version of the article, which is available at
Financial Interest onlinelibrary.wiley.com.
Authors must describe any direct or indirect financial interest they may
have in the subject matter of a submitted manuscript, and must authorize
the ISMRM to publish this financial disclosure with the article if deemed n BOOK REVIEWS
appropriate. The Financial Disclosure policy of the ISMRM is as follows:
Anyone who would like to help in reviewing books for JMRI should contact
‘‘Each author shall describe (a) any direct financial interest which that
author has in the subject matter discussed in the submitted manuscript, Diego R. Martin, MD, PhD
and (b) any affiliation or financial involvement which that author has with or JMRI Associate Editor, Book Reviews
in any organization with a direct financial interest in the subject matter dis- Professor and Head, Radiology Department
cussed in the submitted manuscript. Such information will be held in the
University of Arizona
confidence of the journal editor during the review process. If necessary, in
the editor’s view, this information may be shared with reviewers after dis- University Medical Center
cussion between the editor and author. The author shall grant permission 1501 North Campbell Avenue
for the ISMRM to publish the financial information described above, or an PO Box 245067
appropriate summary thereof, with the manuscript if the manuscript is Tucson, AZ 85724
accepted for publication.’’ E-mail: dmartin@radiology.arizona.edu
CONTINUED
n ACCEPTED MANUSCRIPTS Attribution license (CC-BY) [http://creativecommons.org/licenses/by/3.0/].
All other authors (non- RCUK and Wellcome Trust funded authors) are
Copyright Assignment and Authorship Criteria free to choose Creative Commons Attribution Non-Commercial (CC-BY-
NC) [http://creativecommons.org/licenses/by-nc/3.0/], or Attribution-Non-
Authors submitting a manuscript do so with the understanding that if it
Commercial-NoDerivs (CC-BY-NC-ND) [http://creati vecommons.org/
is accepted for publication, copyright in the article, including the right to
licenses/by-nc-nd/3.0/] license.
reproduce the article in all forms and media, shall be assigned exclusively
More information about open access license terms and conditions are
to the ISMRM. After the acceptance of their paper, authors will be pro-
available at the following website: http://olabout.wiley.com/WileyCDA/Sec-
vided instructions for completing an online copyright transfer agreement
tion/id-815641.html.
(CTA). Articles cannot be published until the publisher has received the
appropriate signed license agreement from the authors. Once your article
has been accepted and received by Wiley for production, the correspond-
Compliant with Funder Mandates - Open Access
ing author will receive an email from Wiley’s Author Services system which Policies
will ask them to log in and will present them with the appropriate license OnlineOpen is fully compliant with open access mandates --- meeting the
for completion. requirements of funding organizations where these apply, including but
JMRI requires that all authors and co-authors satisfy the authorship cri- not limited to:
teria described by the International Committee of Medical Journal Editors Research Councils UK: MRC, BBSRC, AHRC, ESRC, EPSRC, NERC,
(ICMJE), which state: ‘‘Authorship criteria should be based on: (1) substan- STFCThe Wellcome TrustAustrian Science FundTelethon ItalyNIHThe
tial contributions to conception and design, acquisition of data, or analysis Howard Hughes Medical Institute (HHMI)
and interpretation of data; (2) drafting the article or revising it critically for More information on funder mandates and open access policies are
important intellectual content; and (3) final approval of the version to be available at the following website: http://olabout.wiley.com/WileyCDA/Sec-
published. Authors should meet conditions 1, 2, and 3.’’ tion/id-406074.html.
Note to Contributors on Deposit of Accepted

Note to NIH Grantees
Version Pursuant to NIH mandate, Wiley-Blackwell will post the accepted version
Authors are permitted to self-archive the peer-reviewed (but not final) of contributions authored by NIH grant-holders to PubMed Central upon
version of the Contribution on the Contributor’s personal website, in the acceptance. This accepted version will be made publicly available 12
Contributor’s company/institutional repository or archive, and in certain months after publication. For further information, see www.wiley.com/go/
not for profit subject-based repositories such as PubMed Central as listed nihmandate.
at the following website: http://olabout. wiley.com/WileyCDA/Section/id-
820227.html, subject to an embargo period of 12 months for scientific,
technical and medical (STM) journals and 24 months for social science and Rights and Permissions
humanities (SSH) journals following publication of the final Contribution.
There are separate arrangements with certain funding agencies govern- Direct quotations, tables, or illustrations taken from copyrighted material
ing reuse of this version as set forth at the following website: http:// must be accompanied by written permission for their use from the pub-
www.wiley.com/go/funderstatement lisher. The permission is presented as a footnote or addition to the
The Contributor may not update the accepted version or replace it with legend and must provide complete information as to source. Photo-
the published Contribution. graphs of identifiable persons must be accompanied by a signed release
that indicates informed consent. Otherwise, the eyes must be blacked
For Authors Choosing OnlineOpen out to prevent identification.
Choosing OnlineOpen makes your article open access and freely available
to all on Wiley Online Library, including those who don’t subscribe to the Requests to use material that appears in JMRI and other permission
journal. OnlineOpen fulfills funding agency mandates that require grantees inquiries should be submitted via the ‘‘Request Permissions’’ link on Wiley
to archive the final version of their article. Wiley will also deposit your Onli- Online Library. Simply click on the link next to the article in question and
neOpen article in PubMed Central and PMC mirror sites. follow the instructions.
The cost for OnlineOpen is usually US$3,000, which can be paid by the
author, the author’s funding agency, or the author’s institution. Proofs and Reprints
Proofs will be sent electronically to the corresponding author. Authors
Copyright and Open Access Licenses should indicate all corrections in an annotated PDF file.
Authors choosing OnlineOpen will retain copyright in their articles and will To avoid delays in publication, proofs should be returned to the Produc-
be offered a choice of creative commons licenses. tion Editor at jmri@wiley.com within 48 hours of receipt.
All Research Councils UK (RCUK), Wellcome Trust and Austrian Science Authors may order reprints by contacting the address provided in the
Fund (FWF) funded authors will be directed to the Creative Commons proof package.
J_ID: JMRI Customer A_ID: JMRI25336 Cadmus Art: JMRI25336 Ed. Ref. No.: 16-0202.R1 Date: 2-December-16 Stage: Page: 11
REVIEW ARTICLE
Practical Applications of Balanced

Steady-State Free-Precession (bSSFP)
Imaging in the Abdomen and Pelvis
Nicola Schieda MD,* Inga Isupov MD, Andrew Chung MD,
Niamh Coffey MD, and Leonard Avruch MD
Balanced steady-state free-precession (bSSFP) is an important pulse sequence that may be underutilized in abdominal and
pelvic magnetic resonance imaging (MRI). bSSFP offers several advantages for abdominal and pelvic MRI that include:
bright blood effects, a relative insensitivity to the dephasing effects which occur in structures with linear movement, low
specific absorption rate (SAR), high signal-to-noise ratio (SNR), high spatial resolution, and rapid acquisition times. Bright
blood effects can be exploited to diagnose or confirm vascular pathologies when gadolinium-enhanced imaging cannot
be performed, is indeterminate, or is degraded by artifact. The relative insensitivity to dephasing artifact in areas of linear
movement is useful when imaging the biliary, urinary, and gastrointestinal tracts where dephasing artifacts may mimic fill-
ing defects such as calculi or polyps. Low SAR imaging is important in pediatric and pregnant patients and may be useful
in patients with medical devices that restrict SAR levels. Rapid acquisition times and high SNR are extremely valuable
assets in abdominal and pelvic MRI and bSSFP (which can be performed as static or cine acquisitions) and can be added
to most existing abdominal and pelvic protocols when deemed suitable without significantly prolonging examination
times. This article reviews the fundamentals of bSSFP imaging, presents vascular and nonvascular applications of bSSFP in
abdominal and pelvic MRI, and discusses potential limitations (including imaging artifacts) of bSSFP.
Level of Evidence: 5
J. MAGN. RESON. IMAGING 2017;45:11–20.
B alanced Steady-State Free-Precession (bSSFP; True FISP

[Siemens Healthcare, Erlangen, Germany], FIESTA
[GE Healthcare, Milwaukee, WI], Balanced FFE [Philips
dephasing artifacts in linear flow, and comparatively high
spatial resolution of bSSFP render it a useful pulse sequence
for many abdominal and pelvic applications. The purpose
Healthcare, Best Netherlands]) is a form of gradient recalled of this review is to evaluate the fundamentals of bSSFP, dis-
echo (GRE) magnetic resonance imaging (MRI) with dis- cuss potential advantages and disadvantages of bSSFP
tinctive imaging properties that can be exploited in abdomi- (including imaging artifacts), and to highlight applications
nal and pelvic MRI. Compared to the more traditional fast/ of bSSFP in the abdomen and pelvis using clinical
turbo spin-echo (FSE/TSE) and spoiled GRE sequences that examples.
are commonly employed in abdominal and pelvic MRI,
bSSFP may be underutilized and abdominal and pelvic MR Fundamentals of bSSFP
practitioners may be less familiar with this pulse sequence. As a brief review, GRE MR pulse sequences utilize gradient
While bSSFP is commonly used in abdominal and pelvic refocusing of excited spins to produce the recorded signal.2
MRI for assessment of the small bowel with enterography,1 After each excitation and gradient refocusing cycle, there
additional vascular and nonvascular applications of bSSFP will be some residual magnetization that remains in the
may be less well recognized. The rapid acquisition time, transverse plane before the next excitation begins (noting
high signal-to-noise ratio (SNR), low specific absorption that the timing of the next excitation is predetermined by
rate (SAR), bright-blood effects, relative insensitivity to the user-defined time to repetition [TR]).2 This residual
View this article online at wileyonlinelibrary.com. DOI: 10.1002/jmri.25336
Received Feb 22, 2016, Accepted for publication May 24, 2016.
*Address reprint requests to: N.S., Ottawa Hospital, University of Ottawa, Department of Medical Imaging, 1053 Carling Ave.,
Ottawa, Ontario, Canada K1Y 4E9. E-mail: nschieda@toh.on.ca
From the and Ottawa Hospital, University of Ottawa, Department of Medical Imaging, Ottawa, Ontario, Canada.
C 2016 International Society for Magnetic Resonance in Medicine

V 11
ID: srinivasanv Time: 15:48 I Path: //chenas03/Cenpro/ApplicationFiles/Journals/Wiley/JMRI/Vol04501/160160/Comp/APPFile/JW-JMRI160160

Journal of Magnetic Resonance Imaging
tion will be added to the residual transverse magnetization.

TABLE 1. Pulse Sequence Parameters Used for Bal-
anced Steady-State Free-Precession Imaging in the After several cycles, a steady state is achieved between the
Abdomen and Pelvis at 1.5 and 3.0T longitudinal and residual magnetizations. The signal
recorded with balanced GRE will consist of two compo-
Parameter 1.5Ta 3.0Tb nents: 1) free induction decay of the newly excited longitu-
dinal magnetization (T1- and T2 -weighted), and 2)
Field of view (cm) 40 40 refocused residual transverse magnetization (T2-
Slice thickness/ 4/1 c
4/1c weighted).4–6 In fully refocused GRE or bSSFP, both
gap (mm) the free induction decay of the longitudinal magnetization
TR/TE4 (msec) 4.2/2.2 3.2-3.2/ (T1-weighted) and the residual transverse magnetization (T2-
1.2-1.4 weighted) are recorded, which provides a mixture of T1 and
Flip angle (degrees) 70 45-50 T2 weighting.4–6 A full description of the other properties
of the bSSFP pulse sequence, including important aspects of
Bandwidth (Hz) 560 488-977
image contrast and SNR are beyond the scope of this arti-
Matrix 230 3 256 224-256 3 cle, but have been described in detail elsewhere.4–6 The typ-
256-320
ical parameters used for bSSFP imaging in the abdomen
Number of 1 1 and pelvis at our institution at both 1.5 and 3.0T are pro-
signals averaged
vided in Table 1. It is important to emphasize one addi- T1
Acceleration 0 0 tional property of bSSFP that is important in abdominal/
Physiology Breath-hold Breath-hold pelvic MRI, namely, with bSSFP balanced gradients are
a
1.5T systems used at our institution include Magnetom Sym- applied in all three axes (slice selection, phase encoding, and
phony and Magnetom Symphony TIM (Siemens Healthcare). readout) such that gradient-induced dephasing for stationary
b
3.0T systems used at our institution include Magnetom TRIO or linearly moving tissues within a particular TR is mini-
TIM (Siemens Healthcare) and Discovery 750W (GE
Healthcare). mized/eliminated.4–6
c
Slice thickness and gap can be modified to include thinner Intrinsic advantages of bSSFP for abdominal and pel-
slice acquisitions (resulting in less anatomic coverage) or thicker vic MRI include: 1) rapid acquisition times due to the
slice acquisitions/increased gap (resulting in increased anatomic
coverage). For dynamic or cine MRI, acquisitions are parti- reduced TR (bSSFP can be implemented as a breath-hold
tioned into stations where a series of images are obtained at sequence which freezes motion); 2) high SNR due to the
each station. For dynamic pelvic floor MRI (MR defecography) combined signal from both the recovered longitudinal and
a single sagittal plane is selected (including the rectum, vagina,
uterus, and urinary bladder) and images are repeated every sec-
residual transverse magnetization (bSSFP has the highest
ond (during rest Kegel exercising, Valsalva, and defecation). For SNR per unit time of any imaging pulse sequence;4,5 3) T2/
dynamic MR enterography, typically the coronal (or axial) T1 weighting that provides images where blood, bile, and
plane is selected and a series of 6-8 images are obtained at each other fluid like structures are of increased signal intensity;6
station and then repeated at multiple stations covering the
bowel in its entirety. and 4) a relative insensitivity to the dephasing artifacts
encountered in areas of linear movement (due to the bal-
anced gradients in all imaging axes).4–6
transverse magnetization may not be desirable, since it can Compared to single-shot half-Fourier FSE/TSE
alter the T1 and T2 weighting of the signal recorded in the (HASTE or ssFSE), which is more commonly used in
next excitation cycle. With the majority of GRE imaging in abdominal/pelvic MRI,7 bSSFP provides several additional
the body, the residual transverse magnetization is removed advantages including: 1) higher resolution images due to
(spoiled) such that the signal recorded from the next excita- less imaging blur related to the extended echo train of
tion/refocusing cycle will be composed of only the recovered ssFSE, and 2) lower SAR, which may be particularly advan-
longitudinal magnetization.3 This methodology controls the tageous when imaging at 3T or in patients with SAR restric-
amount of T1 and T2 weighting of the final image; how- tions (eg, restricted medical devices, pregnancy, and
ever, it suffers from a lower SNR because the signal of the pediatrics).3,8 Potential disadvantages of bSSFP compared to
residual transverse magnetization is lost.3 single-shot FSE include: a) susceptibility artifacts encoun-
Unlike spoiled GRE, balanced GRE pulse sequences tered with GRE compared to FSE; b) T2/T1 weighting
utilize the residual transverse magnetization in the subse- compared to T2-weighted imaging, which may limit charac-
quent excitation/refocusing cycle. The TR is minimized (eg, terization of tissues when basing assessment on a structures
the TR is usually much less than the T2 relaxation times of T2 signal (for example, when assessing the T2 signal of a
tissues being imaged) so that the residual transverse magnet- hepatic or renal mass); c) etching (india ink), banding, and
ization does not decay completely by the next excitation.4–6 fringe artifacts, which may compromise image quality (dis-
At the next excitation, the recovered longitudinal magnetiza- cussed below).
12 Volume 45, No. 1

Schieda et al.: bSSFP Imaging in the Abdomen and Pelvis
Applications of bSSFP in Abdominal and

pelvic MRI
Localizers
Localizer sequences are meant to provide a rapid three-plane
overview of the anatomy within the imaging field of view
(FOV) to assist the MR practitioner for the prescription of
subsequent pulse sequences. Typical pulse sequences used
for localization include: 2D fast spoiled GRE, single-shot
FSE, and bSSFP. bSSFP is an ideal pulse sequence for local-
ization. It offers advantages over fast spoiled GRE because it
provides better delineation of anatomy and over single-shot
FSE because areas of susceptibility artifact (T2 effect) that
are detected on bSSFP can be used to screen for occult
implants or devices that may not have been identified dur-
ing the preprocedure MR Safety Questionnaire (Fig. 1). A F1
disadvantage of bSSFP is moire (fringe) artifacts that occur
at the extremes of the imaging FOV that may obscure
important anatomy (discussed below).
Vascular Applications
FIGURE 1: An 89-year-old male undergoing magnetic reso- Two properties of bSSFP yield the so-called bright blood
nance cholangiopancreatography (MRCP) for evaluation of ele-
vated liver function tests. MR safety questionnaire was effect (eg, blood is inherently of increased signal intensity
completed by the patient prior to MRI and was unremarkable. unrelated to flow or contrast administration) that is critical
Coronal balanced steady-state free-precession (bSSFP) image for numerous applications in abdominal and pelvic MRI.6
obtained from three plane localizers demonstrates a large area
of susceptibility artifact in the right mid-abdomen (white bSSFP sequences are T2/T1-weighted (blood and other
arrow). The patient was questioned about this while remaining fluid-like structures are of increased signal) and also apply
on the scanner but could offer no additional information. A balanced gradients along all three imaging axes (so that
review of the patient’s electronic medical record revealed a
recent endoscopic procedure for bleeding peptic ulcer 7 days
dephasing effects are minimized/eliminated). bSSFP can
prior with placement of an endoscopy clip to stop bleeding. therefore be used to evaluate for vascular pathologies (com-
Endoscopy clips used at our institution are MR-conditional monly venous thrombosis9 when the use of gadolinium is
(Resolution Clip, Boston Scientific Medical) at 1.5 and 3T and
relatively contraindicated, such as in pregnancy (Figs. 2, 3) F2 F3
the examination was completed without incident.
or in patients with compromised renal function (Fig. 4).10 F4
FIGURE 2: A 27-year-old female with left leg swelling at 19 weeks gestational age. Sagittal gray-scale ultrasound image (a) of the
left common femoral vein (CFV) showed slow-flowing venous blood (white arrow) although the vein was compressible and demon-
strated complete color flow on color Doppler imaging (not shown). Ultrasound evaluation of the pelvic veins was limited by the
gravid uterus and adjacent bowel. The patients’ symptoms persisted and a follow-up pelvic MRI was performed to assess for
thrombus in the pelvic veins. Axial bSSFP images (b,c: Acquisition time 17 seconds breath-hold) show that the left external iliac
vein (EIV) is expanded and shows low signal (white arrow in b) compared to the bright signal in the contralateral EIV (open white
arrow in b). More caudally, the left EIV is expanded and shows low signal (white arrow in c) compared to the normally compressed
right EIV (open white arrow). Axial fat-suppressed (FS) 2D time-of-flight (TOF) pelvic venogram (d: Acquisition time 4 minutes 18
seconds free-breathing) with cranial to caudal saturation band confirms absence of flow in the left EIV (white arrow) and normal
flow in the right EIV (open white arrow). A diagnosis of pelvic deep venous thrombosis was established. Also note gravid uterus
in (b), which preempted the routine use of gadolinium in this patient.
January 2017 13

FIGURE 3: A 31-year-old female 8 days post-cesarean section with fever and elevated white blood cell count. MRI was performed
to rule out pelvic thrombophlebitis. Coronal bSSFP image (a: Acquisition time 18 seconds breath-hold) shows thrombus in the
right gonadal vein (white arrows), which is expanded and of low signal compared to the proximal inferior vena cava (IVC) and
aorta. The cranial extent of the thrombus is well seen (arrowhead). Axial 2D FS TOF images (b: Acquisition time 5 minutes 32 sec-
onds free-breathing) show the thrombus as low signal (arrowhead) protruding into the opacified IVC (white arrow). Also note
banding (moir e) artifacts on the liver in (a) at the extreme of the imaging field of view (FOV) in (a). The patient declined gadolin-
ium injection.
An additional advantage of bSSFP over other noncontrast- other vascular conditions in the abdomen and pelvis includ-
enhanced MR techniques used to image for venous throm- ing venous or arterial aneurysms (Fig. 7) and arterial sten- F7
bosis (eg, time-of-flight [TOF] venography) is comparatively oses or dissection.
shorter acquisition times. The bright blood effects of bSSFP
can also be exploited to determine if a structure is related to Biliary Applications
the vascular system when gadolinium is not administered Similar to blood, bile is also of increased signal intensity on
F5 (Fig. 5) or in cases where postgadolinium-enhanced images bSSFP. bSSFP can be used for MR cholangiopancreatograpy
F6 are indeterminate (Fig. 6). bSSFP may also be complemen- (MRCP) imaging.11 An advantage of bSSFP compared to
tary to gadolinium-enhanced imaging for the evaluation of single-shot FSE for biliary imaging relates to the
FIGURE 4: A 71-year-old female with clear cell renal cell carcinoma (RCC) and tumor thrombus undergoing MRI for preoperative
planning. Due to severe renal failure gadolinium was not administered. Axial T2W FSE (a) shows a large mass centered in the right
kidney (open white arrow) with tumor thrombus (thick white arrow) filling the right renal vein and inferior vena cava (IVC). Note a
thin rim of patent IVC (thin white arrow) medially depicted as a flow void. Coronal bSSFP (b,c) images clearly depict the cranial
extent of the tumor thrombus growing from the right kidney (open white arrow) into the IVC (thick white arrows). Portions of the
IVC that are clear of tumor (thin white arrows) are well delineated on the bSSFP image. The suprahepatic IVC and right atrium
were confidently reported to be clear of tumor and cardiac surgery and bypass was not required for resection.
14 Volume 45, No. 1

FIGURE 5: A 56-year-old female with history of pancreatitis imaged with nongadolinium-enhanced magnetic resonance cholangio-
pancreatograpy (MRCP) to assess for choledocholithiasis. Axial T2W ssFSE image (a) shows a complex mass adjacent to the pan-
creatic head (white arrow) with internal mixed signal intensity. The mass is homogeneously hypointense on T1W 3D GRE (b). A
pancreatic tumor was suspected and follow-up gadolinium-enhanced MRI recommended. In retrospect, coronal bSSFP image (c)
obtained at time of MRCP illustrates that the mass is homogeneously of increased signal (white arrow) and is communicating with
the gastroduodenal artery (GDA) (thin white arrow). Axial T1W 3D GRE image obtained during the arterial phase (d) of enhance-
ment confirms the diagnosis of GDA aneurysm (white arrow).
insensitivity of bSSFP to the dephasing effects of flowing for biliary imaging, which can also be performed as a 3D
bile, which can be misinterpreted for filling defects on FSE acquisition.11 In our practice, we also rely on three the
F8 imaging (Fig. 8) by an inexperienced interpreter. Bright properties of bSSFP (including the opposed-phase etching
blood effects on bSSFP can be used to diagnose vascular [india ink] artifact at fat–water interfaces) to improve assess-
conditions on nongadolinium-enhanced MRCP studies, for ment of vascular invasion in pancreatic cancer in combina-
example, to confirm the suspected diagnosis of a crossing tion with gadolinium-enhanced images, which is a critical
vessel causing an apparent biliary stricture without recalling determinant for surgical resectability (Fig. 10).12–14 A disad- F10
F9 the patient (Fig. 9). Rapid acquisition times and high- vantage of bSSFP for MRCP relates to susceptibility artifacts
resolution imaging further emphasize the utility of bSSFP from surgical clips or gas within duodenal diverticula or
FIGURE 6: A 51-year-old female with history of metastatic clear cell RCC postnephrectomy and isolated lung metastatectomy in
clinical remission. Routine follow-up CT imaging showed a hypervascular mass in the left adnexa (not shown) and MRI was recom-
mended. Axial and coronal T2W FSE images (a) demonstrate a low signal intensity mass (thick white arrows) in close proximity to
pelvic vessels (thin white arrows). The mass is separate from the normal ovaries (arrowheads). Axial and off-axis curved reformat-
ted oblique maximum intensity projection (MIP) postgadolinium-enhanced T1W 3D GRE images show the mass (thick white arrows)
is enhancing as much as the adjacent internal iliac artery (thin white arrow) and appears in contiguity with the vessel. The mass
was diagnosed as an internal iliac artery aneurysm. A follow-up ultrasound performed 3 years later showed the mass had grown
and was solid (not shown). The final diagnosis after laparoscopic resection was a broad ligament fibroid. In retrospect, review of
coronal bSSFP image (c) would have been contributory. The mass was of intermediate signal (thick white arrow) isointense to
obturator internis muscle (arrowhead) and does not show the bright signal that would be expected in a vessel; for example, com-
pare to the left internal iliac artery (thin white arrow).
January 2017 15

FIGURE 7: A 42-year-old male with complex cystic lesion in the right renal pelvis detected on abdominal sonogram undergoing MRI
for further characterization. Transverse and sagittal grayscale images (a) show a complex cystic lesion (calipers) with some low-level
internal echoes located medial to the right renal hilum. Color Doppler was not performed. Axial bSSFP image (b) shows that the
mass (white arrow) is isointense to the IVC (thin white arrow). Axial T2W ssFSE image (c) shows flow void within the mass (white
arrow) and IVC (thin white arrow). Coronal oblique MIP image from MR angiogram (d) confirms the mass is a renal artery aneurysm.
FIGURE 8: Examples of clinical utility of bSSFP during MRCP. Top: An 83-year-old male undergoing MRCP to rule out common
bile duct (CBD) calculus. Axial T2W FSE image (a) demonstrates a prominent CBD (thick white arrow) with a central filling defect
(thin white arrow). Coronal bSSFP image (b) shows no corresponding filling defect in the CBD (thick white arrow). Coronal thick-
slab fat suppressed FS T2W FSE image (c) similarly does not show any filling defects in the CBD (thick white arrow). The apparent
filling defect in (a) is related to FSE flow artifact. Bottom: A 76-year-old female with CBD calculus. Note that the calculus (thin
white arrow) is located in the dependent portion of the CBD (thick white arrow) on axial T2W FSE (d) and is identified on both cor-
onal bSSFP (e) and FS heavily T2W coronal FSE images (f).
FIGURE 9: A 64-year-old male undergoing MRCP to evaluate elevated liver function tests. Coronal FS maximum intensity projec-
tion (MIP) imaged obtained from heavily T2W 3D FSE MRCP (a) shows an apparent stricture in the common hepatic duct (white
arrow). Sequential coronal bSSFP images (b) show the common hepatic artery (thin white arrows) crossing over the common
hepatic duct simulating the stricture in (a). The lack of upstream biliary dilation is another diagnostic feature of a crossing vessel
simulating a stricture.

FIGURE 10: A 73-year-old male with biopsy proven pancreatic adenocarcinoma undergoing MRI for preoperative planning.
Sequential coronal bSSFP images (a–c) at the level of the superior mesenteric artery (SMA) and pancreatic head show the SMA
posteriorly in (a) is completely encircled by a low signal line (etching artifact) which represents the fat water interface between
the vessel and preserved vascular cuff of fat (open white circle). More anteriorly (b,c) at the level of the pancreatic mass (white
arrow) the etching artifact surrounding the SMA is lost (black arrows) where the vessel interfaces with the tumor (white arrow) in
keeping with disruption of the vascular fat cuff and vascular invasion. Coronal FS 3D T1W GRE image obtained after gadolinium
injection (d) shows that the tumor (white arrow) is touching the SMA (black arrows) but only encircling 908 of its’ circumference.
Vascular invasion was confirmed at surgery.
FIGURE 11: A 43-year-old male with left flank pain undergoing MR for further evaluation. Axial T2W FSE image (a) shows fluid sur-
rounding the nondilated renal pelvis (thick white arrow) with an apparent filling defect in the left renal pelvis (thin white arrow). A
differential diagnosis of renal calculus or flow artifact from FSE was provided. In retrospect, review of coronal bSSFP image (b)
also shows the finding of a filling defect in the left renal pelvis (thin white arrow). Axial noncontrast-enhanced CT image (c) from
study performed later the same day shows a renal calculus in the left renal pelvis (white arrow). Also note simple cyst in the lower
pole of the left kidney (asterisk).
FIGURE 12: A 34-year-old male patient with Crohn’s disease postileo-colonic resection with end-to-side ileo-colonic anastomosis
undergoing MR enterography to assess for strictures. Coronal T2W ssFSE (a) image shows several potential abnormalities. There
is a long segment narrowing of the transverse colon (thin white arrow). There is focal narrowing at the ileo-colonic anastomosis
(thick white arrow). There are multiple filling defects in the ileum (arrowheads). Coronal bSSFP image (b) obtained from cine-MR
shows the transverse colon is now distended (thin white arrow) indicating that the finding in (a) was due to peristalsis. There is
persistent narrowing at the neo-terminal ileum (thick white arrow) which also demonstrated decreased peristalsis (not shown) in
keeping with a true stricture. The filling defects in the small bowel in (a) are no longer evident (b) and represented normal flow
artifact, which is commonly encountered at FSE. Reproduced with permission; the original publication is available at: http://link.
springer.com/article/10.1007%2Fs00261-015-0392-1
January 2017 17

FIGURE 13: A 27-year-old female patient 17 weeks gestational age with right lower quadrant pain imaged with MR to rule out appen-
dicitis. Sequential axial bSSFP images from caudal to cranial (a–d) show the dilated fluid filled appendix (thick white arrows) with inter-
nal appendicolith (thin white arrow in [c]) and surrounding free fluid (open white arrows). Appendicitis was confirmed at laparoscopy.
FIGURE 14: A 49-year-old female posthysterectomy status with pelvic floor dysfunction and known rectal prolapse imaged with
MR defecography to evaluate for intussusception and other pelvic floor abnormalities. Sagittal bSSFP image obtained in the mid-
line (a) after administration of ultrasound gel into the rectum shows the normal position of the rectum, vagina, and urinary blad-
der (r, v, ub, respectively) above the pubococcygeal line (dotted line). Sagittal bSSFP image (b) obtained during maximal defeca-
tion shows a small cystocele (open white arrow), a moderate-sized peritineocele (thick white arrow), rectal prolapse (thin white
arrow) and a low rectal intussusception (arrowheads).
FIGURE 15: A 64-year-old female undergoing pelvic MRI to rule out locally recurrent tumor status post-abdominoperineal resec-
tion with increasing carcinoembryonic antigen (CEA). Coronal bSSFP image (a) demonstrates moir e (fringe) artifact at the extreme
of the imaging field of view obscuring the liver (open white arrows). Coronal ssFSE image (b) obtained during the same examina-
tion shows a mass in liver (white arrow) which was obscured in (a). Axial contrast-enhanced CT (c) image confirms the presence of
large mass in the left lobe of the liver (white arrow) which was consistent with metastatic disease.
18 Volume 45, No. 1

FIGURE 16: Axial fat-suppressed bSSFP images (a,b) in a 43-year-old male undergoing MRCP to rule out biliary stricture demon-
strate banding artifacts which are commonly encountered with bSSFP. In (a) the artifact extends over the inferior vena cava
(arrowhead), while in (b) the artifact extends over the aorta (arrowheads). Banding artifacts may simulate pathologies such as inti-
mal flap in vascular dissection.
other adjacent bowel structures which may obscure the bili- with fast spoiled GRE, ssFSE, or bSSFP. bSSFP provides
ary tract. several advantages when imaging the pelvic floor compared
to other pulse sequences related to: high SNR, high spatial
Urinary Tract Applications resolution, and the increased signal intensity of fluid-like
Urine is of increased signal intensity on bSSFP and bSSP is
structures which can be used to opacify the rectum, small
invaluable when imaging the urinary tract because flowing
bowel, urinary bladder, and vagina (Fig. 14). F14
urine will not be mistaken for a potential filling defect such
F11 as renal calculi (Fig. 11). Imaging Artifacts in bSSFP
Artifacts encountered with bSSFP are important to identify
Gastrointestinal Tract Applications
bSSFP is a fundamental pulse sequence in the performance and understand as potential limitations in abdominal and
and evaluation of MR enterography and should be incor- pelvic MRI. bSSFP is a fundamentally a GRE type sequence
porated into routine MRE examinations.1,15 The high spa- and is vulnerable to susceptibility (T2 ) artifact, which may
tial resolution of bSSFP is complementary to ssFSE obscure important imaging findings. Moire (fringe) artifact
sequences that are typically of lower resolution due to the and banding artifacts that occur at the extremes of the imag-
image blur related to the extended echo train and are used ing FOV with bSSFP sequences may limit evaluation of
to identify areas of fistulization, sinus tracts, or deep ulcer- important findings that may be present in these locations
ation.1,16 bSSFP (or ssFSE) may be acquired dynamically (Fig. 15). Fat shows increased signal intensity on bSSFP, F15
to evaluate areas of narrowing as potential strictures versus which can be suboptimal, making areas of pathology more
F12 areas of normal peristalsis (Fig. 12).17,18 Dynamic bSSFP difficult to identify. Fat-suppressed bSSFP images can now
imaging can also be evaluated subjectively or quantitatively be generated on most commercial scanners by applying fre-
for areas of diminished peristalsis as an additional imaging quency selective RF preparatory pulses within a given TR or
finding of active inflammatory bowel disease.19 Potential magnetization preparation.23,24 India ink (etching) artifact
filling defects encountered on ssFSE can be easily inter- occurring at fat–water interfaces should be recognized on
preted in conjunction with bSSFP (Fig. 12). This is partic- bSSFP imaging. This artifact can be exploited to identify
ularly invaluable in patients who are evaluated for fat-containing structures and also, in our experience, can be
suspected small bowel polyps, where bSSFP should be con- used in conjunction with other sequences to determine
sidered mandatory.20 The low SAR of bSSFP, combined whether an expected fat–water interface may be obscured/
with high SNR and spatial resolution, make it an ideal invaded. A common misconception is that, like traditional
pulse sequence for the evaluation of suspected appendicitis chemical shift (in and opposed phase) GRE imaging, bSSFP
F13 in pregnancy (Fig. 13). images are opposed phase because they are acquired using
an opposed phase echo time (TE). Phase coherence or inco-
Pelvic Floor Applications herence with bSSFP is related to the spin-echo-like phase
Evaluation of the pelvic floor with MRI has several advan- coherence at TE 5 TR/2, which is determined by the pre-
tages compared to conventional fluoroscopic studies includ- cessional phase angle for the spins within one TR in the
ing the lack of ionizing radiation and the ability to image rotating frame, which is manipulated by the center reference
the soft tissue structures of the pelvis that are otherwise frequency.25 It is possible, therefore, to acquire in- or
occult with fluoroscopy.21,22 Accurate evaluation of the pel- opposed-phase bSSFP images by manipulating the center
vic floor requires dynamic imaging. This can be performed frequency.25 Lastly, in areas of turbulent or nonlinear flow,
January 2017 19

the insensitivity of bSSFP to dephasing effects may be 10. media ACodac. ACR Manual on Contrast Media Version 9. In: Radiol-
ogy ACo, editor; 2013.
degraded. This can result in unusual banding artifacts and
11. Glockner JF, Lee CU. Balanced steady state-free precession (b-SSFP)
areas of inhomogeneous signal intensity, which in certain imaging for MRCP: techniques and applications. Abdom Imaging
instances may simulate pathologies such as aortic dissection 2014;39:1309–1322.
F16 (Fig. 16). 12. Catalano C, Pavone P, Laghi A, et al. Pancreatic adenocarcinoma:
In conclusion, bSSFP offers several potential advan- combination of MR imaging, MR angiography and MR cholangiopan-
creatography for the diagnosis and assessment of resectability. Eur
tages compared to the more traditional fast/turbo spin-echo Radiol 1998;8:428–434.
and spoiled GRE sequences employed in abdominal/pelvic 13. Arslan A, Buanes T, Geitung JT. Pancreatic carcinoma: MR, MR angi-
MRI, including: high SNR, high resolution, rapid acquisi- ography and dynamic helical CT in the evaluation of vascular invasion.
Eur J Radiol 2001;38:151–159.
tion times, low SAR, bright blood effects, and relative insen-
sitivity to linear flow. Abdominal and pelvic MR 14. Lopez Hanninen E, Amthauer H, Hosten N, et al. Prospective evalua-
tion of pancreatic tumors: accuracy of MR imaging with MR cholan-
practitioners should be familiar with the fundamentals of giopancreatography and MR angiography. Radiology 2002;224:
bSSFP and applications in which bSSFP may improve their 34–41.
body MR practice. 15. Sinha R, Verma R, Verma S, Rajesh A. MR enterography of Crohn dis-
ease: part 1, rationale, technique, and pitfalls. AJR Am J Roentgenol
2011;197:76–79.
Conflict of Interest
16. Sinha R, Verma R, Verma S, Rajesh A. MR enterography of Crohn dis-
The authors declare no conflicts of interest. ease: part 2, imaging and pathologic findings. AJR Am J Roentgenol
2011;197:80–85.
References 17. Leyendecker JR, Bloomfeld RS, DiSantis DJ, Waters GS, Mott R,
1. Quon JS, Quon PR, Lim CS, Abdeen N, Schieda N. Magnetic reso- Bechtold RE. MR enterography in the management of patients with
nance enterography in post-operative inflammatory bowel disease. Crohn disease. Radiographics 2009;29:1827–1846.
Abdom Imaging 2015;40:1034–1049. 18. Allen BC, Leyendecker JR. MR enterography for assessment and man-
2. Hashemi H, Bradley WG, Lisanti CJ. MRI: The basics. Philadelphia: agement of small bowel Crohn disease. Radiol Clin North Am 2014;
Lippincott Williams & Wilkins: 2010. 52:799–810.
3. Chen Q, Stock KW, Prasad PV, Hatabu H. Fast imaging techniques for 19. Guglielmo FF, Mitchell DG, O’Kane PL, et al. Identifying decreased
body magnetic resonance imaging. Appl Radiol 2002:66–73. peristalsis of abnormal small bowel segments in Crohn’s disease using
cine MR enterography: the frozen bowel sign. Abdom Imaging 2015;
4. Bieri O, Scheffler K. Fundamentals of balanced steady state free pre- 40:1150–1156.
cession MRI. J Magn Reson Imaging 2013;38:2–11.
20. Anzidei M, Napoli A, Zini C, Kirchin MA, Catalano C, Passariello R.
5. Scheffler K, Lehnhardt S. Principles and applications of balanced SSFP Malignant tumours of the small intestine: a review of histopathology,
techniques. Eur Radiol 2003;13:2409–2418. multidetector CT and MRI aspects. Br J Radiol 2011;84:677–690.
6. Chavhan GB, Babyn PS, Jankharia BG, Cheng HL, Shroff MM. Steady- 21. Colaiacomo MC, Masselli G, Polettini E, et al. Dynamic MR imaging
state MR imaging sequences: physics, classification, and clinical appli- of the pelvic floor: a pictorial review. Radiographics 2009;29:e35.
cations. Radiographics 2008;28:1147–1160.
22. Law YM, Fielding JR. MRI of pelvic floor dysfunction: review. AJR Am
7. Kim BS, Angthong W, Jeon YH, Semelka RC. Body MR imaging: J Roentgenol 2008;191(6 Suppl):S45–53.
fast, efficient, and comprehensive. Radiol Clin North Am 2014;52:
623–636. 23. Borthne AS, Dormagen JB, Gjesdal KI, Storaas T, Lygren I, Geitung
JT. Bowel MR imaging with oral Gastrografin: an experimental study
8. Chung HW, Chen CY, Zimmerman RA, Lee KW, Lee CC, Chin SC. T2- with healthy volunteers. Eur Radiol 2003;13:100–106.
weighted fast MR imaging with true FISP versus HASTE: comparative
efficacy in the evaluation of normal fetal brain maturation. AJR Am J 24. Scheffler K, Heid O, Hennig J. Magnetization preparation during the
Roentgenol 2000;175:1375–1380. steady state: fat-saturated 3D TrueFISP. Magn Reson Med 2001;45:
1075–1080.
9. Lindquist CM, Karlicki F, Lawrence P, Strzelczyk J, Pawlyshyn N,
Kirkpatrick ID. Utility of balanced steady-state free precession MR 25. Huang TY, Chung HW, Wang FN, Ko CW, Chen CY. Fat and water
venography in the diagnosis of lower extremity deep venous throm- separation in balanced steady-state free precession using the Dixon
bosis. AJR Am J Roentgenol 2010;194:1357–1364. method. Magn Reson Med 2004;51:243–247.
20 Volume 45, No. 1

REVIEW ARTICLE
Whole-Body PET/MRI for Colorectal

Cancer Staging: Is It the Way Forward?
Dong Ho Lee, MD1,2 and Jeong Min Lee, MD1,2,3*
Magnetic resonance imaging (MRI) is presently the modality of choice for the local staging of rectal cancer, with posi-
tron emission tomography (PET) being optional for the evaluation of colorectal cancer. Indeed, previous studies have
demonstrated that liver MRI using hepatocyte-specific contrast agents can provide high diagnostic performance in the
detection of colorectal cancer liver metastases. Recently, however, whole-body PET/MRI, which can provide information
regarding both anatomy and metabolism, has been introduced to clinical imaging, and studies are under way to assess
whether it can improve diagnostic performance for oncologic diseases as well as provide additional information regard-
ing the disease phenotype and biology compared to conventional imaging modalities of computed tomography (CT),
PET, or MRI. This review offers a brief overview of the technical considerations of the PET/MRI system, and the current
status of imaging modalities in the staging of colorectal cancer. The potential of whole-body PET/MRI to improve the
performance of colorectal cancer staging and the results of several recent studies will be discussed, and workflow con-
siderations of whole-body PET/MRI for patients with colorectal cancer will be addressed.
C olorectal cancer is not only one of the most common

malignancies, but also the third leading cause of
cancer-related death worldwide.1,2 Thus, the correct staging
MRI is the modality of choice for this endeavor but several
functional MRI studies including diffusion-weighted imag-
ing (DWI) have been demonstrated to yield even better
of this disease based on imaging studies is of paramount diagnostic performance in the evaluation of the treatment
importance so as to determine the most appropriate treat- response to CRT than conventional MRI.8 PET may also
ment strategy. In this regard, the National Comprehensive provide information regarding the treatment response to
Cancer Network (NCCN) currently recommends intrave- CRT and the prognosis of patients with LARC.9–13 There-
nous contrast-enhanced chest/abdominal/pelvis computed fore, at present, imaging studies including both MRI and
tomography (CT) for the staging of colorectal cancer. In PET are used to assess tumor staging as well as to obtain
patients with rectal cancer, rectal magnetic resonance imag- information regarding the prognosis of patients with colo-
ing (MRI) is also recommended due to the excellent per- rectal cancer.
formance of MRI in evaluating the tumor border, In the midst of these diverse imaging techniques for
mesorectal fascia (MRF), pelvic lymph nodes, and the anal the evaluation of colorectal cancer, an integrated whole-
sphincter compared to CT or transrectal ultrasound body PET/MRI system which can provide both the high-
(TRUS).3–7 In addition, in cases of equivocal findings for resolution anatomic data of MRI and functional imaging
malignancies on contrast-enhanced CT, positron emission data of PET has recently been introduced.14–25 The clinical
tomography (PET)-CT has also been demonstrated to help advantages of PET/MRI relate predominantly to improve-
provide more accurate staging. ments provided by MRI over CT and the potential for
As for locally advanced rectal cancer (LARC), preoper- simultaneous PET and MRI acquisitions.25 Indeed, hybrid
ative chemoradiation therapy (CRT) has now become the PET/MRI can provide several advantages compared with
standard treatment option, and thus, restaging of rectal can- PET/CT: a lack of ionizing radiation, functional imaging
cer after CRT has been found to be increasing. At present, such as DWI or dynamic contrast-enhanced (DCE) MRI,
Received Apr 7, 2016, Accepted for publication May 24, 2016.
*Address reprint requests to: J.M.L., Department of Radiology, Seoul National University Hospital, 101 Daehangno, Jongno-gu,
Seoul, 110-744, Korea. E-mail: jmsh@snu.ac.kr
From the 1Department of Radiology, Seoul National University Hospital, Seoul, Korea; 2Seoul National University College of Medicine, Seoul, Korea; and
3
Institute of Radiation Medicine, Seoul National University Medical Research Center, Seoul, Korea.

V 21

TABLE 1. Whole-Body PET/MR Protocols for Patients With Colorectal Cancer
Whole-body PET/MR (25-30 min) Dedicated MR* (30 min)
PET AC with Dixon Diagnostic MR pulse Rectal MR High resolution T2-WI

sequence (19 sec) sequence
DWI
PET AC with Dixon Diagnostic MR pulse DCE-MRI
Liver MR with Heavily T2-WI and
hepatocyte- TSE/FSE
specific con-
trast agent
PET AC with Dixon Diagnostic MR pulse Dual echo T1-WI
Dynamic phases
PET AC with Dixon Diagnostic MR pulse Hepatobiliary phase
DWI
PET 5 positron emission tomography; MR 5 magnetic resonance imaging; AC 5 attenuation correction; WI 5 weighted image;
DWI 5 diffusion weighted imaging; DCE 5 dynamic contrast enhancement; * 5 dedicated MR protocol is selected according to the
aim of whole-body PET/MR study and individual patient need; TSE 5 turbo spin echo; FSE 5 fast spin echo.
the ability to capture cardiac or respiratory motion for bet- report we briefly address the usefulness and potential of the
ter registration, and improved soft-tissue contrast.25 How- whole-body PET/MRI system for the imaging of colorectal
ever, there are also several disadvantages of hybrid PET/ cancer, and review the results of early clinical experiences.
MRI systems, including 3 to 5 times higher cost than PET/
CT, as well as longer examination times and limited patient Technical considerations for Whole-Body
throughput. Therefore, although combining functional PET/MRI
information with anatomic information is an obvious To combine PET and MRI in the same patient, two main
advantage of PET/MRI over PET/CT, the imaging proto- strategies with different temporal associations between PET
cols are difficult to complete in the time required for PET/ and MRI have been explored: sequential and simultane-
CT acquisitions.22 To the contrary, the main disadvantages ous.33 In the sequential approach, PET and MRI are
of PET/CT are the relatively high radiation exposure from sequentially acquired in a completely independent manner.
the two components, low contrast resolution of noncontrast One of the pros of the sequential approach is that it can
CT for soft tissue, and the relatively low spatial resolution allow the use of state-of-the art time-of-flight (TOF) PET
of PET data.22,25 At present, in clinical practice, PET/MRI scanners. However, although accurate registration of these
is preferentially used over PET/CT in the following situa- independently acquired images are prerequisite, misregistra-
tions: 1) when imaging the brain, head and neck, abdomi- tion and its related artifacts can often arise from changes in
nal solid organs including the liver, the heart, and the the position of the organs during the two scannings.34,35 In
musculoskeletal system due to the better soft-tissue contrast addition, although this approach is relatively economical,
of the MRI component22,24,26–31; 2) when imaging children without the need for technical adjustments inherent to
and women of child-bearing age due to the reduced radia- incorporating different detectors into a single unit, it
tion exposure of PET/MRI; and 3) when the combined ana- requires a large space to house both the PET and MRI
tomic and functional capabilities of MRI together with scanners.35
molecular PET information is needed to provide new To the contrary, simultaneous scanning of PET and
insights into disease phenotypes and biology.22,32 Consider- MRI using the recently introduced fully integrated whole-
ing this theoretical background of whole-body PET/MRI, it body PET/MRI scanner (Biograph mMR, Siemens Health-
may not be far-fetched to assume that it may also provide care, Erlangen, Germany; and Signa PET/MRI, GE Health-
higher performance in the local staging of rectal cancer as care, Milwaukee, WI) can provide more exact anatomic and
well as in detecting distant metastases in patients with colo- temporal registration of the MRI and PET signals.24 Yet
rectal cancer compared to CT or MRI alone. Thus, in this with this technique, several technical obstacles remain to be
22 Volume 45, No. 1

Lee and Lee: Whole-Body PET/MRI for Colorectal Cancer
solved, including interference between the PET and MRI air) or four (water, fat, lung, and air) tissue classes are auto-
systems, motion correction of the PET data during scan- matically segmented without consideration of bone tissue,
ning, and the limited bore size of the magnet. Recent tech- and for four-class tissue segmentation, the Dixon MRI
nical advances in radiofrequency (RF) shielding, gradient sequence is used to differentiate water from fat tissue.44
coils, PET detectors such as avalanche photodiodes and sili- Therefore, reliable representation of bone attenuation based
con photomultiplier detectors, and large-bore magnets of on segmentation remains challenging due to the lack of an
sufficient gradient power have largely addressed these issues. MRI signal from the cortical bone and potential small errors
However, although simultaneous scanning of PET and MRI in quantitative measurement of standardized uptake values
reduces the misregistration between the two image sets, it is (SUV), although the effect of this limitation in clinical prac-
still challenging to match breath-held MR images with free- tice may be small.36 With regard to the quantitative accu-
breathing PET images.25 In abdominal imaging, this misre- racy of PET, previous studies have shown that although
gistration is most pronounced in studies of the upper abdo- SUVs of PET/MRI are highly correlated to values obtained
men, such as dedicated liver PET/MRI.35 Indeed, it has from PET/CT, the failure to segment the cortical bone cor-
previously been demonstrated that the combination of free- rectly can lead to 10–20% lower SUV measurements within
breathing PET data with breath-hold MRI sequences can the cortical bone using PET/MRI compared to PET/
lead to distortion in liver contours, attenuation correction CT.49–51 It should be noted, however, that this has not yet
errors, and imaging registration errors.35 Considering that been shown to change the clinical interpretation.29,42,52
MRI can provide better delineation of liver lesions or meta-
Despite these variations in SUV, PET/MRI has been shown
static disease and perfect registration may be of much
to be sufficiently robust in evaluating malignant bone
greater importance than in PET/CT, MRI-based nonrigid
lesions when T1-weighted turbo spin-echo images are
motion correction in simultaneous PET/MRI may be used
included for spinal assessment.53
to improve diagnostic performance for the evaluation of
metastases.25,36 An alternative approach to reduce misrgistra- Workflow Considerations for Whole-Body
tion could also be the use of free-breathing radial MRI PET/MRI in Patients with Colorectal Cancer
sequences and respiratory-gated PET acquisitions at a cost
Current whole-body PET/MRI generally consists of two
of longer examination time.37
parts: a whole-body PET/MRI part and a dedicated MRI
Another important issue that remains to be overcome
part (Table 1). In order to compete with whole-body PET/ T1
is the MRI-based attenuation correction of PET data. To
CT scans, which require 20 minutes or less of total exami-
generate artifact-free, quantitatively accurate PET images,
nation time, whole-body PET/MRI protocols ideally should
attenuation and scatter correction is of crucial impor-
not exceed 20–30 minutes.35,54 In addition, to obtain
tance.38,39 In the combined PET/CT scanner, a PET attenu-
answers for specific clinical questions in the evaluation of
ation map at 511 KeV is generated from X-ray CT images
abdominal organs and pelvic structures, dedicated MRI
using a piece-wire, linear transformation function.24,40,41 As
sequences should also be performed.
both the emitted 511 KeV photon pairs in the PET scan
Due to the limited z-axis coverage of PET detectors
and X-rays in the CT scan are attenuated during penetration
(25.8 cm) in the current integrated PET/MRI system, multi-
of the body structure, attenuation correction in PET/CT is
somewhat straightforward. However, MRI-based PET ple bed positions are required to acquire whole-body PET/
attenuation correction is much more complicated, as the MR images. Typically, MR images are obtained simultane-
MRI signal is largely determined by the tissue’s hydrogen ously with the PET images in 4–5 stations for vertex-to-
density and relaxation properties, not by the depth of body thigh coverage. Approximately 2–5 minutes of scan time are
structures.24,30,42–45 In addition, MRI RF coils placed above needed for PET imaging at each bed position, and the
the PET detectors in an integrated scanner can cause serious simultaneous MRI signal acquisition has to fit into this scan
attenuation of 511 KeV photons generated from the radio- time. As the Dixon-based MRI sequence for attenuation
tracers.24 Another technical challenge in MRI-based PET correction using the segment-based approach usually takes
attenuation correction is the incidence of truncation artifacts 19 seconds of acquisition time per bed position, two or
during PET reconstruction caused by the limited MRI field- three diagnostic MRI pulse sequences including heavily T2-
of-view.46–48 To this end, segmentation-based, attenuation weighted images, short-tau inversion-recovery (STIR) T2-
correction approaches are mostly used in commercially avail- weighted sequences, and 3D volumetric T1-weighted images
able clinical PET/MRI systems today. In the segmentation- with fat suppression are usually obtained during PET imag-
based attenuation correction approach, individual MR ing acquisition. Whole-body DWI may also be beneficiary
images are segmented into several tissue classes and known for detecting occult metastases in colorectal cancers; how-
attenuation coefficients are assigned to them,24,42–44 ie, for ever, it is not routinely used especially when workflow effi-
whole-body PET/MRI studies, three (soft-tissue, lung, and ciency is a major consideration.35 In total, the whole-body
January 2017 23

PET/MRI part of the examination usually requires 20–30 Local Staging of Colorectal Cancer
minutes for completion.35 T-Staging of Rectal Cancer
After completion of the whole-body PET/MRI part, Precise evaluation of the stage of rectal cancer using imaging
dedicated MRI at high resolution and contrast-enhanced studies is crucial for the determination of surgical resectabil-
MRI confined to the targeted organ is started with or with- ity and the need for neoadjuvant therapy.61 With recent
out simultaneous PET acquisition, depending on the indica- technical advances in phase-array multichannel coils, thin-
tion for the examination. MRI sequences in a dedicated section high-resolution rectal MRI has now become possi-
MRI protocol should be selected to provide maximum diag- ble, with studies reporting T-staging accuracies of 65% to
nostic and prognostic information with added clinical value 86%.62–68 High-resolution T2-weighted imaging is the key
to that of PET/CT or MRI alone in a given examination sequence for the T-staging of rectal cancer. On T2-weighted
time. For this purpose, the clinical situation of the patient MRI, the rectal wall mucosa shows low signal intensity, the
and the aim of the use of whole-body PET/MRI have to be submucosa shows high signal intensity, the muscularis prop-
considered prior to the examination, and the protocol of the ria shows low signal intensity, the mesorectal fat shows high
dedicated MRI part of whole-body PET/MRI should be signal intensity, and MRF shows linear low signal inten-
refined according to the needs of the individual patient. In sity.69 Based on this rectal wall layer differentiation on T2-
patients with rectal cancer, a dedicated rectal MRI protocol weighted imaging, the T-staging of rectal cancer can be
in which a pelvic phase-array multichannel coil providing made. The major limitation of rectal MRI for T-staging is
high spatial resolution with a high signal-to-noise ratio can the differentiation between T2 and T3 cancers. As an exam-
be used for exact local staging both initially and after CRT. ple, the desmoplastic reaction around the tumor can be seen
Its mandatory sequence is T2-weighted imaging with high as an irregular outer border of the rectal wall, and thus,
resolution which provides high spatial resolution for primary MRI can typically overstage T2 lesions as T3 lesions in such
rectal cancer T-staging and the assessment of the relation- cases.70 Another clear advantage of rectal MRI over other
ship between the tumor and MRF. In addition, DWI may imaging modalities including CT or TRUS is the assessment
help in the tumor detection and in the assessment of the of the circumferential resection margin (CRM) at the time
treatment response to CRT in patients with LARC.8,55 of total mesorectal excision (TME), which is currently the
Although the use of contrast agent is generally not recom- standard surgical treatment method for rectal cancer and the
mended in a rectal MRI protocol for local staging, there anal sphincter complex.7 MRF is the most important ana-
have been several studies reporting that perfusion MRI with tomic landmark for the feasibility of TME, which has
DCE using extracellular MRI agents can provide informa- evolved into the standard operative procedure for the resec-
tion regarding the treatment response to CRT as well as the tion of cancers located in the middle or lower third of the
status of the circumferential resection margin.56–60 Further- rectum.61 In this regard, previous studies have shown that
more, screening of focal liver lesions could also be done MRI can clearly show the relationship between tumors and
with the use of extracellular MRI contrast agents as well as MRF, and that this information can predict the CRM for
with the aid of PET and DWI sequences. TME.71–73 It has also been reported that recurrence rates
When colorectal cancer patients have lesions too small are significantly higher in patients with positive CRMs (19–
to characterize or suspected metastatic lesions in the liver on 22%) than in patients with negative CRMs (3–5%).74 Fur-
other imaging modalities, PET/MRI with dedicated liver thermore, the status of CRM has been known to be a signif-
MRI protocol using hepatocyte-specific contrast agents can icant predicting factor for local recurrence as well as for
be a good choice for the exact evaluation of colorectal can- overall survival after TME in rectal cancer.75 Therefore, pre-
cer liver metastases (CRLMs). Dedicated liver MRI sequen- operative assessment of CRM is of great clinical importance.
ces with hepatocyte-specific contrast agents can provide A recent European study of 408 patients with rectal cancer
maximum lesion-to-liver contrast on hepatobiliary phase who underwent rectal MRI using a body coil reported that
images, thereby significantly improving the diagnostic per- rectal MRI using a body coil can achieve 88% accuracy in
formance in detecting focal liver lesions. predicting a clear CRM at the time of TME.76 Therefore,
An additional consideration is whether to acquire a due to the high accuracy of rectal MRI in predicting a clear
separate, longer PET acquisition to match the MRI station, CRM, NCCN guidelines currently recommend rectal MRI
which may be helpful in pelvic malignancies to allow better as the initial imaging modality for rectal cancer T-staging.77
matching of bladder size and position of anatomical struc- In addition, high-resolution rectal MRI is also recom-
tures.25 Finally, for patient comfort and economic considera- mended as the first method of choice for restaging after pre-
tions, whole-body PET/MRI including a dedicated MRI operative chemoradiation, and MRI response to neoadjuvant
scanning should have a reasonable scan time, usually not therapy has been shown to be an indicator of long-term
exceeding 60 minutes.24 survival.78
24 Volume 45, No. 1

To the contrary, PET or PET/CT has been known to cancer,87,88 and did not significantly improve the diagnostic
be inappropriate for the accurate T-staging of rectal cancer, performance of MRI in colorectal cancer N-staging. There-
mainly due to their limited spatial resolution. In addition, it fore, a combination involving DWI would not be consid-
is known that CT can provide more detailed and accurate ered reliable in differentiating malignant from benign lymph
anatomical and structural information than PET; thus, nodes in patients with colorectal cancers.89
PET/CT for rectal cancer T-staging is almost completely As for PET/CT, the maximum standardized uptake
dependent on CT. However, as rectal MRI can provide value (SUVmax) of metastatic lymph nodes is usually more
information regarding T-staging as well as CRM, the dedi- than 2.0,69 and therefore, PET/CT can strengthen the possi-
cated high-resolution rectal MRI part of PET/MRI may be bility of metastatic lymph nodes seen on other imaging
better than PET/CT for the assessment of both the T- modalities. However, according to a recent meta-analysis the
staging of rectal cancer and CRM with high accuracy (Fig. pooled estimates of sensitivity and specificity of PET/CT
F1 1). Indeed, according to a recent study, the diagnostic accu- for preoperative colorectal cancer N-staging were only
racy of PET/MRI for rectal cancer T-staging was demon- 42.9% and 87.9%, respectively, and there was no definite
strated to be significantly higher than that of PET/CT due evidence to support the routine clinical use of PET/CT for
to the high resolution of MRI in assessing rectal cancer T- colorectal cancer N-staging.90
staging compared to CT.79 However, to determine the exact Based on these aforementioned study results, there
role of PET/MRI in rectal cancer T-staging, further studies would be a possibility that the dedicated MRI part of PET/
with a large number of patients are warranted. MRI can provide moderate accuracy in colorectal cancer N-
staging (Fig. 1) and the PET part can provide additional
N-Staging of Colorectal Cancer information regarding glucose metabolism, which can
Unlike the T-stage, N-stage is more challenging to assess via
strengthen the possibility of lymph node metastasis.79
imaging studies including MRI.25 Many previous studies
regarding the N-staging of colorectal cancer have adopted a Restaging After Neoadjuvant Chemoradiation
size criteria for the discrimination between malignant and Therapy for Rectal Cancer
benign lymph nodes, with lymph nodes >5 mm usually In rectal cancer, an advanced T-stage, the presence of lymph
being considered a metastatic lymph node.1 However, node metastasis, extramural venous invasion (EMVI), and a
according to a study done by Dworak, the mean diameter threatened or existing involvement of MRF are known to be
of benign lymph nodes was 3.34 mm and the mean diame- risk factors for local recurrence after TME as well as for
ter of metastatic lymph nodes was 3.84 mm.80 Therefore, poor overall survival.91 For patients with LARC, preopera-
application of a size criteria for the N-staging of colorectal tive CRT has improved survival, possibly due to the increas-
cancer has shown limited success, with only moderate sensi- ing rate of CRM-negative TME.92 Even though rectal
tivity and specificity for the detection of lymph node metas- cancer is not considered highly radiosensitive, preoperative
tasis81–83: CT demonstrated a sensitivity of 55% and a CRT often induces downstaging and increases R0 resection,
specificity of 74% and MRI showed a sensitivity of 66% and it has been reported that pathologic complete response
and a specificity of 76%, which is comparable to CT.1 can be achieved in 8–24% of patients with current CRT
Therefore, in addition to the size criteria, several morpho- regimens.93 In addition, pathologic complete response to
logic criteria have additionally been proposed to improve CRT is known to be a significant predicting factor for better
the diagnostic performance of imaging studies for colorectal overall and disease-free survival.94–98 Therefore, restaging
cancer N-staging. One such criterion is the irregular border imaging studies after preoperative CRT in patients with
of lymph nodes related to the extracapsular extension of the LRAC before planned TME are of clinical importance and
disease, which is more frequently found in metastatic lymph are being done with increasing frequency, along with the
nodes. Metastatic lymph nodes can also show heterogeneous increased use of CRT. During restaging, assessments of
signal intensity on T2-weighted MR images due to the MRF and the interval development of metastasis need to be
tumor foci within the involved node.84,85 With the aid of performed. For the restaging of rectal cancer after CRT, rec-
these morphologic criteria in addition to the size criteria, tal MRI has been considered the modality of choice, as it
higher accuracy with a sensitivity of 80–85% and a specific- can provide excellent soft-tissue contrast as well as informa-
ity of 97–98% was demonstrated with rectal MRI.84,86 tion regarding the relationship between the tumor and
Theoretically, the addition of DWI should also MRF. One previous study reported that MRI can provide
improve the diagnostic performance of MRI for colorectal 76% sensitivity and 86% specificity in the assessment of
cancer N-staging, as it can provide information regarding MRF in patients with rectal cancer who underwent preoper-
cellularity. However, DWI was shown to have moderate sen- ative CRT.99 However, the accuracy of rectal MRI for
sitivity (67–78%) and specificity (60–67%) for the detection restaging after CRT can be expected to be somewhat limited
of lymph node metastasis in patients with colorectal for the following reasons: the differentiation of residual
January 2017 25

FIGURE 1.
26 Volume 45, No. 1

tumors or tumor infiltration from desmoplastic reactions or PET/MRI can provide information regarding the morpho-
fibrosis is difficult; radiation proctitis can be misinterpreted logic changes of the tumor. In addition, when DWI is
as local invasion; and radiation therapy can induce benign added to the dedicated rectal MRI part of PET/MRI, cellu-
reactive lymph node hyperplasia within the mesorectum, larity information can be obtained. In combining this func-
which can be misdiagnosed as lymph node involvement.7,100 tional, morphological, and metabolic information obtained
In addition, rectal MRI would not be reliable to confirm from PET/MRI, more accurate assessment of residual
complete response after CRT as microscopic residual tumors after CRT may be possible. PET/MRI for rectal can-
tumors, which can be detected on histopathologic examina- cer restaging after CRT is thus considered to have the
tions, are hardly detected on MRI.8 Thus, to improve the potential to provide higher diagnostic performance com-
diagnostic accuracy of MRI in rectal cancer restaging after pared to MRI or PET only. A potential synergistic effect of
CRT, several functional MRI studies including DWI and PET and MRI could also be expected in assessing the
DCE-MRI have been examined. Theoretically, CRT can response to neoadjuvant therapy with the goal of predicting
induce tumor necrosis, fibrosis, and cellularity decreases, complete pathologic response in patients with rectal can-
which would result in an increase in the apparent diffusion cer.25,106 Furthermore, MRI assessment of tumor regression
coefficient (ADC) value on DWI, and therefore, DWI may may help predict survival, and reduction in tumor SUV fol-
help differentiate viable tumors from fibrosis after lowing CRT correlates with increased likelihood of complete
CRT.8,101–103 Indeed, according to a recent meta-analysis, pathologic response.25,78,106,107 Despite these results, the
DWI was demonstrated to be more sensitive (62–94%) and usefulness of PET/MRI in rectal cancer restaging after CRT
nearly as specific (74–91%) as conventional MRI in rectal has not been fully satisfactorily evaluated. Thus, for accurate
cancer restaging after CRT.99 Furthermore, the addition of evaluation of the clinical role of PET/MRI in rectal cancer
DWI to conventional MRI was also shown to improve the restaging after CRT, further studies with a larger population
prediction of tumor clearance in the MRF after CRT.55 As are warranted.
for DCE-MRI, several previous studies have reported that a
large decrease in the mean Ktrans (volume transfer constant) M-Staging of Colorectal Cancer
of the tumor after CRT was a significant predictor of good The liver is the most commonly involved site of distant
response in patients with LARC.59,104 metastasis in patients with colorectal cancer, with over half
In addition, as PET can provide information regarding of all colorectal cancer patients experiencing liver metastasis
glucose metabolism within the tumor, one previous study during their disease course.108,109 It is well known that cura-
reported that the decreased uptake of 18-fluorodeoxyglucose tive resection of CRLMs can provide a clear survival benefit.
(18-F FDG) after CRT was a significant prediction factor However, curative resection for CRLMs can be possible in
for long-term survival.105 PET can also differentiate residual only approximately one-third of patients with CRLMs,
tumors that show FDG uptake from fibrosis or hyperemia, mainly due to the extent of disease at detection.110,111 With
which usually does not show FDG uptake. According to recent advances in chemotherapy, however, initially unresect-
Denecke et al, they reported that PET is better than both able or borderline resectable CRLMs can now be resectable
CT and MRI for the prediction of tumor response to CRT after neoadjuvant chemotherapy (NAC), and the resultant
when the mean reduction in SUV after CRT was used.12 resection of CRLMs after NAC has already been demon-
The SUV value of tumors after CRT was also demonstrated strated to provide a clear survival benefit.112 Therefore,
in previous studies to be a predictor of the treatment exact assessment of the extent of CRLMs as well as the eval-
response to CRT, showing that the SUVmax of tumors that uation of concurrent extrahepatic metastases is of crucial
were downstaged after CRT was significantly lower than that clinical importance for proper management.
of nondownstaged tumors.10,11,13 At present, contrast-enhanced CT is most widely used
Therefore, the PET part of PET/MRI can provide for the diagnosis and follow-up after treatment in patients
C with CRLMs, and is also currently recommended by
O
metabolic information of tumors after CRT in patients with
L LARC, and the dedicated high-resolution rectal MRI part of NCCN guidelines. However, according to a recent meta-
O
R
FIGURE 1: PET/MRI images of a 54-year-old female showing the capacity of PET/MRI to assess rectal cancer staging. (a) T2-
weighted axial MR image shows segmental wall thickening of the upper rectum with perirectal fat infiltration (white arrow) indi-
cating a T3 stage rectal cancer. The mesorectal fascia (black arrow) is clearly visualized, and is free from the tumor. (b) On
diffusion-weighted image with a b-value of 1000, rectal wall thickening shows high signal intensity (arrow), indicating diffusion
restriction. (c) On fusion PET/MRI image, the lesion shows avid FDG uptake with an SUVmax of 7.3 (arrow), indicating a malignant
tumor. (d) On T2-weighted axial image, an enlarged lymph node (arrow) is seen above the rectal wall thickening. (e) On diffusion-
weighted image, the enlarged lymph node shows high signal intensity (arrow), suggesting diffusion restriction. (f) On fusion PET/
MRI image, the enlarged lymph node shows FDG uptake with an SUVmax of 5.4 (arrow), indicating a metastatic lymph node. With
the aid of these PET/MRI images, a preoperative diagnosis of T3N 1 could be made.
January 2017 27

FIGURE 2.
28 Volume 45, No. 1

analysis, MRI may be the preferred first-line modality for management plan in 17–25% of patients.122,123 Considering
the assessment of CRLMs owing to its excellent soft-tissue these study results, the dedicated liver MRI part of PET/
contrast.113 With PET or PET/CT, studies have shown that MRI can be considered to provide excellent diagnostic per-
small CRLMs can be missed using this modality.113,114 In formance for the detection of CRLMs when hepatocyte-
addition, NAC can induce tumor necrosis and decrease glu- specific contrast agents with a sophisticated protocol includ-
cose metabolism within the tumor, resulting in decreased ing DWI is used. In addition, the PET part of PET/MRI
FDG uptake. This decreased FDG uptake of tumors after can provide information regarding the presence of current
NAC, in turn, can cause a decrease in the lesion-to-liver extrahepatic metastases as well as prognosis after treatment
contrast on PET or PET/CT, and therefore, the accuracy of of CRLMs. Therefore, there is the potential for PET/MRI
PET or PET/CT for the evaluation of CRLMs can be lim- to be used as a “one-stop shop” tool for the evaluation of
ited in patients who had undergone NAC.115 A potential patients with CRLMs prior to treatment. Moreover, PET/
benefit of PET/MRI over PET/CT in the evaluation of MRI can also create better workflow for patients with
colorectal cancer is the better characterization of "too small CRLMs, as it can provide excellent diagnostic performance
to characterize" lesions or "indeterminate lesions" on CT.35 for CRLMs as well as information regarding whole-body
According to a previous study116 that compared PET/MRI metabolism and patient prognosis. Nevertheless, to assess
with unenhanced MRI to PET/contrast-enhanced CT the exact role of PET/MRI for patients with CRLMs, fur-
(CECT), PET/MRI resulted in similar diagnostic accuracy ther studies with a prospective design and a large number of
for the detection of liver metastases to PET/CECT. There- patients are warranted. Furthermore, the cost-effectiveness of
fore, several studies recommend the use of dynamic CE PET/MRI compared to that of standard contrast-enhanced
imaging sequences and DWI to improve the detection and CT with/without PET also remains to be analyzed in order
characterization of liver metastases.24,35 to arrive at more accurate guidelines for the imaging of
Recently, hepatocyte-specific contrast agents that can patients with CRLMs.
provide excellent liver-to-lesion contrast on hepatobiliary The lung is another frequently involved site of distant
phase images has been introduced in liver MRI, and studies metastasis in patients with colorectal cancer, with 5–15%
have reported excellent diagnostic performance of of patients with colorectal cancer having isolated lung
hepatocyte-specific contrast-enhanced liver MRI with the metastasis.124,125 With recent advances in chemotherapy and
addition of DWI for the evaluation of CRLMs.115,117,118 surgical techniques, many previous studies have reported
According to a recent randomized clinical trial, hepatocyte- that the resection of pulmonary metastases from colorectal
specific contrast-enhanced liver MRI was shown to be better cancer has a good prognosis, with a clear survival bene-
than contrast-enhanced CT and nonhepatocyte-specific fit.126,127 Therefore, detection of pulmonary metastases
contrast-enhanced liver MRI for the evaluation of would be another important step for the management of
CRLMs.119 Therefore, for the evaluation of CRLMs, colorectal cancer patients. However, there are several limita-
hepatocyte-specific contrast-enhanced liver MRI with DWI tions of pulmonary MRI in the detection of lung nodules,
F2 would provide the best diagnostic performance (Fig. 2). including a low signal-to-noise ratio in aerated lungs, sus-
However, PET may yet play a role, as it can provide infor- ceptibility artifacts generated from differences between air
mation regarding the prognosis of patients with CRLMs. and soft tissue, and motion-related artifacts from cardiac
Indeed, investigators have shown that decreased or absent and/or respiratory origins.128–130 As a result, the PET/MRI
FDG uptake after chemotherapy for CRLMs is a significant detection rate of lung nodules is currently lower than that
prediction factor for better treatment response as well as of PET/CT; Chandarana et al reported that the sensitivity
overall survival.45,120,121 In addition, PET can also help of PET/MRI was 70.3% for all lung nodules seen on PET/
detect concurrent extrahepatic metastases, as it provides CT.131 Considering this aforementioned technical back-
information regarding whole-body glucose metabolism ground and previous study results, there is concern in
F3 (Fig. 3). Therefore, although PET is not routinely recom- replacing the CT portion of PET/CT with MRI for the
C
O
mended for preoperative M-staging, recent studies have lung, due to the low rate of lung nodule detection.132,133
L reported that preoperative PET can indeed change the However, in a recent investigation by Raad et al, they
O
R
FIGURE 2: PET/MRI images of a 52-year-old female showing the ability of PET/MRI to detect small liver metastasis. (a) Contrast-
enhanced portal venous phase axial CT image shows no focal lesion in segment VII of the liver. (b) However, arterial phase gadox-
etic acid-enhanced axial MR image of PET/MRI shows an 8 mm peripheral enhancing small nodular lesion in segment VII of the
liver (arrow). (c) On hepatobiliary phase image of PET/MRI, this lesion shows low signal intensity (arrow). (d) This lesion shows
high signal intensity on diffusion-weighted image with a b-value of 1000 (arrow), indicating diffusion restriction. (e) On fusion
PET/MRI image, this lesion shows iso-metabolism compared to the adjacent liver parenchyma, probably due to its small size. (f)
Hepatic resection was done for the small nodule in segment VII detected only on PET/MRI, and a final diagnosis of metastasis
(arrow) was made through histopathologic examination.
January 2017 29

FIGURE 3.
30 Volume 45, No. 1

reported that the majority of lung nodules missed on PET/ regarding the prognosis and response to CRT in patients
MRI was due to their small size, with a mean size of 4 mm, with LARC. In a recent study by Kang et al comparing
and the appearance of benignity showing either disappear- PET/MRI with CECT in 51 patients with colorectal can-
ance or no change on follow-up imaging.130 Therefore, cer,135 PET/MRI added value to CECT in 14 of 51 patients
small lung nodules that can be missed on PET/MRI might (27.5%) in terms of better characterization (12/51 [23.5%])
have little clinical significance in patients with colorectal and additional detection (2/51 [3.9%]) of extracolonic
cancer. In addition, the recently introduced free-breathing lesions, which led to a change in the treatment strategy in
ultrashort echo time (UTE) sequence demonstrated signifi- 11 of 51 (21.6%) patients. Therefore, they concluded that
cantly higher sensitivity for the detection of lung nodules integrated whole-body PET/MRI added value to CECT in
compared to the conventional 3D GRE technique, especially the detection of metastatic lesions and the characterization
for small nodules less than 5 mm.134 Thus, to more conclu- of indeterminate lesions, aiding in the selection of more
sively assess the role of PET/MRI in detecting lung nodules appropriate treatment strategies for patients with colorectal
in patients with colorectal cancer, further studies with a pro- cancer.135
spective design and the use of cutting-edge techniques As for the liver, it is known to be one of the organs in
including UTE as well as comparison with PET/CT are which MRI can offer better diagnostic performance in the
warranted. detection and characterization of focal lesions compared to
Theoretically, whole-body PET/MRI may improve the CT, especially when using hepatocyte-specific contrast
diagnostic performance of abdomino-pelvic lymph node agents. Currently, hepatocyte-specific contrast-enhanced liver
metastases in patients with colorectal cancer compared with MRI with sophisticated DWI is considered a state-of-art
CECT or PET alone, as PET/MRI can provide both ana- imaging tool for the evaluation of CRLMs. By using a liver
tomic and functional information. However, until now there MRI protocol with hepatocyte-specific contrast agents and
has been no trial assessing this issue. Bone is another site DWI in the dedicated MRI part of PET/MRI, PET/MRI
where PET/MRI can provide better diagnostic performance can offer better diagnostic performance in the detection of
compared to PET/CT, as MRI can provide superior soft- CRLMs compared to CT or PET/CT. In addition, the
tissue contrast of bony lesion to CT. However, evidence of whole-body PET/MRI part of PET/MRI can help radiolog-
PET/MRI to assess bone lesion superiority to PET/CT is ists determine whether or not concurrent extrahepatic
also limited. Further studies with a prospective design and a metastasis is present, which would be important in the man-
large number of patients are warranted in order to assess the agement planning of patients with CRLMs. Furthermore,
value of PET/MRI in colorectal cancer M-staging. the degree of FDG uptake assessed using PET can provide
additional information regarding the prognosis and response
Clinical Potential of the Whole-Body to treatment. Therefore, there is the potential for whole-
PET/MRI System for Patients with body PET/MRI to be used as a "one-stop shop" tool for the
Colorectal Cancer evaluation of CRLMs. For this purpose, dedicated MRI pro-
The whole-body PET/MRI system can be better than PET/ tocols should be refined according to the aim of study and
CT due to the superior soft-tissue contrast of MRI, which each individual patient’s needs. The cost-effectiveness of
provides clear advantages.25 In cases of rectal cancer, MRI PET/MRI is another important issue that should be
can offer better soft-tissue contrast for the evaluation of the addressed, as the routine use of PET or liver MRI for colo-
rectal wall and MRF, and PET/MRI may also be superior to rectal cancer staging is not currently recommended.
PET/CT in rectal cancer T-staging.35 In addition, as the Finally, the lung is another frequently involved organ
whole-body PET/MRI part of PET/MRI can provide infor- of distant metastasis in patients with colorectal cancer. How-
mation regarding the presence of concurrent distant metas- ever, the PET/MRI detection rate of small lung nodules has
tasis, PET/MRI can also be more beneficial for rectal cancer been known to be somewhat lower than that of PET/CT,
staging compared to rectal MRI only. Furthermore, the mul- which can be a limitation for replacing the CT portion of
C
O
tiparametric capability of PET/MRI including DWI and PET/CT with MRI in colorectal cancer M-staging. The
L DCE-MRI may have the potential to provide information recently introduced UTE sequence, however, may improve
O
R
FIGURE 3: PET/MRI images of a 58-year-old male showing the ability of PET/MRI to detect distant metastasis. (a) T2-weighted
axial image shows an approximately 6 cm metastasis in the right lobe of the liver. (b) On hepatobiliary phase image, the metasta-
sis shows low signal intensity. (c) On fusion PET/MRI image, the metastasis shows FDG uptake with an SUVmax of 6.2, indicating
a malignant lesion. (d) In addition to the liver metastasis, coronal whole-body fusion PET/MRI image shows an enlarged lymph
node in the left supraclavicular area showing FDG uptake with an SUVmax of 5.3 (arrow), suggesting a metastatic lymph node. (e)
Axial whole-body fusion PET/MRI image also shows an enlarged lymph node in the left supraclavicular area with FDG uptake
(arrow). US-guided biopsy was done for the left supraclavicular lymph node enlargement, and a final diagnosis of metastatic
lymph node was made through histopathologic examination.
January 2017 31

the detection rate of small lung nodules on PET/MRI, and 12. Denecke T, Rau B, Hoffmann KT, et al. Comparison of CT, MRI and
FDG-PET in response prediction of patients with locally advanced
thus, requires additional investigation. rectal cancer after multimodal preoperative therapy: is there a bene-
In conclusion, data on the diagnostic performance of fit in using functional imaging? Eur Radiol 2005;15:1658–1666.
PET/MRI in the evaluation of colorectal cancer are still lim- 13. Kalff V, Duong C, Drummond EG, Matthews JP, Hicks RJ. Findings
on 18F-FDG PET scans after neoadjuvant chemoradiation provides
ited. However, considering that MRI is the anatomic imag-
prognostic stratification in patients with locally advanced rectal carci-
ing modality of choice in rectal cancer and colorectal liver noma subsequently treated by radical surgery. J Nucl Med 2006;47:
metastases, it would be reasonable to convert PET/CT scans 14–22.
to PET/MRI studies if scan durations can be tolerable at 14. Antoch G, Bockisch A. Combined PET/MRI: a new dimension in
whole-body oncology imaging? Eur J Nucl Med Mol Imaging 2009;
acceptable costs. Integrated whole-body PET/MRI has 36(Suppl 1):S113–120.
already been shown to provide added value to CECT in the 15. von Schulthess GK, Schlemmer HP. A look ahead: PET/MR versus
detection of metastatic lesions and the characterization of PET/CT. Eur J Nucl Med Mol Imaging 2009;36(Suppl 1):S3–9.
indeterminate lesions, and may aid in the selection of more 16. Wehrl HF, Sauter AW, Judenhofer MS, Pichler BJ. Combined PET/
appropriate treatment strategies for patients with colorectal MR imaging—technology and applications. Technol Cancer Res
Treat 2010;9:5–20.
cancer. Moreover, the restaging of rectal cancer and the eval-
17. Zaidi H, Montandon ML, Alavi A. The clinical role of fusion imaging
uation of CRLMs after chemotherapy appears to be areas in using PET, CT, and MR imaging. Magn Reson Imaging Clin N Am
which whole-body PET/MRI may also be better than con- 2010;18:133–149.
ventional imaging modalities such as PET or MR/CT only. 18. Balyasnikova S, Lofgren J, de Nijs R, Zamogilnaya Y, Hojgaard L,
Fischer BM. PET/MR in oncology: an introduction with focus on MR
Therefore, it can be expected that the combination of func-
and future perspectives for hybrid imaging. Am J Nucl Med Mol
tional MRI tools and the molecular capability of PET may Imaging 2012;2:458–474.
result in the substantial contribution of PET/MRI in tumor 19. Catalano OA, Rosen BR, Sahani DV, et al. Clinical impact of PET/MR
phenotyping, prediction of response to treatment, and treat- imaging in patients with cancer undergoing same-day PET/CT: initial
experience in 134 patients—a hypothesis-generating exploratory
ment monitoring in patients with colorectal cancer. study. Radiology 2013;269:857–869.
20. Catana C, Guimaraes AR, Rosen BR. PET and MR imaging: the odd
References couple or a match made in heaven? J Nucl Med 2013;54:815–824.
1. Koh DM, Brown G, Husband JE. Nodal staging in rectal cancer. 21. Gaertner FC, Furst S, Schwaiger M. PET/MR: a paradigm shift. Can-
Abdom Imaging 2006;31:652–659. cer Imaging 2013;13:36–52.
2. Adam R, De Gramont A, Figueras J, et al. The oncosurgery approach 22. Czernin J, Ta L, Herrmann K. Does PET/MR Imaging Improve Cancer
to managing liver metastases from colorectal cancer: a multidiscipli- Assessments? Literature Evidence from More Than 900 Patients.
nary international consensus. Oncologist 2012;17:1225–1239. J Nucl Med 2014;55(Suppl 2):59S–62S.
3. Brown G, Kirkham A, Williams GT, et al. High-resolution MRI of the 23. Jadvar H, Colletti PM. Competitive advantage of PET/MRI. Eur J
anatomy important in total mesorectal excision of the rectum. AJR Radiol 2014;83:84–94.
Am J Roentgenol 2004;182:431–439.
24. Yoo HJ, Lee JS, Lee JM. Integrated whole body MR/PET: where are
4. Rao SX, Zeng MS, Xu JM, et al. Assessment of T staging and meso- we? Korean J Radiol 2015;16:32–49.
rectal fascia status using high-resolution MRI in rectal cancer with
rectal distention. World J Gastroenterol 2007;13:4141–4146. 25. Sotoudeh H, Sharma A, Fowler KJ, McConathy J, Dehdashti F. Clini-
cal application of PET/MRI in oncology. J Magn Reson Imaging 2016
5. Giusti S, Buccianti P, Castagna M, et al. Preoperative rectal cancer
[Epub ahead of print].
staging with phased-array MR. Radiat Oncol 2012;7:29.
26. Boss A, Stegger L, Bisdas S, et al. Feasibility of simultaneous PET/
6. Ucar A, Obuz F, Sokmen S, et al. Efficacy of high resolution magnetic
MR imaging in the head and upper neck area. Eur Radiol 2011;21:
resonance imaging in preoperative local staging of rectal cancer.
1439–1446.
Mol Imaging Radionucl Ther 2013;22:42–48.
27. Buchbender C, Heusner TA, Lauenstein TC, Bockisch A, Antoch G.
7. Jhaveri KS, Hosseini-Nik H. MRI of rectal cancer: an overview and
Oncologic PET/MRI, part 2: bone tumors, soft-tissue tumors, mela-
update on recent advances. AJR Am J Roentgenol 2015;205:W42–
noma, and lymphoma. J Nucl Med 2012;53:1244–1252.
55.
28. Buchbender C, Heusner TA, Lauenstein TC, Bockisch A, Antoch G.
8. Kim SH, Lee JM, Hong SH, et al. Locally advanced rectal cancer:
Oncologic PET/MRI, part 1: tumors of the brain, head and neck,
added value of diffusion-weighted MR imaging in the evaluation of
chest, abdomen, and pelvis. J Nucl Med 2012;53:928–938.
tumor response to neoadjuvant chemo- and radiation therapy. Radi-
ology 2009;253:116–125. 29. Al-Nabhani KZ, Syed R, Michopoulou S, et al. Qualitative and quanti-
tative comparison of PET/CT and PET/MR imaging in clinical prac-
9. Oku S, Nakagawa K, Momose T, et al. FDG-PET after radiotherapy is
tice. J Nucl Med 2014;55:88–94.
a good prognostic indicator of rectal cancer. Ann Nucl Med 2002;
16:409–416. 30. Andersen FL, Ladefoged CN, Beyer T, et al. Combined PET/MR
imaging in neurology: MR-based attenuation correction implies a
10. Calvo FA, Domper M, Matute R, et al. 18F-FDG positron emission
strong spatial bias when ignoring bone. NeuroImage 2014;84:206–
tomography staging and restaging in rectal cancer treated with pre-
216.
operative chemoradiation. Int J Radiat Oncol Biol Phys 2004;58:528–
535. 31. Bailey DL, Barthel H, Beuthin-Baumann B, et al. Combined PET/MR:
Where are we now? Summary report of the second international
11. Guillem JG, Moore HG, Akhurst T, et al. Sequential preoperative
workshop on PET/MR imaging April 8-12, 2013, Tubingen, Germany.
fluorodeoxyglucose-positron emission tomography assessment of
Mol Imaging Biol 2014;16:295–310.
response to preoperative chemoradiation: a means for determin-
ing longterm outcomes of rectal cancer. J Am Coll Surg 2004; 32. Czernin J, Herrmann K. The potential of PET/MRI imaging in oncol-
199:1–7. ogy: a comment to a summary report of the First PET/MRI Workshop
32 Volume 45, No. 1

in Tuebingen in 2012. Mol Imaging Biol the official publication of bone and liver standardized uptake value with oncologic PET/CT
the Academy of Molecular Imaging 2013;15:372–373. data. J Nucl Med 2012;53:1878–1882.
33. von Schulthess GK, Veit-Haibach P. Workflow considerations in PET/ 53. Eiber M, Takei T, Souvatzoglou M, et al. Performance of whole-body
MR imaging. J Nucl Med 2014;55(Suppl 2):19S–24S. integrated 18F-FDG PET/MR in comparison to PET/CT for evaluation
of malignant bone lesions. J Nucl Med 2014;55:191–197.
34. Vaska P, Cao T. The state of instrumentation for combined positron
emission tomography and magnetic resonance imaging. Semin Nucl 54. Fowler KJ, McConathy J, Narra VR. Whole-body simultaneous posi-
Med 2013;43:11–18. tron emission tomography (PET)-MR: optimization and adaptation of
MRI sequences. J Magn Reson Imaging 2014;39:259–268.
35. Fraum TJ, Fowler KJ, McConathy J, et al. PET/MRI for the body
imager: abdominal and pelvic oncologic applications. Abdom Imag- 55. Park MJ, Kim SH, Lee SJ, Jang KM, Rhim H. Locally advanced rectal
ing 2015;40:1387–1404. cancer: added value of diffusion-weighted MR imaging for predicting
tumor clearance of the mesorectal fascia after neoadjuvant chemo-
36. Ouyang J, Li Q, El Fakhri G. Magnetic resonance-based motion cor-
therapy and radiation therapy. Radiology 2011;260:771–780.
rection for positron emission tomography imaging. Semin Nucl Med
2013;43:60–67. 56. Lim JS, Kim D, Baek SE, et al. Perfusion MRI for the prediction of
treatment response after preoperative chemoradiotherapy in locally
37. Pepin A, Daouk J, Bailly P, Hapdey S, Meyer ME. Management of advanced rectal cancer. Eur Radiol 2012;22:1693–1700.
respiratory motion in PET/computed tomography: the state of the
art. Nucl Med Commun 2014;35:113–122. 57. Gollub MJ, Cao K, Gultekin DH, et al. Prognostic aspects of DCE-
MRI in recurrent rectal cancer. Eur Radiol 2013;23:3336–3344.
38. Bailey DL. Transmission scanning in emission tomography. Eur J
Nucl Med 1998;25:774–787. 58. Oberholzer K, Menig M, Pohlmann A, et al. Rectal cancer: assess-
ment of response to neoadjuvant chemoradiation by dynamic
39. Zaidi H, Hasegawa B. Determination of the attenuation map in emis- contrast-enhanced MRI. J Magn Reson Imaging 2013;38:119–126.
sion tomography. J Nucl Med 2003;44:291–315.
59. Kim SH, Lee JM, Gupta SN, Han JK, Choi BI. Dynamic contrast-
40. Burger C, Goerres G, Schoenes S, Buck A, Lonn AH, Von Schulthess enhanced MRI to evaluate the therapeutic response to neoadjuvant
GK. PET attenuation coefficients from CT images: experimental eval- chemoradiation therapy in locally advanced rectal cancer. J Magn
uation of the transformation of CT into PET 511-keV attenuation Reson Imaging 2014;40:730–737.
coefficients. Eur J Nucl Med Mol Imaging 2002;29:922–927.
60. Lollert A, Junginger T, Schimanski CC, et al. Rectal cancer: dynamic
41. Kinahan PE, Hasegawa BH, Beyer T. X-ray-based attenuation correc- contrast-enhanced MRI correlates with lymph node status and epi-
tion for positron emission tomography/computed tomography scan- dermal growth factor receptor expression. J Magn Reson Imaging
ners. Semin Nucl Med 2003;33:166–179. 2014;39:1436–1442.
42. Eiber M, Martinez-Moller A, Souvatzoglou M, et al. Value of a Dixon- 61. Klessen C, Rogalla P, Taupitz M. Local staging of rectal cancer: the
based MR/PET attenuation correction sequence for the localization current role of MRI. Eur Radiol 2007;17:379–389.
and evaluation of PET-positive lesions. Eur J Nucl Med Mol Imaging
2011;38:1691–1701. 62. Beets-Tan RG, Beets GL, van der Hoop AG, et al. High-resolution
magnetic resonance imaging of the anorectal region without an
43. Keereman V, Fierens Y, Broux T, De Deene Y, Lonneux M, endocoil. Abdom Imaging 1999;24:576–581; discussion 582–574.
Vandenberghe S. MRI-based attenuation correction for PET/MRI
using ultrashort echo time sequences. J Nucl Med 2010;51:812–818. 63. Brown G, Richards CJ, Newcombe RG, et al. Rectal carcinoma: thin-
section MR imaging for staging in 28 patients. Radiology 1999;211:
44. Martinez-Moller A, Souvatzoglou M, Delso G, et al. Tissue classifica- 215–222.
tion as a potential approach for attenuation correction in whole-
body PET/MRI: evaluation with PET/CT data. J Nucl Med 2009;50: 64. Blomqvist L, Machado M, Rubio C, et al. Rectal tumour staging: MR
520–526. imaging using pelvic phased-array and endorectal coils vs endo-
scopic ultrasonography. Eur Radiol 2000;10:653–660.
45. Findlay M, Young H, Cunningham D, et al. Noninvasive monitoring
of tumor metabolism using fluorodeoxyglucose and positron emis- 65. Beets-Tan RG, Beets GL, van der Hoop AG, et al. Preoperative MR
sion tomography in colorectal cancer liver metastases: correlation imaging of anal fistulas: Does it really help the surgeon? Radiology
with tumor response to fluorouracil. J Clin Oncol 1996;14:700–708. 2001;218:75–84.
46. Delso G, Martinez-Moller A, Bundschuh RA, et al. Evaluation of the 66. Beets-Tan RG, Beets GL, Vliegen RF, et al. Accuracy of magnetic res-
attenuation properties of MR equipment for its use in a whole-body onance imaging in prediction of tumour-free resection margin in rec-
PET/MR scanner. Phys Med Biol 2010;55:4361–4374. tal cancer surgery. Lancet 2001;357:497–504.
47. Delso G, Martinez-Moller A, Bundschuh RA, Nekolla SG, Ziegler SI. 67. Beets-Tan RG, Morren GL, Beets GL, et al. Measurement of anal
The effect of limited MR field of view in MR/PET attenuation correc- sphincter muscles: endoanal US, endoanal MR imaging, or phased-
array MR imaging? A study with healthy volunteers. Radiology 2001;
tion. Med Phys 2010;37:2804–2812.
220:81–89.
48. MacDonald LR, Kohlmyer S, Liu C, Lewellen TK, Kinahan PE. Effects
68. Gagliardi G, Bayar S, Smith R, Salem RR. Preoperative staging of rec-
of MR surface coils on PET quantification. Med Phys 2011;38:2948–
tal cancer using magnetic resonance imaging with external phase-
2956.
arrayed coils. Arch Surg 2002;137:447–451.
49. Samarin A, Burger C, Wollenweber SD, et al. PET/MR imaging of
69. Kijima S, Sasaki T, Nagata K, Utano K, Lefor AT, Sugimoto H. Preop-
bone lesions—implications for PET quantification from imperfect
erative evaluation of colorectal cancer using CT colonography, MRI,
attenuation correction. Eur J Nucl Med Mol Imaging 2012;39:1154–
and PET/CT. World J Gastroenterol 2014;20:16964–16975.
1160.
70. Beets-Tan RG, Beets GL. Rectal cancer: review with emphasis on MR
50. Aznar MC, Sersar R, Saabye J, et al. Whole-body PET/MRI: the effect
imaging. Radiology 2004;232:335–346.
of bone attenuation during MR-based attenuation correction in
oncology imaging. Eur J Radiol 2014;83:1177–1183. 71. Taylor A, Slater A, Mapstone N, Taylor S, Halligan S. Staging rectal
cancer: MRI compared to MDCT. Abdom Imaging 2007;32:323–327.
51. Wiesmuller M, Quick HH, Navalpakkam B, et al. Comparison of
lesion detection and quantitation of tracer uptake between PET from 72. Maizlin ZV, Brown JA, So G, et al. Can CT replace MRI in preopera-
a simultaneously acquiring whole-body PET/MR hybrid scanner and tive assessment of the circumferential resection margin in rectal can-
PET from PET/CT. Eur J Nucl Med Mol Imaging 2013;40:12–21. cer? Dis Colon Rectum 2010;53:308–314.
52. Kim JH, Lee JS, Song IC, Lee DS. Comparison of segmentation- 73. Maier A, Fuchsjager M. Preoperative staging of rectal cancer. Eur J
based attenuation correction methods for PET/MRI: evaluation of Radiol 2003;47:89–97.
January 2017 33

74. Gowdra Halappa V, Corona Villalobos CP, Bonekamp S, et al. Rectal for rectal cancer: a pooled analysis of individual patient data. Lancet
imaging: part 1, High-resolution MRI of carcinoma of the rectum at Oncol 2010;11:835–844.
3 T. AJR Am J Roentgenol 2012;199:W35–42.
94. Janjan NA, Crane C, Feig BW, et al. Improved overall survival
75. Taylor FG, Quirke P, Heald RJ, et al. Preoperative magnetic reso- among responders to preoperative chemoradiation for locally
nance imaging assessment of circumferential resection margin pre- advanced rectal cancer. Am J Clin Oncol 2001;24:107–112.
dicts disease-free survival and local recurrence: 5-year follow-up
95. Theodoropoulos G, Wise WE, Padmanabhan A, et al. T-level down-
results of the MERCURY study. J Clin Oncol 2014;32:34–43.
staging and complete pathologic response after preoperative che-
76. Diagnostic accuracy of preoperative magnetic resonance imaging in moradiation for advanced rectal cancer result in decreased
predicting curative resection of rectal cancer: prospective observatio- recurrence and improved disease-free survival. Dis Colon Rectum
nal study. BMJ 2006;333:779. 2002;45:895–903.
77. Benson AB, 3rd, Venook AP, Bekaii-Saab T, et al. Rectal cancer, ver- 96. Garcia-Aguilar J, Hernandez de Anda E, Sirivongs P, Lee SH, Madoff
sion 2.2015. J Natl Compr Canc Netw 2015;13:719–728; quiz 728. RD, Rothenberger DA. A pathologic complete response to preopera-
tive chemoradiation is associated with lower local recurrence and
78. Patel UB, Taylor F, Blomqvist L, et al. Magnetic resonance imaging- improved survival in rectal cancer patients treated by mesorectal
detected tumor response for locally advanced rectal cancer predicts sur- excision. Dis Colon Rectum 2003;46:298–304.
vival outcomes: MERCURY experience. J Clin Oncol 2011;29:3753–
3760. 97. Hartley A, Ho KF, McConkey C, Geh JI. Pathological complete
response following pre-operative chemoradiotherapy in rectal can-
79. Paspulati RM, Partovi S, Herrmann KA, Krishnamurthi S, Delaney CP, cer: analysis of phase II/III trials. Br J Radiol 2005;78:934–938.
Nguyen NC. Comparison of hybrid FDG PET/MRI compared with
PET/CT in colorectal cancer staging and restaging: a pilot study. 98. Stipa F, Chessin DB, Shia J, et al. A pathologic complete response
Abdom Imaging 2015;40:1415–1425. of rectal cancer to preoperative combined-modality therapy results
in improved oncological outcome compared with those who achieve
80. Dworak O. Morphology of lymph nodes in the resected rectum of no downstaging on the basis of preoperative endorectal ultrasonog-
patients with rectal carcinoma. Pathol Res Pract 1991;187:1020– raphy. Ann Surg Oncol 2006;13:1047–1053.
1024.
99. van der Paardt MP, Zagers MB, Beets-Tan RG, Stoker J, Bipat S.
81. Hadfield MB, Nicholson AA, MacDonald AW, et al. Preoperative Patients who undergo preoperative chemoradiotherapy for locally
staging of rectal carcinoma by magnetic resonance imaging with a advanced rectal cancer restaged by using diagnostic MR imaging: a
pelvic phased-array coil. Br J Surg 1997;84:529–531. systematic review and meta-analysis. Radiology 2013;269:101–112.
82. Kim NK, Kim MJ, Park JK, Park SI, Min JS. Preoperative staging of 100. Del Vescovo R, Trodella LE, Sansoni I, et al. MR imaging of rectal
rectal cancer with MRI: accuracy and clinical usefulness. Ann Surg cancer before and after chemoradiation therapy. Radiol Med 2012;
Oncol 2000;7:732–737. 117:1125–1138.
83. Matsuoka H, Nakamura A, Sugiyama M, Hachiya J, Atomi Y, Masaki 101. Lambregts DM, Vandecaveye V, Barbaro B, et al. Diffusion-weighted
T. MRI diagnosis of mesorectal lymph node metastasis in patients MRI for selection of complete responders after chemoradiation for
with rectal carcinoma. what is the optimal criterion? Anticancer Res locally advanced rectal cancer: a multicenter study. Ann Surg Oncol
2004;24:4097–4101. 2011;18:2224–2231.
84. Brown G, Richards CJ, Bourne MW, et al. Morphologic predictors of 102. Barbaro B, Vitale R, Valentini V, et al. Diffusion-weighted magnetic
lymph node status in rectal cancer with use of high-spatial-resolution resonance imaging in monitoring rectal cancer response to neoadju-
MR imaging with histopathologic comparison. Radiology 2003;227: vant chemoradiotherapy. Int J Radiat Oncol Biol Phys 2012;83:594–
371–377. 599.
85. Kim JH, Beets GL, Kim MJ, Kessels AG, Beets-Tan RG. High-resolu- 103. Jung SH, Heo SH, Kim JW, et al. Predicting response to neoadjuvant
tion MR imaging for nodal staging in rectal cancer: are there any cri- chemoradiation therapy in locally advanced rectal cancer: diffusion-
teria in addition to the size? Eur J Radiol 2004;52:78–83. weighted 3 Tesla MR imaging. J Magn Reson Imaging 2012;35:110–
86. Kim CK, Kim SH, Chun HK, et al. Preoperative staging of rectal can- 116.
cer: accuracy of 3-Tesla magnetic resonance imaging. Eur Radiol 104. Intven M, Reerink O, Philippens ME. Dynamic contrast enhanced MR
2006;16:972–980. imaging for rectal cancer response assessment after neo-adjuvant
87. Cho EY, Kim SH, Yoon JH, et al. Apparent diffusion coefficient for chemoradiation. J Magn Reson Imaging 2015;41:1646–1653.
discriminating metastatic from non-metastatic lymph nodes in pri- 105. O’Neill BD, Brown G, Heald RJ, Cunningham D, Tait DM. Non-oper-
mary rectal cancer. Eur J Radiol 2013;82:e662–668. ative treatment after neoadjuvant chemoradiotherapy for rectal can-
cer. Lancet Oncol 2007;8:625–633.
88. Heijnen LA, Lambregts DM, Mondal D, et al. Diffusion-weighted MR
imaging in primary rectal cancer staging demonstrates but does not 106. Cascini GL, Avallone A, Delrio P, et al. 18F-FDG PET is an early predictor
characterise lymph nodes. Eur Radiol 2013;23:3354–3360. of pathologic tumor response to preoperative radiochemotherapy in
locally advanced rectal cancer. J Nucl Med 2006;47:1241–1248.
89. Beets-Tan RG, Lambregts DM, Maas M, et al. Magnetic resonance
imaging for the clinical management of rectal cancer patients: rec- 107. Huh JW, Min JJ, Lee JH, Kim HR, Kim YJ. The predictive role of
ommendations from the 2012 European Society of Gastrointestinal sequential FDG-PET/CT in response of locally advanced rectal can-
and Abdominal Radiology (ESGAR) consensus meeting. Eur Radiol cer to neoadjuvant chemoradiation. Am J Clin Oncol 2012;35:340–
2013;23:2522–2531. 344.
90. Veit-Haibach P, Kuehle CA, Beyer T, et al. Diagnostic accuracy of 108. Manfredi S, Lepage C, Hatem C, Coatmeur O, Faivre J, Bouvier AM.
colorectal cancer staging with whole-body PET/CT colonography. Epidemiology and management of liver metastases from colorectal
JAMA 2006;296:2590–2600. cancer. Ann Surg 2006;244:254–259.
91. Beets-Tan RG, Beets GL. MRI for assessing and predicting response 109. Jemal A, Siegel R, Xu J, Ward E. Cancer statistics, 2010. CA: Cancer
to neoadjuvant treatment in rectal cancer. Nat Rev Gastroenterol J Clin 2010;60:277–300.
Hepatol 2014;11:480–488.
110. Belli G, D’Agostino A, Ciciliano F, Fantini C, Russolillo N, Belli A.
92. Roh MS, Colangelo LH, O’Connell MJ, et al. Preoperative multimo- Liver resection for hepatic metastases: 15 years of experience.
dality therapy improves disease-free survival in patients with carci- J Hepatobiliary Pancreat Surg 2002;9:607–613.
noma of the rectum: NSABP R-03. J Clin Oncol 2009;27:5124–5130.
111. Smith MD, McCall JL. Systematic review of tumour number and out-
93. Maas M, Nelemans PJ, Valentini V, et al. Long-term outcome in come after radical treatment of colorectal liver metastases. Br J Surg
patients with a pathological complete response after chemoradiation 2009;96:1101–1113.
34 Volume 45, No. 1

112. Leonard GD, Brenner B, Kemeny NE. Neoadjuvant chemotherapy meta-analysis and systematic review. Eur J Nucl Med Mol Imaging
before liver resection for patients with unresectable liver metastases 2015;42:152–163.
from colorectal carcinoma. J Clin Oncol 2005;23:2038–2048.
124. Villeneuve PJ, Sundaresan RS. Surgical management of colorectal
113. Niekel MC, Bipat S, Stoker J. Diagnostic imaging of colorectal liver lung metastasis. Clin Colon Rectal Surg 2009;22:233–241.
metastases with CT, MR imaging, FDG PET, and/or FDG PET/CT: a
125. Park HS, Jung M, Shin SJ, et al. Benefit of adjuvant chemotherapy
meta-analysis of prospective studies including patients who have not
previously undergone treatment. Radiology 2010;257:674–684. after curative resection of lung metastasis in colorectal cancer. Ann
Surg Oncol 2016;23:928–935.
114. Akhurst T, Kates TJ, Mazumdar M, et al. Recent chemotherapy
reduces the sensitivity of [18F]fluorodeoxyglucose positron emission 126. Treasure T. Pulmonary metastasectomy for colorectal cancer: weak
tomography in the detection of colorectal metastases. J Clin Oncol evidence and no randomised trials. Eur J Cardiothorac Surg 2008;
2005;23:8713–8716. 33:300–302.
115. van Kessel CS, Buckens CF, van den Bosch MA, van Leeuwen MS, 127. Landes U, Robert J, Perneger T, et al. Predicting survival after pul-
van Hillegersberg R, Verkooijen HM. Preoperative imaging of colo- monary metastasectomy for colorectal cancer: previous liver metasta-
rectal liver metastases after neoadjuvant chemotherapy: a meta-anal- ses matter. BMC Surg 2010;10:17.
ysis. Ann Surg Oncol 2012;19:2805–2813. 128. Bergin CJ, Glover GH, Pauly JM. Lung parenchyma: magnetic sus-
116. Reiner CS, Stolzmann P, Husmann L, et al. Protocol requirements ceptibility in MR imaging. Radiology 1991;180:845–848.
and diagnostic value of PET/MR imaging for liver metastasis detec- 129. Suga K, Ogasawara N, Okada M, Hara A, Matsunaga N. Potential of
tion. Eur J Nucl Med Mol Imaging 2014;41:649–658. noncontrast electrocardiogram-gated half-fourier fast-spin-echo mag-
117. Chung WS, Kim MJ, Chung YE, et al. Comparison of gadoxetic acid- netic resonance imaging to monitor dynamically altered perfusion in
enhanced dynamic imaging and diffusion-weighted imaging for the regional lung. Invest Radiol 2002;37:615–625.
preoperative evaluation of colorectal liver metastases. J Magn Reson 130. Raad RA, Friedman KP, Heacock L, Ponzo F, Melsaether A,
Imaging 2011;34:345–353. Chandarana H. Outcome of small lung nodules missed on hybrid
118. Lee KH, Lee JM, Park JH, et al. MR imaging in patients with sus- PET/MRI in patients with primary malignancy. J Magn Reson Imaging
pected liver metastases: value of liver-specific contrast agent gadox- 2016;43:504–511.
etic acid. Korean J Radiol 2013;14:894–904. 131. Chandarana H, Heacock L, Rakheja R, et al. Pulmonary nodules in
119. Zech CJ, Korpraphong P, Huppertz A, et al. Randomized multicentre patients with primary malignancy: comparison of hybrid PET/MR and
trial of gadoxetic acid-enhanced MRI versus conventional MRI or CT PET/CT imaging. Radiology 2013;268:874–881.
in the staging of colorectal cancer liver metastases. Br J Surg 2014; 132. Sieren JC, Ohno Y, Koyama H, Sugimura K, McLennan G. Recent
101:613–621.
technological and application developments in computed tomogra-
120. de Geus-Oei LF, Vriens D, van Laarhoven HW, van der Graaf WT, phy and magnetic resonance imaging for improved pulmonary nod-
Oyen WJ. Monitoring and predicting response to therapy with 18F- ule detection and lung cancer staging. J Magn Reson Imaging 2010;
FDG PET in colorectal cancer: a systematic review. J Nucl Med 32:1353–1369.
2009;50(Suppl 1):43S–54S.
133. Stolzmann P, Veit-Haibach P, Chuck N, et al. Detection rate, loca-
121. Choi M, Kollepara SL, Heilbrun LK, Smith D, Shields AF, Philip PA. tion, and size of pulmonary nodules in trimodality PET/CT-MR: com-
PET scans as a predictive marker of survival in advanced colorectal parison of low-dose CT and Dixon-based MR imaging. Invest Radiol
cancer. Clin Colorectal Cancer 2015;14:35–40. 2013;48:241–246.
122. Heriot AG, Hicks RJ, Drummond EG, et al. Does positron emission 134. Burris NS, Johnson KM, Larson PE, et al. Detection of small pulmo-
tomography change management in primary rectal cancer? A pro- nary nodules with ultrashort echo time sequences in oncology
spective assessment. Dis Colon Rectum 2004;47:451–458. patients by using a PET/MR system. Radiology 2016;278:239–246.
123. Maffione AM, Lopci E, Bluemel C, Giammarile F, Herrmann K, 135. Kang B, Lee JM, Song YS, et al. Added value of integrated whole-
Rubello D. Diagnostic accuracy and impact on management body PET/MRI for evaluation of colorectal cancer: comparison with
of (18)F-FDG PET and PET/CT in colorectal liver metastasis: a contrast-enhanced MDCT. AJR Am J Roentgenol 2016;206:W10–20.
January 2017 35

COMMENTARY
Guidelines for Documentation and

Consent for Nonclinical, Nonresearch
MRI in Human Subjects
Scott B. Reeder, MD, PhD,1* Vera Kimbrell, BSRT,2 Titti Owman, RT, RN,3
Michael Steckner, PhD, MBA,4 and Fernando Calamante, PhD,5
on behalf of the ISMRM Safety Committee†
Magnetic resonance imaging (MRI) of human subjects is widely performed for clinical and research purposes. Clinical
MRI requires a physician order, while research MRI typically requires an approved protocol from a local Institutional
Review Board, as well as informed consent. However, there are several circumstances in which it is appropriate to
perform MRI in human subjects, that constitute neither clinical nor research activities. Examples include clinical protocol
development, training and teaching, and quality assurance testing. We refer to such activities as nonclinical, nonresearch
MRI. The purpose of this document is to provide principles and guidelines for appropriate and safe use of MRI in
human subjects for nonclinical, nonresearch purposes.
M agnetic resonance imaging (MRI) is a noninvasive

cross-sectional imaging technique that has revolution-
ized the diagnosis of disease and is commonly used around
tion varies widely, the United States Federal Government
defines research as “a systematic investigation, designed to
develop or contribute to generalizable knowledge.”1 Other
the world in clinical practice. Furthermore, there are tre- governments and regions of the world have related but dif-
mendous continued research activities developing new MRI ferent definitions.1–4 It will be incumbent upon the reader
techniques as well as using established MRI methods as part of these guidelines to familiarize himself or herself with the
of important research studies including human subjects. definition of research that is relevant to their country, and
If an MRI study is performed for clinical purposes, a phy- consider this concept in the proposed guidelines.
sician order is required and informed consent from the patient There are many circumstances where MRI in human
is usually not necessary given the high safety protocol and low subjects (hereafter “volunteers”) falls into a gray zone that
risk of MRI, relative to many medical procedures. If the activity constitutes neither research nor clinical imaging. This leads
constitutes research, standard ethical principles typically require to tremendous uncertainty as to whether these activities
the use of locally approved Institutional Review Board (IRB) or should be performed as part of an IRB approved protocol,
Ethics Board (both hereafter referred to as IRB) protocols for or whether any form of informed consent is even necessary.
research studies. In particular, most prospective research proto- We refer to such an activity as “nonclinical, nonresearch”
cols also involve some form of informed consent. MRI.
The appropriate use of an IRB protocol depends Currently, the use of volunteers in MRI scanners
highly on the definition of research. Although this defini- for nonclinical, nonresearch purposes is highly variable. At
*Address reprint requests to: S.B.R., Department of Radiology, University of Wisconsin, 600 Highland Avenue, CSC E1/374, Madison, WI 53792-3252.
E-mail: sreeder@wisc.edu
From the 1Departments of Radiology, Medical Physics, Biomedical Engineering, Medicine, and Emergency Medicine, University of Wisconsin, Madison,
Wisconsin, USA; 2Brigham and Women’s Hospital, Boston, Massachusetts, USA; 3Department of Medical Imaging and Physiology, Lund University Hospital,
Lund, Sweden; 4Toshiba Medical Research Institute, Mayfield Village, Ohio, USA; and 5Florey Institute of Neuroscience and Mental Health, University of
Melbourne, Melbourne, Victoria, Australia
Additional supporting information may be found in the online version of this article
36 C 2016 International Society for Magnetic Resonance in Medicine

V
ID: jw3b2server Time: 05:43 I Path: //chenas03/Cenpro/ApplicationFiles/Journals/Wiley/JMRI/Vol04501/160157/Comp/APPFile/JW-JMRI160157

Reeder et al.: Guidelines for Documentation and Consent
many institutions, no formal consent is used, while at other Quality Assurance / Quality Improvement
institutions, IRB research protocols are being used. Testing of existing, repaired or new equipment occasionally
Although MRI is an exceptionally safe imaging modality, it requires volunteers to evaluate system performance for qual-
is not free from all risk. For this reason, guidelines for the ity assurance and quality improvement purposes. Evaluation
appropriate use of nonclinical, nonresearch MRI scanning may include (but is not limited to) testing of signal-to-noise
and for appropriate use of informed consent are needed. ratio performance, evaluation of image artifacts, and any
A structured process is needed to ensure safe and appropri- metric related to the performance of the MRI system.
ate scanning of volunteers, and also to provide the necessary In addition to the above applications, there may be other
information about the potential risks of MRI to volunteers appropriate applications for the use of nonclinical, nonre-
undergoing nonclinical, nonresearch MRI. search MRI in volunteers. These may vary by site and by
The purpose of this document is to provide suggested region, and depend on the specific needs of a particular insti-
principles and guidelines for appropriate MRI of volunteers tution. It is important to note, however, that establishing and
for nonclinical, nonresearch purposes. The essence of these implementing a research MRI protocol using volunteers,
guidelines closely follows the general principles of research before embarking on a research protocol with patients, does
and safety pertaining to the appropriate screening of volun- not constitute a nonclinical, nonresearch MRI activity. A
teers, informed written consent, and all standard safety separate IRB protocol may be required, if the overall activity is
precautions. It is essential to note that the definitions of related to research.
research, authority of local IRBs, and the laws and regula-
tions of different countries and regions of the world are Guidelines for Nonclinical, Nonresearch MRI
highly variable. The guidelines described below may not be The following guidelines aim to describe the basic principles
applicable in all regions. The intention of this document is and suggested requirements for implementing a careful and
not to dictate how such an approach should be performed orderly process for institutions that wish to implement proc-
but rather provide helpful guidelines, should the need for a esses for nonclinical, nonresearch MRI scanning activities.
structured process arise at a specific institution. Any imple- The following guidelines attempt to provide helpful sugges-
mentation of the guidelines described in this document may tions to ensure safe MRI practices and to protect volunteers
benefit from consultation with the local IRB and must be by following accepted principles of informed consent.
consistent with local guidelines, laws, and customary prac- 1. The use of volunteers for the purposes of clinical protocol
tices. Please note that the following guidelines have been development, training and teaching, and quality assur-
written on behalf of and approved by the International Soci- ance testing should be minimized wherever possible. The
ety for Magnetic Resonance in Medicine (ISMRM) Safety use of phantoms is advised whenever possible. However,
Committee. A list of Safety Committee members at the there are currently no phantoms that adequately mimic
time of drafting this document is found in the on-line all the physiological processes that affect an MRI exami-
Supporting Information. nation, including human anatomy, physiological activity,
body composition, and MRI relaxation parameters, and
Definition of Nonclinical, Nonresearch MRI
other important features that may impact an MRI acqui-
There are many potential uses of MRI for nonresearch, non- sition, such as ECG signals and respiratory gating. It is
clinical purposes. While this varies depending on regional recognized, therefore, that in many circumstances that
regulation, the most common applications of nonclinical, MRI in volunteers is ultimately necessary for testing clin-
nonresearch MRI include the following: ical protocols, training, and teaching, and for some qual-
ity assurance applications.
Clinical Protocol Development
When developing new clinical MRI protocols or refining 2. It is recommended that sites contact their local IRB, before
existing clinical MRI protocols, it is often necessary to image implementing a nonclinical, nonresearch MRI activity, to
volunteers to ensure adequate image quality has been achieved, ensure that the proposed MRI activities in human subjects
before using these protocols in patients for clinical purposes. do not constitute human subjects research requiring IRB
These protocols are generally comprised of sequences approval. Nonclinical, nonresearch MRI should never be
approved by local regulatory bodies for use as defined by the used to circumvent the appropriate use of IRB approved
supervising physician. research protocols.
In general, MRI involving nonapproved software and/
Training and Teaching or hardware (“devices”) requires regulatory and/or local
It is sometimes necessary to image volunteers for the IRB oversight, although this may vary from region to
purposes of teaching students and for training operators on region. Before using nonapproved devices as part of
the safe and effective use of MRI systems. nonclinical, nonresearch MRI, it is recommended that the
January 2017 37
ID: jw3b2server Time: 05:43 I Path: //chenas03/cenpro/Support/3B2/Wiley/3B2Server/PS_Generation/Issue_3G/3d_Process/JW-JMRI160157

site consult with appropriate regulations and the local IRB (NSF). However, it is recognized that there may be rare
regarding the use of nonapproved devices. For example, in and unforeseen circumstances where the use of a contrast
the United States, the Food and Drug Administration agent is needed for an appropriate nonclinical, nonresearch
Code of Federal Regulations (CFR) 812 may provide the MRI activity that is not considered research and, therefore,
necessary guidance.5 is inappropriate to use under an IRB approved protocol.
3. In general, all nonclinical, nonresearch MRI should be This determination should be made in conjunction with
performed in healthy adults who can participate in the the local IRB, if possible.
informed consent process without additional consent Should contrast agents be used for nonclinical, nonre-
required from a parent or guardian. Nonclinical, search MRI, all standard risks of contrast agents, including
nonresearch MRI should not, in any way, replace a allergic reactions, NSF, and other known risks, should be
medically necessary imaging study. included in the consent form, similar to that listed in an
4. Archiving of images, as well as anonymization and/or IRB approved consent form for the use of contrast
deidentification of images should follow local institu- enhanced MRI for research purposes. All standard institu-
tional practices similar to those used for research studies. tional procedures for the safe use of contrast, including the
For some applications, such as teaching and training, availability of appropriate equipment, drugs and trained
there may be no need to archive images. individual(s) needed to treat contrast reactions should be
5. Individuals from vulnerable populations,1 which includes followed. Finally, contrast agents are drugs, and like any
children, wards of the state, prisoners, pregnant patients, drug, the administration of a contrast agent always requires
persons who are mentally disabled or cognitively prescription and oversight from a licensed physician.
impaired, or are economically or educationally disadvan- Recently, there have been several reports describing
taged should be excluded from participation. Further- increased signal with T1 weighted imaging in the deep
more, individuals who are impaired and unable to nuclei of the brain in patients receiving multiple doses of
provide informed consent, including those individuals gadolinium based contrast agents.6–9 This imaging obser-
unable to speak the local language adequately, or have vation has raised concern for the possibility of toxicity
communication impairments should also not participate related to cumulative gadolinium deposition. However, no
in nonclinical, nonresearch MRI activities. paired data demonstrating any adverse biological or clini-
6. All nonclinical, nonresearch MRI should be performed cal outcomes have been reported, despite hundreds of mil-
under the direct supervision of an experienced and lions of administered doses. As there are no known risks
qualified (in accordance with local guidelines) MRI related to gadolinium deposition, no specific recommen-
operator. In the case of teaching, it may be appropriate dations regarding informed consent and gadolinium depo-
for an inexperienced MRI operator to perform the sition can be made by the ISMRM at this time. Should
examination, but only under direct supervision of an any adverse biological or clinical effects related to gadolin-
experienced and qualified operator. ium deposition be discovered subsequent to this publica-
7. A major purpose of developing guidelines for nonclini- tion, it may be appropriate to include these risks as part of
cal, nonresearch MRI is to ensure that all volunteers the consent process.
undergo appropriate MRI safety screening. MRI safety 10. Use of nonclinical, nonresearch scanning should never
screening should be performed at a level equivalent to be performed for any medical purpose. It should be
that used for research and/or clinical purposes. As there made clear to volunteers that this MRI activity should
is no benefit to the volunteer, any concern whatsoever never be used to replace clinical imaging. Volunteers
about scanning a volunteer with a contraindicated who are symptomatic or have concerns of underlying
implant or other questionable contraindication (e.g., disease should consult their physician.
claustrophobia) should exclude the volunteer from par- 11. Careful consideration should be made to whether to scan
ticipating. Furthermore, if it were necessary to perform employees or trainees within your institution or volunteers
involved procedures such as radiographs to exclude from outside the institution. It may be preferable to recruit
orbital metallic fragments, for example, the volunteer volunteers who have a working knowledge of MRI safety
should not undergo nonclinical, nonresearch MRI. and the local MRI environment to facilitate the safest pos-
8. Documentation of appropriate MRI safety screening (e.g., sible scanning activity. There may also be important impli-
signed safety screening forms) is considered essential. cations related to workers’ compensation and, if needed,
9. The use of contrast agents for the purpose of nonclinical, appropriate wavers of liability may be necessary within the
nonresearch MRI is highly discouraged. Although gadolin- consent form. Consultation with the appropriate depart-
ium based contrast agents are considered safe, they are not mental human resources team is encouraged, to ensure
free from adverse events including allergic reactions and in that nonclinical, nonresearch policies are consistent with
patients with renal failure, nephrogenic systemic fibrosis institutional policies. If employees or trainees are recruited,
38 Volume 45, No. 1

a supervisor with an authority relationship over volunteers clinical, nonresearch MRI activities should have undergone
should never be involved in the recruitment of volunteers. standard research training and certification. For example,
It is recommended that a standardized approach for standard CITI (Collaborative Institutional Training Initia-
approval of employee participation in nonclinical, nonre- tive) and HIPAA (Health Insurance Portability and
search MRI be instituted. Accountability Act) training is recommended for those indi-
Should volunteers who are not employees of the institu- viduals performing nonclinical, nonresearch MRI in the
tion be recruited, standard recruiting mechanisms (e.g., United States
billboards, emails, etc.) such as those used by the IRB may All of the basic elements of informed consent, similar
be appropriate based on local institutional practices. Any to that required in research protocol should be captured in
process that introduces an element of coercion, such as a the consent form. The specific elements of informed consent
status relationship or excessive compensation should be may vary from region to region. One suggestion is to begin
avoided. In general, it is recommended that minimal with an existing IRB approved consent form and modify it
payment, if any, to compensate subjects for their time and appropriately to suit the scanning activities for nonclinical,
expenses (e.g., travel, parking) should be used. nonresearch purposes. A new MRI safety screening form
12. Oversight of nonclinical, nonresearch MRI activities by and informed consent should be performed every time a
an appropriate institutional oversight committee, such as volunteer enters the MRI environment for nonclinical,
an MRI Safety Committee or a dedicated “NonClinical, nonresearch MRI. In circumstances where the volunteer is
NonResearch MRI Committee” is highly recommended. an employee and an advanced MRI operator (e.g., MRI
Similar to an IRB committee, this committee should technologist / radiographer) with an existing safety screening
have experts with experience and training in the princi- form on file, other safety screening procedures that ensure
ples of human subjects research. Guidance and oversight rigorous MRI safety screening may be adopted. Note that
by a professional with expertise in MRI safety, perhaps knowledgeable operators may have pre-existing conditions
as a member of the oversight committee, is also recom- that are acceptable as an MRI operator, but not as a volun-
mended. The oversight committee should appoint one teer (e.g., MR conditional device, or medication that inter-
or more qualified individuals who are authorized to feres with the thermoregulatory system).
approve nonclinical, nonresearch imaging activities. When seeking informed consent for research, there are
Finally, in challenging situations, escalation to broader several key elements that should be included in the consent
institutional oversight such as the institutional Safety process. Based on widely accepted principles of informed
Committee or Risk Management may be necessary. It is consent,10 we advise that the following basic elements of
incumbent upon the oversight committee to identify informed consent used for research should also apply to vol-
and understand the appropriate institutional pathway(s) unteers for nonclinical, nonresearch MRI:
should escalation be necessary, before implementing local
1. A statement that explains the purposes of the MRI activ-
nonclinical, nonresearch guidelines.
ity, the expected duration of the imaging session, and a
The above list of suggestions provides the basic framework for description of the procedures that will be followed.
performing nonclinical, nonresearch MRI scanning in human 2. A description of any foreseeable risks or discomfort to
subjects. It is recognized that institutional guidelines and the subject should also be included. This includes physi-
regional differences in laws and other regulations supersede all cal or psychological effects, which for MRI may include:
of the above guidelines. If appropriate for your local institu- (a) claustrophobia, (b) skin burns, (c) risk of projectiles,
tion, the above guidelines may provide helpful suggestions for (d) nerve stimulation, (e) heating, (f ) acoustic noise, (g)
providing an organized process and framework on the appro- risk of disclosing personal medical information to those
priate nonclinical, nonresearch use of MRI in human subjects. who may see the images, (h) risk of discovery of a find-
In this way, the safe and appropriately documented use of this ing of uncertain medical significance, (i) risks of contrast
MRI activity can be performed. agents, if used.
3. An explicit explanation that there is no direct benefit to
Informed Consent the volunteer, although there may be benefit to future
Informed consent for nonclinical, nonresearch MRI should patients by establishing and improving clinical protocols,
be performed to the same standards as performed for teaching of MRI operators and through quality assurance
research studies. Key elements of informed consent, as listed and quality improvement of the medical practice.
below, should be included in the consent process. It is also 4. A statement regarding whether or not an expert interpret-
highly recommended that signed informed consent, to ing physician will review images for incidental findings,
document the consent process, is performed. Furthermore, should be included (please see next section for further
all individuals involved in the consent procedure for non- details regarding incidental findings).
January 2017 39

5. If images will undergo a formal review, the potential In general, we recommend that no formal image review
consequence of identifying findings of unknown clinical be performed for nonresearch, nonclinical MRI, so long as this
significance and potential psychological and health con- is disclosed in the consent form. This approach may help avoid
sequences, as well as potential consequences for insur- potential abuse by volunteers seeking MRI as a surrogate for a
ability for healthcare should also be included as a clinically necessary MRI exam. It should be noted, however,
potential risk. that at some institutions, IRB protocols require review of cer-
6. A statement describing the extent, if any, to which confi- tain research MRI studies by board-certified radiologist or other
dentiality of images and any identifying information will clinician. This practice varies between specific IRB protocols,
be maintained. institutions, and regions. The local IRB may be helpful to pro-
7. A statement regarding compensation, if any, and/or vide guidance on this issue, although it should be noted that
medical treatments that are available should injury occur oversight of the IRB applies only to human subjects research
as part of this activity. All statements must follow local activities, and does not apply to nonclinical, nonresearch MRI
laws and regulations. It should be noted that the sample activities.
consent forms provided in the on-line supporting infor- As discussed above, any guidelines that choose to
mation are from the United States. review images for incidental findings should include a state-
8. Information on an institutional contact for any questions ment in the consent form describing the process for inform-
that the volunteer may have related to the scanning ing the subject and/or his/her physician.
activity. The specific choice of institutional contact listed In conclusion, the principles and guidelines described in
on the consent form should be based on the same this document are aimed at providing a useful framework and
approach taken for local IRB protocols. process for appropriate and safe nonclinical, nonresearch MRI of
9. A statement that participation in this activity is volun- human volunteers. Without a physician order and without an
tary and that refusal to participate will involve no pen- approved IRB protocol, many institutions do not have well
alty or other loss of benefits to which the subject would developed processes including elements of informed consent
be otherwise entitled. Volunteers are permitted to with- that ensure that standard MRI safety practices, and the principles
draw at any time without penalty. A copy of the consent of beneficence and informed consent are adhered to. We believe
form should be made available to the volunteer. that this document provides a reasonable approach to provide an
orderly process for human scanning in the gray zone between
Incidental Imaging Findings clinical and research MRI activities.
Incidental findings in nonclinical research activities present
an unusual challenge. There is a difficult balance weighing
Acknowledgments
the potential benefits such as a life-saving diagnosis of an
The authors wish to acknowledge support from the NIH
unsuspected abnormality versus the potential harm created
(K24 DK102595).
by the anxiety of a finding of uncertain significance and
potential implications for employment and insurability.
There are no clear consensus approaches for managing inci-
dental imaging findings in the context of research related References
imaging, and it is recognized that there remains a general 1. Department of Health and Human Services, Code of Federal Regula-
“. . . lack of evidence on which to base practice . . . on the tions, Part 46: protection of human subjects. Available at: http://www.
balance of harm versus benefit in telling research partici- hhs.gov/ohrp/regulations-and-policy/regulations/45-cfr-46/index.html.
Accessed April 3, 2016.
pants about findings . . .”.11 Formalized review of nonclini-
2. Health Canada Research Ethics Board - Ethics Review of Research
cal, nonresearch incidental findings may also place an undue Involving Humans - Administrative Policy and Procedures Manual.
burden on sites to find an available licensed physician who Available at: http://www.hc-sc.gc.ca/sr-sr/pubs/advice-avis/reb-cer/
is willing to perform formal image reviews.11,12 index-eng.php. Accessed April 3, 2016.
Despite these challenges and lack of evidence, guid- 3. Australian Code for the Responsible Conduct of Research. Available
at: http://www.nhmrc.gov.au/_files_nhmrc/file/research/research-integ
ance on the development of specific policies and procedures
rity/r39_australian_code_responsible_conduct_research_150811.pdf.
for managing incidental findings are becoming increasingly Accessed April 9, 2016.
available.11–13 For example, Thorogood et al propose that 4. Frascati Manual. Proposed standard practice for surveys on research
feedback polices for incidental findings should describe pro- and experimental development. Available at: http://www.keepeek.
com/Digital-Asset-Management/oecd/science-and-technology/frascati-
cedures for determining which data (if any) should warrant
manual-2015_9789264239012-en#page1. Accessed April 9, 2016.
review, evaluating incidental findings, re-identifying subjects
5. Food and Drug Administration Code of Federal Regulations Title 21
with findings discovered on anonymized image data, and 2015. Available at: http://www.accessdata.fda.gov/scripts/cdrh/cfdocs/
contacting subjects with incidental findings.13 cfcfr/CFRsearch.cfm?CFRPart=812. Accessed April 3, 2016.
40 Volume 45, No. 1

6. Errante Y, Cirimele V, Mallio CA, Di Lazzaro V, Zobel BB, Quattrocchi 10. Department of Health and Human Services, Code of Federal Regula-
CC. Progressive increase of T1 signal intensity of the dentate nucleus tions, Part 46.116: informed consent checklist - basic and additional
on unenhanced magnetic resonance images is associated with cumu- elements. Available at: http://www.hhs.gov/ohrp/policy/consentckls.
lative doses of intravenously administered gadodiamide in patients html. Accessed date April 3, 16.
with normal renal function, suggesting dechelation. Invest Radiol
11. A report by Representatives of Research Imaging Centers PS,
2014;49:685–690.
Regulatory Bodies, Funding Organisations, Royal Colleges involved
7. Kanda T, Ishii K, Kawaguchi H, Kitajima K, Takenaka D. High signal in research imaging and Patient Organisations, in the UK.
Management of Incidental Findings Detected During Research
intensity in the dentate nucleus and globus pallidus on unenhanced
Imaging 2011. Available at: https://www.rcr.ac.uk/management-inci
T1-weighted MR images: relationship with increasing cumulative dose of a
dental-findings-detected-during-research-imaging. Accessed date
gadolinium-based contrast material. Radiology 2014;270:834–841.
April 3, 16.
8. McDonald RJ, McDonald JS, Kallmes DF, et al. Intracranial gadolinium dep- 12. Wardlaw JM, Davies H, Booth TC, et al. Acting on incidental findings
osition after contrast-enhanced MR imaging. Radiology 2015;275:772–782. in research imaging. BMJ 2015;351:h5190.
9. Radbruch A, Weberling LD, Kieslich PJ, et al. Gadolinium retention in 13. Thorogood A, Joly Y, Knoppers BM, et al. An implementation frame-
the dentate nucleus and globus pallidus is dependent on the class of work for the feedback of individual research results and incidental
contrast agent. Radiology 2015;275:783–791. findings in research. BMC Med Ethics 2014;15:88.
January 2017 41

ORIGINAL RESEARCH
MRI and the Distribution of Bone Marrow

Fat in Hip Osteoarthritis
Jennifer. S. Gregory, PhD,1 Rebecca J. Barr, PhD,1 Victor Varela, PhD,2
Trevor S. Ahearn, PhD,2 Jennifer Lee Gardiner, PhD,3 Fiona J. Gilbert, MD,2
Thomas W. Redpath, PhD,2 James D. Hutchison, PhD,1 and
Richard M. Aspden, DSc1*
Purpose: To characterize the distribution of bone marrow fat in hip osteoarthritis (OA) using magnetic resonance imag-
ing (MRI) and to assess its use as a potential biomarker.
Materials and Methods: In all, 67 subjects (39 female, 28 male) with either total hip replacement (THA) or different
severities of radiographic OA, assessed by Kellgren–Lawrence grading (KLG), underwent 3T MRI of the pelvis using the
IDEAL sequence to separate fat and water signals. Six regions of interest (ROIs) were identified within the proximal
femur. Within each ROI the fractional-fat distribution, represented by pixel intensities, was described by its mean, stand-
ard deviation, skewness, kurtosis, and entropy.
Results: Hips were graded: 12 as severe symptomatic (THA), 33 had KLG0 or 1, 9 were KLG2, 11 with KLG3, and 2 with KLG4
were analyzed together. The fractional-fat content in the whole proximal femur did not vary with severity in males (mean (SD)
91.2 (6.0)%) but reduced with severity in females from 89.1 (6.7)% (KLG0,1), 91.5 (2.9)% (KLG2), 85.8 (16.7)% (KLG3,4) to 77.5
(11.9)% (THA) (analysis of variance [ANOVA] P 5 0.029). These differences were most pronounced in the femoral head, where
mean values fell with OA severity in both sexes from 97.9% (2.5%) (KLG0,1) to 73.0% (25.9%) (THA, P < 0.001) with the largest
difference at the final stage. The standard deviation and the entropy of the distribution both increased (P < 0.001).
Conclusion: Descriptors of the fractional fat distribution varied little with the severity of OA until the most severe stage,
when changes appeared mainly in the femoral head, and have, therefore, limited value as biomarkers.
I N 2013 the World Health Organization classified osteoar-

thritis (OA) as the most common condition affecting the
musculoskeletal system.1 Worldwide, it affects an estimated
Although traditionally thought of as a cartilage disease,
there is increasing recognition that OA is a disorder affect-
ing the whole joint.3 There is a recognized link with obesity,
9.6% of men and 18% of women over 601 and in more eco- especially with knee OA, but also with hip and, curiously,
nomically developed countries OA has been reported in hand OA.4–7 This latter finding indicates that increased
40% of those over 70 years of age.2 The incidence increases weight-bearing is not the primary problem but that meta-
with age and, in an increasingly elderly population, the num- bolic factors may be more important.8 Epidemiological
ber of sufferers continues to increase. Little is known about studies have suggested a systemic etiology independent of
the pathogenesis of primary OA and early detection is diffi- weight-bearing.9,10 It has been proposed that generalized
cult. Treatment is limited to analgesia, exercise, and weight OA may be a systemic disorder affecting the whole muscu-
loss, where appropriate, until joint destruction and pain are loskeletal system driven by lipid metabolism.11 The degener-
severe enough to warrant a surgical joint replacement. ation and loss of cartilage have been the main focus of
Received Dec 2, 2015, Accepted for publication May 5, 2016.
*Address reprint requests to: R.M.A., Arthritis and Musculoskeletal Medicine, School of Medicine, Medical Sciences and Nutrition, IMS Building, University
of Aberdeen, Foresterhill, Aberdeen AB25 2ZD, UK. E-mail: r.aspden@abdn.ac.uk
From the 1Arthritis and Musculoskeletal Medicine, Institute of Medical Sciences, School of Medicine, Medical Sciences and Nutrition, University of
Aberdeen, Foresterhill, Aberdeen, UK; 2Aberdeen Biomedical Imaging Centre, Lillian Sutton Building, School of Medicine, Medical Sciences and Nutrition,
University of Aberdeen, Foresterhill, Aberdeen, UK; and 3Wyeth Research, Collegeville, Pennsylvania, USA
Current affiliations for Victor Varela: 9616 Castle Ridge Circle, Highlands Ranch, CO; Jennifer Lee Gardiner, Bill & Melinda Gates Foundation, Seattle, WA;
Fiona J. Gilbert, Department of Radiology, University of Cambridge School of Clinical Medicine, Cambridge Biomedical Campus, Cambridge, UK.

V

Gregory et al.: Marrow Fat Distribution in Hip OA by MRI
diagnostic and therapeutic studies, despite changes in the radiographic definition based on Kellgren–Lawrence grading
bone figuring highly in the radiographic signs; namely, sub- (KLG). This cross-sectional analysis formed part of a prospective
chondral sclerosis, cyst formation, and marginal osteophyto- study using subjects recruited from the local Radiology Informa-
sis. There is a proliferation of poorly mineralized tion System (RIS). Computerized searches of the database identi-
subchondral bone12 and increased bone containing an fied subjects greater than 30 years old who had undergone an
anteroposterior pelvic radiograph or bilateral radiographs of the
altered profile of growth factors in the iliac crest, remote
hips in the previous 12 months. Radiographic reports were exam-
from the weight-bearing regions of the joints.13
ined by a clinician to assess suitability for the study. Subjects were
Bone-forming osteoblasts share a common mesenchy-
excluded if any of the following were reported: surgical interven-
mal stem cell precursor with adipocytes.14 Defective coregu- tions (including joint prostheses and osteotomies), inflammatory
lation and alterations in lipid metabolism are possible arthropathies, congenital or developmental dysplasias, avascular
mechanisms for the bone pathologies observed in both OA necrosis, metabolic bone disease, or absence of a formal report on
and osteoporosis.11,15 A greater concentration of fatty acids the RIS.
has been reported in cartilage of OA patients16 and an Having identified possible subjects, letters were sent to the
increased fat content in bone marrow from osteoarthritic referring physician to seek their help in recruiting the subject into
femoral heads.17 Not only was the mass of fat per unit vol- the study (no incentive was offered). The referring physician sent
ume of bone tissue doubled in tissue from OA patients,17 an information pack to the subject, who was asked to complete a
despite the reduced marrow space due to the bone prolifera- form and return it to indicate interest in participating in the study.
They were then invited to attend the hospital for MRI. The radio-
tion, but fractional levels of (n 5 6) fatty acids, precursors
graphs of these individuals were graded using the KLG27,28 by a
to proinflammatory eicosanoids, were also increased.17 A
single independent reader with 4 years’ experience who was blinded
report of a pilot study using magnetic resonance imaging
to the clinical diagnosis. The most severely affected hip joint was
(MRI) demonstrated a difference in the lipid fractions in used, except if both hips were graded the same, when the right hip
femoral bone marrow and muscles around the hip joint was chosen in subsequent analyses.
between healthy volunteers and OA patients.18
We currently lack responsive measures, or biomarkers, MRI
that could be used to assess the risk of OA in individuals, Imaging was done using a 3T Philips Achieva MRI scanner (Phi-
detect early disease, monitor progression, or evaluate thera- lips Medical Systems, Best, Netherlands). A protocol was developed
pies. Radiography and MRI offer several biomarkers reflect- based on the Dixon method29 and tested on phantoms comprising
tubes containing different ratios of soy oil and water. The imple-
ing structural changes in cartilage and bone. Radiographic
mentation (Philips Healthcare Clinical Science) was based on the
joint space narrowing is currently the only biomarker
iterative decomposition of water and fat with echo asymmetry and
accepted by the Federal Drug Administration in the USA least-squares estimation (IDEAL) sequence30 and used three
but its lack of sensitivity requires studies to have large sam- gradient-echo images acquired sequentially, with the initial TE
ple sizes and prolonged duration. Measures of joint space equal to 2.1 msec, followed by additional images having TE
width using MRI19 and the application of newer MR meth- increased by an increment (DTE) of 0.76 msec and then by
ods, such as dGEMRIC to assess cartilage composition,20,21 2DTE. This corresponds to the initial TE value having water and
assessment of glycosaminoglycan content using saturation fat in phase, with the early and late TE values having water and fat
transfer,22 and T1rho23 still focus on changes in cartilage, 61208 out of phase. The 3-echo Dixon method allows signals to
but the detection of bone marrow lesions24,25 is broadening be assigned to either water or lipid protons unambiguously follow-
the whole area of MRI biomarkers.26 In this study we used ing suitable data analysis of the three images corresponding to each
MRI to measure the fractional fat content within the bone TE value.
The sequence was used to acquire five slices in the coronal
marrow in the proximal femur in patients with different
plane with a slice thickness of 5 mm, and slice separation of
severities of OA and characterized its distribution using sta-
0.5 mm. The acquired in-plane resolution was 2.5 mm with an
tistical measures of image texture with a view to assessing its
acquisition time of 6 minutes 11 seconds. TR was 160 msec with
potential as an imaging biomarker. an RF pulse angle of 208. These values give a proton density
weighting so that the observed relative fat/water signal ratios are
Materials and Methods not overly affected by differences in relaxation times between the
Subject Recruitment water and lipid protons. Relative fat/water signal values were calcu-
Patients with endstage symptomatic OA were recruited from the lated directly from the decomposed water-fat images using pro-
preoperative assessment clinic while attending the hospital in prep- grams written in-house. The slice in which the femoral head had
aration for a total hip arthroplasty (THA). At this clinic they were its largest diameter was chosen for analysis. Images were also exam-
given an information pack to take home and invited to participate. ined and the presence of bone marrow lesions (BMLs) and cysts
For comparison with this group we required individuals to noted; both were taken to be regions of high signal in the “water”
represent a range of severities of OA. Because there is no recog- images but BMLs were assumed to have diffuse edges, whereas
nized grading of severity for symptomatic OA, we chose to use a cysts presented with demarcated boundaries.
January 2017 43


X
S5ð1=N Þ 2Xi lnðxi Þ
The results were tested for normality (Shapiro–Wilk) and, if

normally distributed, are presented as mean (standard deviation),
otherwise as median [25%, 75%] values. Analysis of variance
(ANOVA) was used to investigate the relationship between OA
severity and the statistical descriptor of the fractional-fat distribu-
tion in each ROI separately and in the total femur, adding all the
femoral areas together. If the normality test failed (P < 0.05) a
Kruskal–Wallis ANOVA on ranks was performed. Tests were cor-
rected for sex.
Results
A total of 67 subjects (39 female and 28 male) with a mean
age of 65.2 years (range 40.5–81.5) took part in this cross-
sectional study; 12 were about to undergo THA and 55
were identified from the RIS and were classified from their
grade of radiographic OA. From the radiographs, 6 hips
C
O were classified as KLG0 and 27 with KLG1, 9 with KLG2,
L 11 with KLG3, and 2 with KLG4. Precision of KL grading
O
R was determined in a previous study using quadratic-weighted
FIGURE 1: Segmentation into regions of interest was done kappa (QWK) with an intraobserver repeatability for this
using statistical shape modeling to define one acetabular and reader of 0.88 and interobserver repeatability of 0.81 against
six femoral regions.
a radiologist with more than 10 years’ experience.37 For anal-
ysis, KLG0 and 1 were combined, as were KLG3 and 4.
The method was calibrated against a phantom comprising a
Table 1 shows the groups and results of tests for differences T1
set of tubes with oil:water volume proportions from 70:30 to
100:0 in steps of 3. These were made up from soy oil and pure
in the distribution of radiographic severity with sex, age, and
water as described by Bernard et al.31 A Bland–Altman analysis of body mass index (BMI). No difference was found between
agreement32 showed the MRI measures to underestimate the true groups with age or BMI but chi-square tests showed they
value with a bias of –14% (95% confidence interval [CI] –15.6%, were not evenly distributed between sexes. Accordingly, subse-
–12.6%). In order to generate a correction factor the MRI value quent tests for significance were corrected for sex.
was regressed on the true value yielding: Qualitatively, the images from severe OA patients
appeared very different from the no-OA group with muscle
MRIfat% 5 0:797oil%13:16 ðR 2 50:99Þ delineation being obscured and a different distribution of
where oil% is the percentage volume of the soy oil and MRIfat% marrow fat apparent in the femur. Representative images of
is that calculated. This equation was used to correct the values for each of the four grades of radiographic severity are shown in
the mean and standard deviation of the percentage fat calculated Fig. 2. Characteristics of the fractional-fat distribution in F2
from the images in each region of interest (ROI) as described the total femur are shown in Fig. 3 and none was very sensi- F3
below. The remaining calculated parameters of the distribution tive to increasing severity of radiographic OA. Small differen-
required no correction. ces were found only in the females for the mean (ANOVA
P 5 0.029), while differences in entropy bordered on tradi-
tional values for significance (P 5 0.055). Closer examination
Statistical Analysis using post-hoc tests showed that the difference arose in
Images of the hip displaying calculated fractional fat values were females undergoing THA in which fractional fat content fell
segmented into one acetabular and six femoral ROIs using statisti-
to 77.5% (11.9%) compared with 88.9% (8.0%) for those
F1 cal shape modeling33–36 (Fig. 1). Image texture in each ROI was
with no-OA to moderate OA. There was, therefore, a signifi-
obtained from the statistical distribution of values of fractional fat
cant difference between the mean fat content of the total
content. The histogram of calculated pixel fat values was character-
ized by calculating the mean, standard deviation, skewness, kurto- proximal femur in males and females in the THA group
sis, and entropy of the distribution. Cysts were automatically (P 5 0.022).
detected (MRIfat% <20%) and removed from each ROI before Considering each region separately, differences in char-
analysis. The entropy of the distribution in each area containing N acteristics of the fractional fat distribution were found in
pixels in which xi is the number of pixels with intensity i was the most proximal regions, especially in the femoral head
defined as: (regions 1 and 2). The region most affected was ROI 1, the
44 Volume 45, No. 1

TABLE 1. Numbers, Age, Sex, and BMI
N Sex N female (%) Age Mean (SD) BMI (kg/m2)
KLG0,1 33 25 (76%) 62.6 (11.3) 27.8 (4.2)

KLG2 9 3 (33%) 65.6 (7.1) 27.4 (4.0)
KLG3,4 13 4 (31%) 67.7 (6.6) 27.2 (3.0)
THA 12 7 (58%) 67.8 (10.5) 28.8 (4.7)
Total 67 39 (58%) 65.2 (10.1) 27.8 (4.0)
P-value 0.015 0.19 0.79
Participants showing no differences in age or BMI with severity of OA (ANOVA) but a significant difference in the sex distributions
between each group (chi-squared) with too many females and too few males in the KLG0,1 group which was reversed in the mild
and moderate groups.
F4 load-bearing area of the femoral head (Fig. 4), in which the 5 (P 5 0.52), ROI 6 (P 5 0.19), or the acetabulum
mean, standard deviation, and entropy all changed with (P 5 0.17).
increasing severity of radiographic OA. No difference was Cysts were identified in three, and BMLs in nine, of
found in the skewness or kurtosis of the fat distributions. the images analyzed. Measurements from the “water” images
There was no difference between males and females and the showed that the fractional water content was highest in cysts
mean value reduced from 97.9% (2.5%) in those with (90.0% [2.0%]) and slightly lower in the BMLs (75%
KLG0,1 to 73.0% (25.9%) in the THA group (P <0.001), [18%]).
with the largest difference being found in this endstage
group. The standard deviation and the entropy of the distri- Discussion
F5 bution increased (both P < 0.001). In region 2 (Fig. 5) the The appearance of images from THA patients was very dif-
differences were similar but not as marked; the mean ferent from that of the other groups. As well as less distinct
decreased from 95.7% (5.4%) to 84.6% (14.2%) (P < muscle delineation, there were apparent differences in mar-
0.001) and, again, the standard deviation (P 5 0.003) and row fat and hence the objective of this study was to quantify
the entropy (P 5 0.007) both increased. The differences these and explore whether characteristics of the statistical
became smaller in the more distal regions and only the aver- distributions could be used as imaging biomarkers. Calcula-
age showed a significant difference with its median value in tions of the fractional fat content in the bone marrow from
ROI 3 falling from 91.8% [84.7%, 95.5%] in the KLG0,1 the MR signals support the results from a previous study18
group to 80.6% [70.8%, 85.0%] in the THA group that there is a significantly smaller fractional-fat content in
(P 5 0.042) and, in ROI 4, from 88.6% [80.4%, 91.9%] in the proximal femur of patients with OA. This difference is
the KLG0,1 group to 68.7% [66.6%, 82.2%] in the THA only apparent in those with endstage disease and occurs
group (P 5 0.021). In all of these, the largest differences mainly in the femoral head, with the differences between
appeared in the THA group. No significant differences in the groups diminishing with increasing distance from the
mean fractional fat content were found with severity in ROI joint. The femoral medulla is the source of hematopoiesis,
FIGURE 2: Representative MR images for each of the four groups classified either by radiographic KLG or by THA.
January 2017 45

C
O
L
O
R
FIGURE 3: Graphs of (a) mean, (b) standard deviation, (c) skewness, (d) kurtosis, and (e) entropy of the fractional fat distribution
in the total proximal femur for THA compared with degrees of radiographic OA.
giving marrow in young individuals its red color. With age obtained are comparable with those previously reported for
there is a known reduction in red marrow with a corre- yellow marrow.
sponding increase in yellow, or fatty, marrow. The rate of The method used three echo times and provided uni-
change is dependent on sex but is usually complete by the form fat saturation by using an iterative least squares
age of 65 for both sexes.38 Red marrow is 40% lipid and method to estimate B0 inhomogeneities. Asymmetrically
yellow marrow is 80% lipid39 and the values we have placed echoes also improve the signal-to-noise ratio from
46 Volume 45, No. 1

C
O
L
O
R
FIGURE 4: Graphs of (a) mean, (b) standard deviation, and (c) entropy of the fractional fat distribution in the most superior region
(ROI 1) of the femoral head for the different groups.
that in the original Dixon method.40 While excellent agree- severities in which all data were acquired using the same
ment has been achieved for fat contents, generally less than methods.
40%, comparing imaging with MR spectroscopy in liver ste- A limitation of the method is that only fractional con-
atosis,41 the high fat content of bone marrow and the pres- tents of water and fat can be measured, not absolute values.
ence of trabecular bone complicates the measurement of BMLs and cysts had high water contents and resulted in a
fractional-fat by shortening the values of T2 for both water reduction in the calculated mean fractional fat% and an
and fat, leading to rapid gradient echo signal decay with increase in the standard deviation, especially in ROI 1,
TE.42 Recent studies have shown that water components where they were most commonly found. The presence of
with short T2 and either very short or long T2 values have BMLs and cysts, however, did not explain all the reduction
negligible contributions to the MR signal at the values for in fractional fat content, as we excluded them as far as pos-
TE accessible in clinical gradient echo sequences.42 Conse- sible and implementation of this measure as a biomarker
quently, MR proton-density fat-fraction images overestimate would be easiest by taking values over the whole ROI. In
the percentage of fat. Modeling T2 corrections indicated doing this, however, a low mean fractional fat% might arise
that the greatest fat-fraction bias, just over 6%, occurred at from BMLs as well as a generally low percentage fat con-
fractional-fat values close to 50% and that this bias tent. A previous histological study indicated that the amount
decreased with increasing fat content so that for fractional- of bone marrow edema represented only a very small frac-
fat values greater than about 85% the bias was less than tion of the head and neck region of examined specimens,44
3%.43 These corrections were not available to us and this but here, where we subdivided the head, they represented a
might explain why our values were unexpectedly high. So, considerable proportion of the ROI and had a marked effect
while the values measured may not be accurate, this should on the values calculated for that region. This same study,
not invalidate their comparison across the range of OA using fractional MR signals calculated from fat and water
January 2017 47

C
O
L
O
R
FIGURE 5: Graphs of (a) mean, (b) standard deviation, and (c) entropy of the fractional fat distribution in the femoral head (ROI 2)
for the different groups.
suppressed proton density images, reported an increase in ness is negative if there are more values than expected to the
the fractional amount of water in OA from 42% water in left of the distribution, is zero for a Gaussian, and positive
the control group to 60% in the OA femora.44 While this if there is a tail to the right. Kurtosis takes the value of
indicates similar trends to our study, the values are hard to three for a normal distribution and a value of greater than
reconcile with our results and with the traditional observa- three indicates a distribution with a sharper peak and fatter
tion of large amounts of fatty, yellow marrow in the proxi- tails. The fractional-fat distributions measured here were
mal femur of elderly individuals. skewed to the left, with sharper peaks and broader tails than
Texture analysis is a common means of characterizing a strictly normal distribution but neither skewness nor kur-
features in images and there are numerous methods available tosis showed any significant differences with severity in any
that describe the distribution of pixel intensities in a digital of the ROIs. The entropy and standard deviation of the dis-
image. Here we used simple measures of the statistical distribution, however, did increase in the femoral head, indi-
tribution and added “entropy”; a statistical measure of ran- cating that the fat distribution became broader and more
domness to describe the changes observed in each ROI. The random in the final stages of the disease.
mean and standard deviation of a Gaussian distribution are These findings of a smaller mean fractional-fat content
common descriptors and need no introduction, other than in severe OA, however, seemingly contradict previous labo-
to note that they are sometimes called the first and second ratory findings17 and observations from surgery in which fat
moments of the distribution. If a distribution is not quite is often expressed as the femoral head is excised. Measuring
Gaussian, skewness is a measure of whether it is lopsided, the total fat content and expressing it as a mass of fat per
the third moment, and kurtosis measures whether it is more unit gross volume of bone tissue biopsy demonstrated a
“peaky” or slightly “squashed,” the fourth moment. Skew- doubling of lipid content in bone cores from patients with
48 Volume 45, No. 1

OA17; the water content was not measured. In this study, we measured, however, occur late in the disease process and
the MR signal only enables the fractional lipid content to would be enhanced by contributions from cysts and BMLs
be measured as a percentage of total signal, with the remain- if added to the lower fat content. They do not appear capa-
der assumed to be due to water. Comparing bone from ble, however, of providing a novel imaging biomarker of
patients with severe OA with that from less severe or no OA disease incidence or sensitive enough to be able to monitor
is further complicated by the increased amount of cancellous progression.
bone in OA,12 which will reduce the volume available for
bone marrow and alter the MR signal. Previous studies,
however, reported that bone in severe OA is hypomineral-
Acknowledgments
ized and the mass fraction of water in the bone increases to
Contract grant sponsor: Translational Medicine Research
about 24%, compared with 17% in normal bone matrix.12
So, along with more bone and smaller spaces between tra- Collaboration, a consortium made up of the Universities of
beculae, there is an increase in water within the bone Aberdeen, Dundee, Edinburgh, and Glasgow, the four asso-
matrix. If, however, this water is closely associated with the ciated NHS Health Boards (Grampian, Tayside, Lothian,
trabecular bone it may be that it has a short T2 and will and Greater Glasgow & Clyde), Scottish Enterprise, and
not contribute to the MR signal as described above. Wyeth; contract grant number: WHMSB-AU119. The fun-
A further limitation arises from the cross-sectional der played no part in the design, execution, analysis, or
nature of this study. While it provides initial data on possi- publication of this article We are grateful to Mrs D. Younie
ble associations between marrow fractional-fat content and for kindly arranging the imaging sessions and Mrs B.
OA severity, a longitudinal study would be needed to show MacLennan (research radiographer) for acquiring the MR
whether more subtle changes could be detected in individu- images. We also thank Dr S. Galea-Soler, Dr G. Waiter, Dr
als with the incidence and progression of OA. Radiographs Zhiqing Wu, and Dr K. Yoshida for Kellgren-Lawrence
were graded by only one reader, not directly involved in this grading, help and advice, and Mr G. Buchan for his exper-
study, which may have led to some ROA cases being mis- tise making the phantoms.
classified, although calibration in previous studies showed
intra- and interobserver repeatability for this reader37 to fall
into the “almost perfect” (QWK 0.81) as defined by
Landis and Koch.45 Study numbers are also relatively small, References
1. Wittenauer R, Smith L, Aden K. Background Paper 6.12. Osteoarthri-
especially in the KLG2 and 4 groups, and when further sub- tis. Geneva, Switzerland: World Health Organization; 2013.
divided by sex led to some results being close to, but greater
2. Valdes AM, Spector TD. Genetic epidemiology of hip and knee osteo-
than, traditional measures denoting significance. Greater arthritis. Nat Rev Rheumatol 2011;7:23–32.
numbers may improve the statistical power but in the search 3. Radin EL, Burr DB, Caterson B, Fyhrie D, Brown TD, Boyd RD.
for a biomarker, while there may be scientific value in find- Mechanical determinants of osteoarthrosis. Semin Arthritis Rheum
1991;21(3 Suppl 2):12–21.
ing a significant association, such results will be of limited
use in individuals or small cohort studies. The calibration 4. Cicuttini FM, Baker JR, Spector TD. The association of obesity with
osteoarthritis of the hand and knee in women: a twin study.
procedure we used has been used elsewhere but it needs to J Rheumatol 1996;23:1221–1226.
be recognized that the types of lipids and their MR signal 5. Oliveria SA, Felson DT, Cirillo PA, Reed JI, Walker AM. Body weight,
are unlikely to match those found in the bone and it simply body mass index, and incident symptomatic osteoarthritis of the
hand, hip, and knee. Epidemiology 1999;10:161–166.
serves to adjust the values closer to what might be expected
in vivo. Further studies are required to compare MR meas- 6. Yusuf E, Nelissen RG, Ioan-Facsinay A, et al. Association between
weight or body mass index and hand osteoarthritis: a systematic
urements of lipid and water with laboratory measurements review. Ann Rheum Dis 2010;69:761–765.
in order to obtain absolute values of total and fractional 7. Carman WJ, Sowers M, Hawthorne VM, Weissfeld LA. Obesity as a
lipid contents. risk factor for osteoarthritis of the hand and wrist: a prospective study.
Am J Epidemiol 1994;139:119–129.
In conclusion, fat-enhanced MR images showed little
variation with increasing severity of radiographic OA; differ- 8. Aspden RM. Obesity punches above its weight in osteoarthritis. Nat
Rev Rheumatol 2011;7:65–68.
ences were only found in patients with endstage disease
9. Hart DJ, Doyle DV, Spector TD. Association between metabolic fac-
awaiting THA. The differences were then confined to the tors and knee osteoarthritis in women: the Chingford Study.
most proximal parts of the femur. Calculation of fractional- J Rheumatol 1995;22:1118–1123.
fat contents from these regions showed them to be signifi- 10. Cooper C, Egger P, Coggon D, et al. Generalized osteoarthritis in
cantly lower, by about 25%, in the THA group and they women: pattern of joint involvement and approaches to definition for
epidemiological studies. J Rheumatol 1996;23:1938–1942.
were accompanied by a change in the texture of the images
11. Aspden RM, Scheven BAA, Hutchison JD. Osteoarthritis is a systemic
within the bone, indicating a broader and more random fat disorder involving stromal cell differentiation and lipid metabolism.
distribution in the final stages of the disease. The differences Lancet 2001;357:1118–1120.
January 2017 49

12. Li B, Aspden RM. Composition and mechanical properties of cancel- 30. Reeder SB, Pineda AR, Wen Z, et al. Iterative decomposition of water
lous bone from the femoral head of patients with osteoporosis or and fat with echo asymmetry and least-squares estimation (IDEAL):
osteoarthritis. J Bone Miner Res 1997;12:641–651. application with fast spin-echo imaging. Magn Reson Med 2005;54:
636–644.
13. Dequeker J, Mokassa L, Aerssens J. Bone density and osteoarthritis.
J Rheumatol 1995;22:98–100. 31. Bernard CP, Liney GP, Manton DJ, Turnbull LW, Langton CM. Com-
parison of fat quantification methods: a phantom study at 3.0T.
14. Grigoriadis AE, Heersche JN, Aubin JE. Differentiation of muscle, fat,
J Magn Reson Imaging 2008;27:192–197.
cartilage, and bone from progenitor cells present in a bone-derived
clonal cell population: effect of dexamethasone. J Cell Biol 1988;106: 32. Bland JM, Altman DG. Statistical methods for assessing agreement
2139–2151. between two methods of clinical measurement. Lancet 1986;1:307–
15. Kruger MC, Horrobin DF. Calcium metabolism, osteoporosis and 310.
essential fatty acids: a review. Prog Lipid Res 1997;36:131–151. 33. Cootes TF, Hill A, Taylor CJ, Haslam J. Use of active shape models
16. Lippiello L, Walsh T, Fienhold M. The association of lipid abnormal- for locating structures in medical images. Image Vis Comput 1994;12:
ities with tissue pathology in human osteoarthritic articular cartilage. 355–365.
Metabolism 1991;40:571–576. 34. Gregory JS, Waarsing JH, Day JS, et al. Early identification of radio-
17. Plumb MS, Aspden RM. High levels of fat and (n-6) fatty acids in can- graphic osteoarthritis of the hip using an active shape model to quan-
cellous bone in osteoarthritis. Lipids Health Dis 2004;3:12. tify changes in bone morphometric features: Can hip shape tell us
anything about the progression of osteoarthritis? Arthritis Rheum
18. Gregory JS, Ahearn TS, Redpath TW, et al. MRI measurements of the 2007;56:3634–3643.
fat content of bone and muscle in hip osteoarthritis. Osteoarthritis
Cartilage 2007;15(Suppl C):C192. 35. Goodyear SR, Barr RJ, McCloskey E, et al. Can we improve the pre-
diction of hip fracture by assessing bone structure using shape and
19. Eckstein F, Collins JE, Nevitt MC, et al. Brief report: cartilage thick- appearance modelling? Bone 2013;53:188–193.
ness change as an imaging biomarker of knee osteoarthritis progres-
sion: data from the Foundation for the National Institutes of Health 36. Barr RJ, Gregory JS, Reid DM, et al. Predicting OA progression to
Osteoarthritis Biomarkers Consortium. Arthritis Rheumatol 2015;67: total hip replacement: can we do better than risk factors alone using
3184–3189. active shape modelling as an imaging biomarker? Rheumatology
(Oxford) 2012;51:562–570.
20. Williams A, Gillis A, McKenzie C, et al. Glycosaminoglycan distribution
in cartilage as determined by delayed gadolinium-enhanced MRI of 37. Yoshida K, Barr RJ, Galea-Soler S, Aspden RM, Reid DM, Gregory JS.
cartilage (dGEMRIC): potential clinical applications. AJR Am J Roent- Reproducibility and Diagnostic Accuracy of Kellgren-Lawrence Grad-
genol 2004;182:167–172. ing for Osteoarthritis Using Radiographs and Dual-Energy X-ray
Absorptiometry Images. J Clin Densitom 2015;18:239–244.
21. Crema MD, Hunter DJ, Burstein D, et al. Association of changes in
delayed gadolinium-enhanced MRI of cartilage (dGEMRIC) with 38. Dunnill MS, Anderson JA, Whitehead R. Quantitative histological
changes in cartilage thickness in the medial tibiofemoral compartment studies on age changes in bone. J Pathol Bacteriol 1967;94:275–291.
of the knee: a 2 year follow-up study using 3.0 T MRI. Ann Rheum Dis
39. Vande Berg BC, Malghem J, Lecouvet FE, Maldague B. Magnetic res-
2014;73:1935–1941.
onance imaging of the normal bone marrow. Skeletal Radiol 1998;27:
22. Ling W, Regatte RR, Navon G, Jerschow A. Assessment of glycosamino- 471–483.
glycan concentration in vivo by chemical exchange-dependent saturation
transfer (gagCEST). Proc Natl Acad Sci U S A 2008;105:2266–2270. 40. Chen CA, Lu W, John CT, Hargreaves BA, Reeder SB, Delp SL, et al.
Multiecho IDEAL gradient-echo water-fat separation for rapid assess-
23. Keenan KE, Besier TF, Pauly JM, et al. T1rho dispersion in articular ment of cartilage volume at 1.5 T: initial experience. Radiology 2009;
cartilage: relationship to material properties and macromolecular con- 252:561–567.
tent. Cartilage 2015;6:113–122.
41. Hines CD, Frydrychowicz A, Hamilton G, Tudorascu DL, Vigen KK, Yu
24. Roemer FW, Frobell R, Hunter DJ, et al. MRI-detected subchondral H, et al. T(1) independent, T(2) (*) corrected chemical shift based fat-
bone marrow signal alterations of the knee joint: terminology, imag- water separation with multi-peak fat spectral modeling is an accurate
ing appearance, relevance and radiological differential diagnosis. and precise measure of hepatic steatosis. J Magn Reson Imaging
Osteoarthritis Cartilage 2009;17:1115–1131. 2011;33:871–881.
25. Ahedi H, Aitken D, Blizzard L, Cicuttini F, Jones G. A population- 42. Karampinos DC, Melkus G, Baum T, Bauer JS, Rummeny EJ, Krug R.
based study of the association between hip bone marrow lesions, Bone marrow fat quantification in the presence of trabecular bone:
high cartilage signal, and hip and knee pain. Clin Rheumatol 2014;33: initial comparison between water-fat imaging and single-voxel MRS.
369–376. Magn Reson Med 2014;71:1158–1165.
26. Eckstein F, Peterfy C. A 20 years of progress and future of quantita- 43. Karampinos DC, Ruschke S, Dieckmeyer M, Eggers H, Kooijman H,
tive magnetic resonance imaging (qMRI) of cartilage and articular tis-
Rummeny EJ, et al. Modeling of T2* decay in vertebral bone marrow
sues—a personal perspective. Semin Arthritis Rheum 2015 [Epub
fat quantification. NMR Biomed 2015;28:1535–1542.
ahead of print].
44. Taljanovic MS, Graham AR, Benjamin JB, et al. Bone marrow edema
27. Kellgren JH, Lawrence JS. Radiological assessment of osteo-arthrosis.
pattern in advanced hip osteoarthritis: quantitative assessment with
Ann Rheum Dis 1957;16:494–502.
magnetic resonance imaging and correlation with clinical examination,
28. The atlas of standard radiographs of arthritis. Rheumatology 2005; radiographic findings, and histopathology. Skeletal Radiol 2008;37:
44(Suppl 4):iv43–iv72. 423–431.
29. Dixon WT. Simple proton spectroscopic imaging. Radiology 1984; 45. Landis JR, Koch GG. The measurement of observer agreement for
153:189–194. categorical data. Biometrics 1977;33:159–174.
50 Volume 45, No. 1

ORIGINAL RESEARCH
Quantifying Metal-Induced Susceptibility

Artifacts of the Instrumented Spine at
1.5T Using Fast-Spin Echo and
3D-Multispectral MRI
S. Sivaram Kaushik, PhD,1* Robin Karr, BS,1 Matthew Runquist, BS,1
Cathy Marszalkowski, BS,1 Abhishiek Sharma, MD,2 Scott D. Rand, MD, PhD,1
Dennis Maiman, MD, PhD,2 and Kevin M. Koch, PhD1
Purpose: To evaluate magnetic resonance imaging (MRI) artifacts near metallic spinal instrumentation using both con-
ventional metal artifact reduction sequences (MARS) and 3D multispectral imaging sequences (3D-MSI).
Materials and Methods: Both MARS and 3D-MSI images were acquired in 10 subjects with titanium spinal hardware on
a 1.5T GE 450W scanner. Clinical computed tomography (CT) images were used to measure the volume of the implant
using seed-based region growing. Using 30–40 landmarks, the MARS and 3D-MSI images were coregistered to the CT
images. Three independent users manually segmented the artifact volume from both MR sequences. For five L-spine
subjects, one user independently segmented the nerve root in both MARS and 3D-MSI images.
Results: For all 10 subjects, the measured artifact volume for the 3D-MSI images closely matched that of the CT implant
volume (absolute error: 4.3 6 2.0 cm3). The MARS artifact volume was 8-fold higher than that of the 3D-MSI images
(30.7 6 20.2, P 5 0.002). The average nerve root volume for the MARS images was 24 6 7.3% lower than the 3D-MSI
images (P 5 0.06).
Conclusion: Compared to 3D-MSI images, the higher-resolution MARS images may help study features farther away
from the implant surface. However, the MARS images retained substantial artifacts in the slice-dimension that result in a
larger artifact volume. These artifacts have the potential to obscure physiologically relevant features, and can be miti-
gated with 3D-MSI sequences. Hence, MR study protocols may benefit with the inclusion both MARS and 3D-MSI
sequences to accurately study pathology near the spine.
A cute back pain is a consequence of a number of clini-

cal conditions,1 for which surgical intervention has
been the preferred course of treatment. The last two deca-
ure to relieve patient symptoms in the spine using surgical
intervention is generally described as failed back surgery
syndrome (FBSS).4
des have seen a respective increase of 90% and 141% in Most fusion surgeries require the installation of metal-
the number of cervical (C) spine and lumbar (L) spine lic hardware to stabilize the spine. Postsurgical radiographic
fusion surgeries.2,3 Despite this prevalence, surgical inter- evaluation is first done with conventional radiography, and
ventions in the spine are prone to complications and have sometimes with computed tomography (CT).5 Metal-
high failure rates (50% for C-spine, 18% for thoracic induced beam scattering and hardening artifacts in CT
spine, and 36% for L-spine2). Unsatisfactory surgical inter- images can been reduced using iterative algorithms,6 high-
ventions can be caused by a host of factors such as energy acquisitions, and/or smoother reconstruction ker-
untreated foraminal stenosis, unrecognized secondary disc nels.7 While CT scans can be extremely useful to monitor
herniation, and traumatization of the nerve root.4 The fail- the placement or mechanical loosening of hardware and
Received Mar 2, 2016, Accepted for publication May 9, 2016.
*Address reprint requests to: S.S.K., Medical College of Wisconsin, 8701 Watertown Plank Rd., Milwaukee, WI 53226. E-mail: sivkaushik@gmail.com
From the 1Department of Radiology, Medical College of Wisconsin, Milwaukee, Wisconsin, USA; and 2Department of Neurosurgery, Medical College of
Wisconsin, Milwaukee, Wisconsin, USA

V 51

trabecular bone continuity, these scans are not well suited Traditionally, 3D-MSI applications were implemented
for assessment of soft-tissue pathology. to acquire sufficient spectral bins (MAVRIC/MAVRIC SL)
Magnetic resonance imaging (MRI) offers a superior or slice encodes (SEMAC) that will overcome the high sus-
soft-tissue contrast, and in the case of hip or knee replace- ceptibility of stainless steel or cobalt-chromium alloyed
ments, advanced susceptibility artifact reduction pulse implants. However, when imaging the spine with titanium
sequences have helped distinguish conditions such as synovi- implants, these conservative 3D-MSI parameters prove to be
tis and osteolysis close to the implant surface.8 However, extremely inefficient. This inefficiency results in inordinately
conventional MR images around metal implants are plagued long scan times, and low signal-to-noise ratio (SNR) in the
by the high susceptibility of certain implant materials, images, which significantly degrades the ability to delineate
which induces extensive image distortions near the implant physiologically relevant details near titanium fusion compo-
interfaces. nents. Recently, a technique was introduced to calibrate the
Recent advances in material science has afforded a spectral coverage needed for every implant material.12 This
transition from high magnetic susceptibility stainless steel affords the ability to tailor 3D-MSI acquisition parameters
implants9 to titanium-alloy implants that have superior osse- for every implant, and has significantly improved image
ous integration and similar strength properties compared to resolution while also reducing the scan time when imaging
those made of stainless steel.10 Given their lower magnetic titanium spinal fusion devices. These improvements help
susceptibility (182 ppm11), titanium implants produce bring the native 3D-MSI image quality closer to that of
reduced MRI susceptibility artifacts and are therefore more conventional MARS images.
amenable to MRI diagnostic assessment. While this may The advent of these newer sequences can greatly
improve the diagnostic capability of MRI around metal, the impact the use of MRI in the study of subjects with spinal
resulting distortions in the images are not negligible. Typical hardware. A number of neurological complications such as
MR images are acquired with in-plane pixel-encoding band- neural foraminal stenosis, arachnoiditis, and epidural fibrosis
widths of about 6200 Hz/pixel and 1 kHz slice-selection are common findings in MR images. Foraminal stenosis is
bandwidths. With titanium-induced Larmor frequency offset responsible for at least 25–29% of FBSS cases17 and epidu-
distributions as high as 3 kHz at field strengths of 1.5T,12 ral fibrosis has been reported as the cause of persistent pain
this can induce in-plane pixel shifts of roughly 8 pixels and in about 30% of FBSS subjects.18 Hence, the excellent soft-
through-plane shifts of roughly 1.5 slice-widths. In addition tissue contrast of MR can play an important part in improv-
to these bulk displacements, the susceptibility of the metal ing the management of subjects with spinal hardware.
subtly shears the structures in the image, which we term Designing an optimum MRI protocol to image the instru-
“lensing.” In order to lower these distortions caused by tita- mented spine can benefit from a quantified analysis of
nium implants, effort has gone into finding “ideal” MR imaging artifacts using MARS and 3D-MSI techniques.
parameters (echo time, receiver bandwidth, slice thick- While other studies have focused on ex vivo studies or have
ness).13 This has led to the use of a high-bandwidth 2D fast made qualitative comparisons between these types of
spin echo sequence with a reduced slice thickness—often sequences,8,19,20 here we used an implant volume measured
known as metal artifact reduction sequences (MARS). from CT implant images as a reference standard. It is
Although MARS has also been demonstrated with the inclu- hypothesized that the artifact volume will be lower in
sion of a view angle tilting (VAT) gradient,14 most conven- images obtained with 3D-MSI sequences, which may clarify
tionally applied MARS sequences operate as just high- the value of these sequences in MR protocols that assess the
bandwidth fast spin-echo sequences. While the MARS instrumented spine.
approach can reduce in-plane distortions, the impact in the
slice dimension is largely overlooked. Materials and Methods
Three-dimensional multispectral imaging (3D-MSI) All MR images were acquired on a GE (General Electric Health-
sequences help overcome the deficiencies of conventional care, Milwaukee, WI) 450W 1.5T scanner. The typical scan proto-
MARS or MARS1VAT sequences. 3D-MSI sequences such col consisted of standard-of-care sagittal and axial T1- and T2-
weighted high-bandwidth 2D-FSE images (MARS), and T1-
as MAVRIC,11 SEMAC,15 and MAVRIC-SL16 operate
weighted MAVRIC-SL 3D-MSI.16 The T1-weighted sequences
under similar underlying MR physics principles and over-
were also acquired after the injection of 10 ml of MAGNEVIST
come the broadened frequency distribution around the
to study the enhancement in the intervertebral discs. For our study,
metal implants by acquiring subimage volumes at distinct the focus remained on the precontrast MARS and 3D-MS images.
frequency offsets from the Larmor frequency. These image Imaging data was obtained in six subjects with lumbar (L)
volumes, or spectral bins, are then combined in a sum-of- implants, and four subjects with cervical (C) spine implants (N 5
squares fashion to produce the artifact-minimized magni- 10, seven men, three women, average age 5 55 6 8 years). All the
tude images. subjects scanned had titanium-based pedicle screws and rods that
52 Volume 45, No. 1

Kaushik et al.: Metal-Induced Artifacts in the Spine
TABLE 1. Implant, artifact, and nerve root volumes.
Age Sex Bins CT MAVRIC MARS

a b c a b c a
Volume SE CV Volume SE CV N.R Volume SEb CVc N.R
L-SPINE
1 65 M 14 21.7 0.9 7.4 24.9 0.8 4.4 4.77 73.8 6.5 13.2 3.09
2 59 M 10 11.6 0.7 10.6 16.8 0.4 3.6 3.5 47.2 7.9 18.9 2.78
3 51 M 24 28.5 2.2 13.4 32.9 4.2 13.1 — 102.2 13.8 13.7 —
4 65 F 10 14.0 0.5 6.4 19.8 0.2 4.5 2.98 38.4 1.6 23.6 2.27
5 59 M 24 16.9 0.5 4.7 24.4 1.6 9.2 2.18 56.5 0.9 2.1 1.82
6 53 F 12 13.7 0.7 9.4 19.2 1.2 6.4 3.15 41.1 0.7 6.7 2.24
a
Absolute error 5.3 6 1.5 42.1 6 18.3
C-SPINE
1 59 M 6 6.8 0.2 5.3 10.5 0.4 18.5 — 16.7 0.4 14.1 —
2 51 M 24 8.2 0.0 0.3 9.6 0.5 6.1 — 25.1 3.0 12.0 —
3 65 F 24 12.9 0.4 5.6 13.8 0.1 4.5 — 26.2 0.3 29.2 —
4 59 M 24 14.7 0.4 4.9 19.7 0.3 2.1 — 28.5 1.4 16.9 —
a
Absolute error 2.5 6 1.8 13.5 6 2.9
a 3
cm .
b
Standard error.
c
Coefficient of variation (%).
NR: nerve root volume (cm3).
were attached at the site of spinal fusion. Written, informed con- between 3:24 minutes and 6:24 min. The scan times for the C-
sent was obtained prior to scanning, and the human research pro- spine subjects were a little shorter, and ranged between 3:31
tocol was approved by the local Institutional Review Board. The minutes and 5:24 minutes for the MAVRIC-SL scans, and 2:54
T1 subject demographics are shown as a part of Table 1. minutes and 3:22 minutes for the MARS scans. To demonstrate
For five of the 3D-MSI images, before the acquisition of the impact of the frequency-encoding directions in the presence of
3D-MSI data, a spectral calibration scan was acquired. This scan metal implants, one MARS dataset was acquired twice, with the
used a low-resolution 3D-MAVRIC pulse sequence to calculate a phase and frequency-encoding swapped, with these imaging param-
map of the frequencies around the implant. This frequency map eters: matrix 5 320 3 224, slice thickness 5 4 mm, TE/TR 5
was then run through an offline algorithm to calculate the ideal 6.2/755 msec, pixel bandwidth 5 325.5 Hz/pixel.
number of spectral bins needed for the 3D-MSI scan.12 With the As a routine procedure after the installation of the fusion
spectral-range calibrated, the spectral-bins were reduced to cover a device, the subjects also underwent a clinically ordered CT scan. The
narrower frequency range. This helped significantly lower the scan CT images were acquired with an FOV of 12–15 cm, a tube voltage
time, and the gain in scan time was used to reduce the extent of of 100–140 kVp, and a voxel size of 0.23 3 0.23–0.35 3 0.35 mm2,
undersampling, which resulted in higher quality images with a bet- and a slice thickness of 1.25–3 mm2. These parameters yielded CT
ter SNR. With fewer spectral bins, the slab thickness was kept images that demonstrated limited beam hardening artifacts near the
moderately large to ensure minimal loss of signal at the edge sli- titanium spinal hardware. Given these limited artifacts, the CT
ces.12 3D-MSI parameters were: matrix 5 256–320 3 224, field images were then used to determine the volume of the implant, and
of view (FOV) 5 20–36 cm, slice thickness 5 4 mm, TE/TR 5 this was used to judge the extent of artifacts seen in the 3D-MSI and
7.2/718 msec, pixel bandwidth (710–977 Hz/pixel), signal averages MARS images as detailed in the sections below.
5 0.5. MARS imaging parameters were: 2D-fast-spin-echo, matrix
5 320–414 3 224, FOV 5 20–28 cm, slice thickness 5 4 mm, Image Registration and Segmentation
TE/TR 5 6.02–6.2/450–576 msec, pixel bandwidth (401–521 All image analysis was implemented using the HOROS software
Hz/pixel), signal-averages 5 2. The MAVRIC-SL voxel sizes package (www.horosproject.org). To aid in the analysis of the data,
ranged between 0.4 3 0.4–0.9 3 0.9 mm2, and the MARS voxel the CT, 3D-MSI, and MARS images were coregistered using a
sizes were smaller, between 0.4 3 0.4–0.6 3 0.6 mm2. The landmark-based affine registration algorithm; 30–40 easily identifia-
MAVRIC-SL scan times for the L-spine subjects were between ble anatomical landmarks were chosen along the sharp edges of the
3:18 minutes and 7:12 min, and the MARS scan times were vertebrae and in the pedicle and were used to first register the CT and
January 2017 53

C
O
L
O
R
FIGURE 1: Representative data showing co-registered CT, MAVRIC-SL 3D-MSI, and MARS data. The in-plane artifact in the MARS
image is larger than the corresponding MAVRIC-SL 3D-MSI image, which agrees well with the CT images. Additionally, the artifact
in the MARS image (shown with arrows) obscures the transverse process of the nerve root.
the 3D-MSI datasets. In the cases where the differences in the CT and Statistical Analysis
MR slice thickness made it difficult to pick landmarks, the CT slice As there were only 10 subjects used for this study, the variability of
thickness was closely matched to the MR images by averaging multi- the implant/artifact volume among the different users was tested
ple slices. Subsequently, a similar number of landmarks were used to using a Wilcoxon signed-rank test. The final implant volume and
register the MARS and the registered 3D-MSI dataset, thus bringing absolute errors were calculated using the mean values of the differ-
F1 all three datasets in the same coordinate space. Figure 1 shows a repre- ent users. The significance of the implant volume, and the nerve
sentative 3D-MSI and MARS dataset registered to the CT image. root volume, were both tested using a Wilcoxon signed-rank test.
The registered datasets were then used to segment the
implant and artifact volumes. For the CT dataset, as there was a Results
dramatic difference between the foreground and the implant signal The implant volume and corresponding artifacts in the MR
intensities, a seed-based region growing was used to segment the images were successfully measured by three independent
implant volume. Additionally, the holes in the implant were filled users and showed only a moderate variability. Using coregis-
in to minimize the error between the implant and artifact volumes tered images enabled improved segmentation, which is high-
in the MR images. The segmented implant was then converted to
lighted in Fig. 1. The segmented volumes for the different
a binary mask, which was used to calculate the volume of the
subjects and different users are given in Table 1.
implant. The artifact region in the 3D-MSI and the MARS images
Given the high contrast between the background and the
were segmented by manually drawing regions of interest (ROIs),
paying close attention to through-plane lensing artifacts in the implant, and the robust seed based region-growing algorithm,
MARS images. Subsequently, these ROIs were used to generate the measured CT implant volumes showed minimal user-
binary masks for both images, which were then used to compute dependent variability. This variability—which was measured
the artifact 1 implant volume. In order to account for variability using the coefficient of variation—was as low as 0.2%, and as
in the manual segmentation procedure, three observers repeated high as 12%, with a mean of 6.8 6 3.6%. Similarly, the corre-
the entire segmentation procedure (S.S.K., R.K., and M.R.). This sponding artifact volume for the 3D-MSI images showed limited
multiple-observer segmentation procedure helped minimize variability among the different users. The variability was as low
observer-induced bias, and the final volume was taken as an aver- as 0.53% and as high as 12.6%, with a mean variability of 7.2 6
age of the numbers reported by all three observers. The differences 5.1% (CT-3D-MSI P 5 0.845). The variability for the MARS
between the average CT implant volume and the average MR arti- images was marginally, yet significantly, higher than both the CT
fact volume was used to compute the “absolute error” for both
and the 3D-MSI images. The variability had a minimum of
MARS and 3D-MSI images. The variability between the different
1.8%, a maximum of 22%, and a mean value of 15.1 6 7.8%
users was quantified using the coefficient of variation. In addition
(MARS-3D-MSI P 5 0.049, MARS-CT P 5 0.02).
to the implant and artifact volume, one user (S.S.K.) segmented
the nerve-root exiting the thecal sac of the spinal column. In slices
Figure 2 shows a 3D rendering from a representative F2
where the through-slice distortion even partially obscured the nerve L-spine subject’s CT implant volume, and the segmented
root, the nerve root was not segmented. The conspicuity of the 3D-MSI and MARS artifact volume. The CT implant vol-
nerve root, and its segmented volume, was used to test the limita- ume for this subject was 21.7 cm3, and the 3D-MSI arti-
tions of both 3D-MSI and MARS in delineating physiologically fact/implant volume was subtly higher, at 24.9 cm3. The
relevant features in the MR images. MARS artifact volume, however, was 3.5-fold higher, at
54 Volume 45, No. 1

C
O
L
O
R
FIGURE 2: Coregistered CT, MAVRIC-SL 3D-MSI, and MARS images. The MARS images show pile-up artifacts that are indicated by
the arrow. The 3D renderings show that the artifacts are greater in the slice dimension of the MARS images.
73.8 cm3. These numbers are largely representative of all the 3D-MSI image was 4.8 cm3, and 35% lower in the
the L-spine subjects, for whom the 3D-MSI absolute error MARS image, at 3.1 cm3. This finding held true for the
was 5.3 6 1.5 cm3, and the MARS absolute error was 9- four other L-spine subjects studied, and the average percent-
fold higher, at 42.1 6 18.3 cm3 (P 5 0.03). age difference between the nerve-root volume of the 3D-
Given the smaller implant volume, the absolute error MSI and MARS images was 24 6 7.3% (P 5 0.06).
for the C-spine subjects was lower. For the four C-spine sub- Additionally, swapping the phase and frequency-
jects studied, the 3D-MSI absolute error was 2.5 6 1.8 cm3, encoding directions merely provides cosmetic changes to the
and the MARS absolute error was 13.5 6 2.9 cm3 (P 5 artifact presentation in MARS acquisitions and does not
0.13). While these numbers clearly show an 5-fold differ- seem to impact the underlying artifact volume. Figure 5 F5
ence, they did not reach significance, as we studied only four shows MARS images in a titanium L-spine subject (not
C-spine subjects. Furthermore, it should be noted that while
the in-plane artifacts can be minimized in the MARS images,
the significant portion of their artifacts are seen in the slice
dimension. These numbers are summarized in the bar chart
F3 shown in Fig. 3. For all 10 subjects, the average absolute
error for the 3D-MSI scans was 4.3 6 2.0 cm3, and 8-fold
higher for the MARS images, at 30.7 6 20.2 cm3 (P 5
0.002). Also, as seen in Table 1, qualitatively, the extent of
the artifacts seen in the 3D-MSI images were not affected by
the number of spectral-bins used for the acquisition.
An example of a partially obscured nerve root can be
F4 seen in Fig. 1, and is indicated with a red arrow. Figure 4 FIGURE 3: Averaged absolute errors in the MARVIC-SL 3D-MSI
shows a 3D rendering of the artifact volume in the 3D-MSI and MARS images for the L- and C-spine subjects. For the L-spine
images, and the MARS images, along with the segmented subjects, there was almost a 9-fold difference between the 3D-
MSI and MARS absolute errors (P 5 0.03). For the C-spine sub-
nerve root volume in yellow. The through-plane artifacts in
jects, the absolute errors showed an 5-fold difference. How-
the MARS image have limited the visibility of this nerve ever, in spite of large difference in errors, as we studied only four
root at a number of locations. The nerve root volume for C-spine subjects, this difference was not significant.
January 2017 55

C
O
L
O
R
FIGURE 4: 3D renderings of the implant/artifact volume (red) and the nerve root (yellow). The slice distortion in the MARS image
obscures the visibility of the nerve root at a number of locations.
included in analysis) where the phase and frequency- Discussion

encoding dimensions were swapped. While we can see In this study we compared the image artifacts obtained in
subtle changes in the artifact presentation, the net artifact conventional MARS and a 3D-MSI sequence (MAVRIC-
volume remains identical (52.3 vs. 51.9 cm3). SL) to reference standard CT images. For all the 10 subjects
C
O
L
O
R
FIGURE 5: Changing the phase- (PE) and frequency-encoding (FE) gradient directions of the MARS imagines did not reduce image
distortions. The artifact presentation shows a marginal change, but the net artifact volume remained the same (3D rendering).
56 Volume 45, No. 1

studied, the MARS images show an artifact volume that is scan utilized here helped determine the number of spectral
roughly 8-fold higher than the corresponding 3D-MSI. bins required for each individual case. This in turn increased
A variety of subtleties uncovered in this analysis the efficiency of the 3D-MSI acquisition and further
require further explanation. A broader treatment of metal allowed images with higher resolution to be acquired in
artifacts in MRI is covered in detailed review articles by clinically viable scan times. Also, with the small number of
Hargreaves et al21 and Koch et al.22 In the context of our subjects in this study, the calibrated images acquired for this
observations in this study, two particular metal-artifact phe- study showed no qualitative change in the artifact volume
nomena are important. First, frequency-encoding gradients when compared to the 24-bin datasets. Hence, these cali-
can cause spatial shifts of signal intensities that may result brated scans may aid in the comprehensive study of pathol-
in an artificial increase in the signal intensity (pile-up), or ogy in the presence of metal implants.
an artificial signal loss.23 Second, in regions close to the This study has a few limitations. First, in spite of the
implant surface the intended slice excitation shapes get large effect size, the small sample sizes limit any broad sta-
warped into one that resembles a “potato-chip.”21 The tistical inferences. In this work we focused on studying in
acquisition of multiple distorted slices can significantly alter vivo image artifacts. Understanding the diagnostic impact of
the shape of the implant, and the surrounding tissues. The MARS and 3D-MSI sequences in spine MR protocols will
hyperintense rings seen in the MARS images are a result of require a larger-cohort clinical study. Second, while the
frequency-encoding distortions, or pile-up. The slice- image registration was used to visually guide artifact quanti-
encoding distortions can be seen more clearly in the 3D fication efforts, resolution and partial volume differences
rendering, where the shape of the implant, or artifact in this
between the MR and CT images could cause the registration
case, has been significantly warped from the original shape
to subtly influence the reported artifact volumes. However,
seen in the rendering of the CT data. 3D-MSI overcomes
it is expected that any registration impact on artifact quanti-
these in-plane and slice dimension distortions and provides
fication is much lower than the interobserver variation
a more accurate delineation of implant boundaries.
introduced by our manual segmentation design. Finally,
In addition to the distortions induced by slice and
while it is out of the scope of this smaller study, a larger
frequency-encoding gradients, there is an additional artifact
cohort study should be performed to include analyses per-
that induces a shearing of signal intensities through a linear
formed by multiple trained radiologists to validate the quan-
combination of these two effects. Here, we refer to this
titative findings shown in this article with a semiquantitative
effect as image “lensing.” While this effect may not be obvi-
clinical scoring system.
ously apparent, it can induce warping of the spatial features
In conclusion, this study quantitatively assessed the
in an image. In the case of spinal imaging, it has the poten-
tial to obscure foraminal pathology in the MARS images. benefits of 3D-MSI in acquiring accurate MR images near
As this effect should be isolated to the MARS images, it is spinal fusion hardware. In spite of the lower magnetic sus-
important to compare both 3D-MSI and MARS images to ceptibility of titanium fusion hardware, conventional MARS
mitigate the impact of lensing in clinical diagnoses. This images show substantial artifacts near such hardware. The
diagnostic effect warrants an independent analysis, and will artifact volumes in the MARS images were 8-fold higher
be probed within a larger clinical cohort study. than the corresponding 3D-MSI images. However, the
One of the primary advantages of 3D-MSI is its MARS images, which have a moderately higher resolution
orientation-independent artifact suppression. The magnetic than 3D-MSI, could be valuable for assessing regions that
field generated by an implant is highly sensitive to its rela- are farther away from the implant surface. A larger clinical
tive orientation to the main magnetic field. While ensuring study is needed to truly ratify the diagnostic differences
the long-axis of the implant is parallel to main magnetic between MARS and 3D-MSI images. However, when used
field may reduce artifacts,24 this is not practical. Addition- in conjunction with MARS sequences, the distinct artifact
ally, as shown earlier, swapping the phase and frequency- reduction advantages of 3D-MSI could potentially help
encoding directions merely provides cosmetic changes to the improve diagnostic assessments of FBSS.
artifact presentation in MARS acquisitions but does not
impact the underlying artifact volume.
Compared to 3D-MSI, the resolution of the MARS
images remains superior (0.7 mm2 vs. 0.9 mm2), and hence
the MARS images hold substantial value in studying regions Acknowledgments
farther away from the implant. The resolution of the 3D- Contract grant sponsor: Advancing a Healthier Wisconsin
MSI images have historically been limited by the large num- Research and Education Fund; contract grant number:
ber of spectral bins that are needed to robustly reduce the 5520357
artifact across a wide variety of implants. The calibration
January 2017 57

References Magn Reson Med 2015 [Epub ahead of print] doi: 10.1002/
mrm.26065.
1. Van Goethem JWM, Parizel PM, Jinkins JR. Review article: MRI of the
postoperative lumbar spine. Neuroradiology 2002;44:723–739. 13. Petersilge CA, Lewin JS, Duerk JL, Yoo JU, Ghaneyem AJ. Optimizing
imaging parameters for MR evaluation of the spine with titanium ped-
2. Rajaee SS, Kanim LEA, Bae HW. National trends in revision spinal icle screws. AJR Am J Roentgenol. 1996;166:1213–1218.
fusion in the USA: Patient characteristics and complications. Bone
Joint J 2014;96 B:807–816. 14. Olsen RV, Munk PL, Lee MJ. Metal artifact reduction sequence: early
clinical applications. Radiographics 2000;20:699–712.
3. Pfunter A, Wier LM, Stocks C. Most frequent procedures performed in
US hospitals. Healthcare Cost Ultilization Project [Internet] 2013:1–9. 15. Lu W, Pauly KB, Gold GE, Pauly JM, Hargreaves BA. SEMAC: slice
encoding for metal artifact correction in MRI. Magn Reson Med 2009;
4. Fritsch EW, Heisel J, Rupp S. The failed back surgery syndrome. Spine 62:66–76.
(Phila 1976) 1996;21:626–633.
16. Koch KM, Brau ACS, Chen W, et al. Imaging near metal with a
5. Sanders WP, Truumees E. Imaging of the postoperative spine. Semin MAVRIC-SEMAC hybrid. Magn Reson Med 2011;65:71–82.
Ultrasound CT MRI [Internet] 2004;25:523–535.
17. Waguespack A, Schofferman J, Slosar P, Reynolds J. Etiology
6. Lemmens C, Faul D, Nuyts J. Suppression of metal artifacts in CT of long-term failures of lumbar spine surgery. Pain Med 2002;3:
using a reconstruction procedure that combines MAP and projection 18–22.
completion. IEEE Trans Med Imaging [Internet] 2009;28:250–260.
18. Chan C, Peng P. Failed back surgery syndrome. Pain Med [Internet]
7. Stradiotti P, Curti A., Castellazzi G, Zerbi A. Metal-related artifacts in 2011;12:577–606.
instrumented spine. Techniques for reducing artifacts in CT and MRI:
State of the art. Eur Spine J 2009;18:102–108. 19. Fritz J, Lurie B, Miller TT, Potter HG. MR imaging of hip arthroplasty.
Radiographics 2014;34:E106–E132.
8. Hayter CL, Koff MF, Shah P, Koch KM, Miller TT, Potter HG. MRI after
arthroplasty: Comparison of MAVRIC and conventional fast spin-echo 20. Koff MF, Shah P, Koch KM, Potter HG. Quantifying image distortion
techniques. Am J Roentgenol 2011;197:405–411. of orthopedic materials in magnetic resonance imaging. J Magn
Reson Imaging 2013;38:610–618.
9. Rupp R, Ebraheim NA, Savolaine ER, Jackson TW. Magnetic reso-
nance imaging evaluation of the spine with metal implants. Spine 21. Hargreaves BA, Worters PW, Pauly KB, Pauly JM, Koch KM, Gold GE.
(Phila 1976) 1993;18:379–385. Metal-induced artifacts in MRI. Am J Roentgenol 2011;197:1–9.
10. Christensen FB, Dalstra M, Sejling F, Overgaard S, B€ unger C. Tita- 22. Koch KM, Hargreaves BA, Pauly KB, Chen W, Gold GE, King KF.
nium-alloy enhances bone-pedicle screw fixation: mechanical and his- Magnetic resonance imaging near metal implants. J Magn Reson
tomorphometrical results of titanium-alloy versus stainless steel. Eur Imaging 2010;32:773–787.
Spine J 2000;9:97–103.
23. Koch KM, King KF, Carl M, Hargreaves BA. Imaging near metal: The
11. Koch KM, Lorbiecki JE, Hinks SR, King KF. A multi spectral three- impact of extreme static local field gradients on frequency encoding
dimensional acquisition technique for imaging near metal implants. processes. Magn Reson Med 2014;71:2024–2034.
Magn Reson Med 2009;61:381–390.
24. Suh J-S, Jeong E-K, Shin K-H, et al. Minimizing artifacts caused by
12. Kaushik SS, Marszalkowski C, Koch KM. External calibration metallic implants at MR imaging: experimental and clinical studies.
of the spectral coverage for three-dimensional multispectral MRI. Am J Roentgenol 1998;171:1207–1213.
58 Volume 45, No. 1

ORIGINAL RESEARCH
MRI of the Knees in Asymptomatic

Adolescent Soccer Players:
A Case–Control Study
Simone B. Matiotti, MD, Msc,1 Ricardo B. Soder, MD, PhD,1*
Rafaela G.Becker, Medical Student,2 Francisco S. Santos, Medical Student,2 and
Matteo Baldisserotto, MD, PhD1
Purpose: To determine the range of asymptomatic abnormal findings in adolescent soccer players at 3.0T MRI of the
knee.
Materials and Methods: In all, 87 knees of asymptomatic 14–17-year-old male adolescents were evaluated at 3T, using
a standardized examination protocol comprising four sequences: two fat-suppressed T2-weighted fast spin-echo sequen-
ces (T2 FSE), in the sagittal (repetition time / echo time [TR/TE], 5.300/71, echo train length [ETL] 17) and coronal planes
(TR/TE, 4234/70, ETL 17), one fat-suppressed proton density (PD) sequence in the axial plane (TR/TE, 2.467/40, ETL 9),
and one T1-weighted spin-echo (T1 SE) sequence in the sagittal plane (TR/TE, 684/12.5). Soccer players (46 knees) were
paired with controls (41 knees) by age and weight. Bone marrow, articular cartilage, meniscus, tendons, ligaments, fat
pad abnormalities, and joint fluid were assessed.
Results: One or more abnormalities were detected in 31 knees (67.4%) in the soccer player group, compared to 20
knees (48.8%) in the control group. The prevalence of bone marrow edema was higher in the soccer group (19 knees,
41.3%) than in the control group (3 knees, 7.3%), P 5 0.001. Other abnormalities found in this sample (joint effusion, car-
tilage lesions, tendinopathy, ganglion cysts, and infrapatellar fat pat edema) were not significantly different between
the two study groups.
Conclusion: Asymptomatic adolescents had a high prevalence of abnormal findings on knee imaging, especially bone
marrow edema. This prevalence was higher among soccer players.
by Soder et al,20 who found that 64.3% of the knees of a

B
ries.
oth professional and amateur high-performance athletes
who play soccer have a high prevalence of knee inju-
1,2
Even among younger athletes, studies have found
sample of under-17 soccer players had injuries and observed
bone marrow edema in more than 50% of the athletes.
that soccer causes more knee injuries than other sports.1,3,4 These authors did not detect any cartilage, meniscus, or ten-
Several types of injuries have been reported, such as osteo- don abnormalities. However, the study conducted by Soder
chondral defects, ligament damage, and tendinopathies.5,6 et al20 used low-field MRI equipment to study athletes’
In response to this, preventative measures have been adopted knees, which is relatively limited for detecting earlier and
among the younger population with the objective of reduc- initial phase injuries. Since their release, high-field, 3.0T
ing the prevalence of these injuries.4,7–10 MRI scanners have exhibited high accuracy and are consid-
Magnetic resonance imaging (MRI) is the standard ered superior to lower-field MRI for detection of osteoartic-
modality for diagnosis and description of knee injuries in ular and musculoskeletal knee injuries, particularly cartilage
soccer players. Attempts to detect injuries at the early stages defects.21–29 It is possible that studying the knees of asymp-
gave rise to studies that investigated populations of asymptomatic soccer players using high-field MRI could detect
tomatic athletes.11–19 One of these studies was conducted early abnormalities that are potentially reversible and which
*Address reprint requests to: R.B.S., Av. Mariland, 1372/501, 90440-190, Porto Alegre, RS, Brazil. E-mail: ricsoder@gmail.com
From the 1Radiology, Pontifıcia Universidade Cat

olica do Rio Grande do Sul (PUCRS), Porto Alegre, RS, Brazil; and 2School of Medicine, PUCRS, Porto
Alegre, RS, Brazil

V 59

are imperceptible using low magnetic field scanners, thereby tocol comprised four sequences; two fat-suppressed T2-weighted
allowing the institution of prophylactic or readaptive meas- fast spin-echo sequences (T2 FSE), in the sagittal (repetition time /
ures or even indicating that the athlete should temporarily echo time [TR/TE], 5300/71, echo train length [ETL] 17) and
refrain from training. coronal planes (TR/TE, 4234/70, ETL 17), one fat-suppressed
In view of this, the objective of this study was to com- proton density (PD) sequence in the axial plane (TR/TE, 2467/40,
ETL 9), and one T1-weighted spin-echo (T1 SE) sequence in the
pare, via 3T MRI, the knees of asymptomatic adolescents
sagittal plane (TR/TE, 684/12,5). The parameters used for image
who take part in high-performance soccer with a control
acquisition were a 352 3 320 matrix, 16 cm field of view (FOV),
group of asymptomatic adolescents who do not play soccer.
3.03 mm slice thickness with 0.3 mm interslice gap, and number of
excitations (NEX) 5 2.
Materials and Methods
The resulting images were analyzed for the presence or
Participants absence of abnormalities. The following findings were noted: joint
All participants and their legal representatives signed consent forms effusion, bone marrow edema, cartilage injuries, and abnormalities
and informed consent forms, respectively. The study was approved of the tendons, meniscus, ligaments, bursas, and fat pads. The fol-
by the Research Ethics Committee at our institution. lowing criteria were used to assess these findings: tendons, rupture,
This cross-sectional, case–control study was conducted from thickening or changes to the signal in T1(SE), fat-suppressed T2-
August 2013 to December 2014 with two groups of adolescents weighted (FSE) and PD sequences; ligaments, rupture, thickening,
aged 14 to 17 years. One of the groups comprised 23 soccer play- or changes to the signal in T1(SE), FSE, and PD sequences; menis-
ers recruited from the feeder categories of a professional soccer cus, increase in signal intensity in T1(SE), FSE, and PD sequences,
team in Brazil. All participants enrolled in the group of athletes irregularity or discontinuity of the outline of the joint surface, with
engaged in intensive soccer training sessions lasting 2 hours and meniscus damage graded as follows: classification: 0 5 normal,
with frequencies varying from 2 to 6 times per week and had been 1 5 intrasubstance signal abnormality not extending to the joint
following this routine for at least 2 years, during which period surface, 2 5 linear intrasubstance signal abnormality not extending
they also took part in regional and national soccer championships. to the joint surface, 3 5 linear intrasubstance signal abnormality
The training routine consisted of circuit training with aerobic and that does extend to the joint surface48; joint cartilage, increase in
anaerobic exercises, with and without the ball, supplemented with signal intensity in T1(SE), FSE, and PD sequences, indefinite or
running, plyometric and isometric exercises, and ball skills training irregular outline or fissures, chondral injuries were classified as fol-
(shooting, heading, dribbling, passing, and ball control). lows: 0 5 normal, 1 5 surface fissures, 2 5 damage compromising
The control group comprised volunteers who did not engage <50% of the thickness of the cartilage, 3 5 damage compromising
in regular sporting activities, with the exception of curricular Physi- >50% of the thickness of the cartilage, but not involving subchon-
cal Education classes at school with no competitive element. dral bone, 4 5 damage compromising the entire thickness as far as
Each participant had both knees examined by MRI, with a the subchondral bone4; bone marrow edema, increased signal
total of 46 knees in the group of athletes and 41 knees in the con- intensity in FSE, and PD sequences and reduced signal intensity in
trol group. T1(SE) weighted sequences; presence of bone marrow edema was
The inclusion criteria for recruitment to the group of athletes quantified as 0 5 absent, 1 5 discrete, 2 5 moderate, 3 5 intense;
were as follows: athletes who were members of the feeder category joint effusion was considered absent when the quantity of synovial
soccer teams, free from knee-related symptoms, aged from 14 to 17 liquid was less than 5 mm in the suprapatellar bursa and present
years, participating in team training for at least 2 years previously in when there was a quantity larger than 5 mm30; edema of the supra-
sessions lasting at least 1 hour, two or more times per week. Exclusion and infrapatellar fat pads, the quadriceps (anterior suprapatellar),
criteria were as follows: history of knee trauma or surgery, other inju- prefemoral (posterior suprapatellar or supratrochlear), and Hoffa
ries that caused breaks from the training routine or a reduction in (infrapatellar) fat pads,31 increased signal intensity in FSE and PD
training frequency, osteoarticular diseases or malformations, history sequences; and reduced signal intensity in T1(FSE) sequences. Gan-
of chronic disease, and body mass index (BMI) >25. glion cysts and liquid in bursas were described in terms of their
The inclusion criteria for the control group were as follows: locations and classified as present or absent.
free from symptoms related to the knees and aged from 14 to 17
years. Exclusion criteria were as follows: high impact sports train- Analysis of the Images
ing with a competitive objective during the previous year, history All of the images acquired from the total of 87 knees were saved
of knee trauma or surgery, osteoarticular diseases or malformations, on a workstation for later analysis. Two radiologists (S.B.M. and
history of chronic disease and BMI >25. R.B.S.) with 8 years’ experience in musculoskeletal imaging
Participants completed a questionnaire on pain, functional assessed all images independently. These observers were blind to
limitations, mechanical symptoms, or any type of knee discomfort which group each subject was from. In cases of disagreement
during the 6 months prior to the MRI examinations. between results, each case was reassessed to arrive at a consensus.
MRI Statistical Analysis

All examinations were conducted using a 3.0T MRI machine The sample size was calculated using PEPI v. 4.0 software (Pro-
(Signa, GE Medical Systems, Waukesha, WI) and an 8-channel grams for Epidemiologists), based on the study conducted by Soder
knee-dedicated HDTR-PA coil. The standardized examination pro- et al.20 Considering a 64.3% prevalence of injuries in the knees of
60 Volume 45, No. 1

Matiotti et al.: 3.0T MRI of Soccer Players’ Knees
TABLE 1. Comparative Table of the Characteristics of

Soccer Players and Controls
Characteristics Players Controls P

(n 5 23) (n 5 21)
Age (years) 15.2 6 0.8 15.7 6 1.1 0.493

Weight (kg) 65.7 6 7.1 67.6 6 10.2 0.483
Height (m) 1.73 6 0.06 1.74 6 0.07 0.680
Body mass 21.6 6 2.1 22.0 6 2.4 0.512
index (BMI)
*Variables expressed as mean 6 standard deviation and com-
pared with Student’s t-test for independent samples.
adolescents who play soccer and a 32.1% prevalence in the knees

of adolescents who do not, an alpha of 5% and power of 80%, it
would be necessary to include 43 knees (22 participants) in each FIGURE 2: Asymptomatic, 15-year-old soccer player. Axial PD-
group in order to detect a difference of 30% between groups. weighted MRI image with fat suppression (TR/TE, 2.467/40;
Data were analyzed using the statistical package SPSS v. 18.0 ETL 9), showing grade 2 patellar chondropathy.
(Chicago, IL) and the Microsoft Excel 2010 (Redmond, WA) elec-
tronic spreadsheet software. Results
Data were input to Excel and later exported to SPSS v. 18.0 for Characteristics of the Participants
statistical analysis. Categorical variables were expressed as frequencies The study and control groups had similar demographic char-
and percentages and compared using the chi-square test or Fisher’s exact
acteristics, which are shown in Table 1. None of the partici- T1
test. Interobserver agreement was evaluated using the kappa coefficient
pants exhibited symptoms or a need for medical examination
of agreement. The McNemar test was used to compare abnormal knee
or supplementary tests or examinations. Median time engaged
findings between groups. Quantitative variables with asymmetrical dis-
tribution were expressed as medians with interquartile ranges (25th and in physical exercise per week was 100 minutes for those in
75th percentiles) and were compared between groups using the Mann– the no soccer group (interquartile range: 71 to 185) and the
Whitney test. The significance level was set at 5%. median for those who engaged in soccer training was 600
minutes (interquartile range: 480 to 700 minutes).
FIGURE 1: Asymptomatic, 16-year-old soccer player. Coronal T2 FIGURE 3: Asymptomatic, 17-year-old soccer player. Sagittal T2
FSE-weighted MRI image with fat suppression (TR/TE, 4234/70; FSE-weighted MRI image with fat suppression (TR/TE, 5.300/
ETL 17) showing bone marrow edema of the medial femoral 71, ETL 17), showing patellar tendinopathy combined with
condyle (long arrow) and the medial tibial plateau (arrow). edema and fragmentation of the infrapatellar pole (arrow).
January 2017 61

TABLE 2. Descriptive and Comparative Table of Abnormalities Detected in the Knees of Soccer Players
and Controls
Players Controls P
n % n %
One or more abnormalities 31 67.4 20 48.8 0.123

Bone marrow edema 19 41.3 3 7.3 <0.001
Joint effusion 9 19.6 8 19.5 0.999
Ganglion cysts/bursae 6 13.0 8 19.5 0.598
Tendinopathy 5 10.9 2 4.9 0.439
Meniscus 5 10.9 — — 0.087
Cartilage 4 8.7 4 9.8 0.999
Hoffa’s fat pad edema 4 8.7 4 9.8 0.999
Abnormalities of the Knees ited abnormalities when studied with MRI. The most fre-
The results for 87 knees were analyzed. One of the patients quent findings were minimal or minor joint effusion
did not complete the examination of the second knee and (19.4%) and fluid in the medial/semimembranosus gastro-
this knee was therefore excluded from the analysis. In the cnemius bursa. Four knees exhibited edema of the infrapa-
group of 23 soccer players, totaling 46 knees, 31 knees tellar fat adjacent to the infrapatellar plica (9.8%). Class 1
(67.4%) exhibited at least one type of abnormality. The cartilage injuries were observed in four knees (9.7%). Just
most prevalent abnormality was bone marrow edema, which two knees exhibited edema involving the iliotibial band.
was detected in 19 knees (41.3%) and the most common Only three knees exhibited bone marrow edema (7.2%) and
location was the medial femoral condyle. Other frequent none of the controls had meniscus abnormalities or patellar
sites of bone marrow edema were the medial tibial plateau tendinopathy.
and the infrapatellar pole. Twelve knees exhibited multiple There was a statistically significant difference in the
foci of bone marrow edema. One of the participants exhib- presence of bone marrow edema between the two groups
ited bone marrow edema compatible with a fracture of the (P 5 0.001), with considerably higher prevalence among the
microtrabecular bone of the medial femoral condyle in both soccer players. Differences between the two groups in terms
F1 knees, suggestive of a stress reaction (Fig. 1). of the other abnormalities studied were not statistically
The second most prevalent abnormality in the group significant.
of soccer players was joint effusion, seen in nine patients There was no significant difference in the numbers of
(19%), classified as absent or present. Abnormalities of the abnormalities when left and right sides were compared, in
patellar cartilage were observed in four knees (8.7%), one of either group. Interobserver agreement (k) was 0.73 for all
F2 which was class 2 (Fig. 2). Four knees exhibited signs of assessments.
patellar tendinopathy, and one of the soccer players also The results are summarized in Table 2. T2
exhibited edema and fragmentation of the anterior tibial
tubercle in both knees and another one exhibited edema
and fragmentation of the infrapatellar pole (jumper’s knee) Discussion
F3 (Fig. 3). Abnormalities of the posterior horn of the medial The objective of this study was to compare the prevalence
meniscus were seen in five knees. Other abnormalities of 3.0T MRI findings in the knees of asymptomatic adoles-
described were fluid in the bursa of the medial/semimem- cent soccer players with findings in the knees of controls.
branosus gastrocnemius, edema of the infrapatellar fat adja- The most important finding of this study was a significantly
cent to the infrapatellar plica, cysts involving the posterior higher prevalence of bone marrow edema in the knees of
cruciate ligament, and edema at the iliotibial band. In this the athletes. Bone marrow edema was detected in 41.3% of
group, 56.5% of participants had abnormalities in both the soccer players compared with only 7% of controls (P <
knees, of whom five (38%) had symmetric, but diverse 0.001), providing support for the belief that this abnormal-
abnormalities. ity is directly related to their sporting activity. These find-
The control group comprised 21 individuals and a ings are in agreement with Soder et al, who also found a
total of 41 knees were analyzed, 20 (48.8%) of which exhib- markedly higher prevalence of bone marrow edema in a
62 Volume 45, No. 1

group of soccer players (50%) compared with controls of patellar tendinopathy in four knees of soccer players,
(3.6%).20 affecting both knees of one player and, in this case, associ-
It was also observed that asymptomatic individuals ated with edema and fragmentation of the anterior tibial
exhibited a high prevalence of abnormalities on MRI tubercule, related to Osgood-Schlatter disease.38 Patellar ten-
images, with a wide range of different types of abnormal- dinopathy was observed in another two athletes, in one of
ities, such as joint effusion, tendinopathies, ganglion cysts, whom it was associated with fragmentation and edema of
liquid in bursas, meniscus abnormalities, chondral injuries, the infrapatellar pole. These findings characterize “jumper’s
edema of the infrapatellar fat, and bone marrow edema, also knee,” an abnormality commonly detected in athletes from
in agreement with findings of previous studies. There was a a range of different sports players.39 The observation of
slightly higher frequency of abnormalities detected by imag- asymptomatic patellar tendinopathy in the soccer players
ing among the group of soccer players than in the control leads us to suspect the possibility that this type of injury
(67.4% vs. 48.8%, respectively), which is similar to findings could come to cause symptoms in the future and that early
described by Soder et al with relation to an earlier study of detection could possibly lead to preventative actions or at
asymptomatic adolescent soccer players employing open- least to closer monitoring of these cases.2,16,40
field 0.35T equipment (64.2% vs. 32.1%, respectively).20 In comparison with the study conducted by Soder
Bone marrow edema is an abnormality that can only et al20 in which other abnormalities such as joint effusion
be detected using MRI and, more specifically, using sequen- or tendon and cartilage abnormalities were not detected, in
ces with fat suppression, and so it is becoming more widely the present study a higher prevalence of these abnormalities
studied as MRI becomes more widespread in orthopedic was observed in the knees of the participants (joint effusion,
clinical practice. When it is linked with sport, it is generally 19.5%, tendon injuries, 8.0%, abnormalities of the patellar
the result of trauma, which may be acute or chronic.32,33 cartilage, 9.2%, and edema at the iliotibial band, 3.4%) and
With increased use of MRI for musculoskeletal assessment although they were not statistically significant in terms of
of athletes, the attention of radiologists and orthopedists has the comparison between groups, they are indicative that
been drawn to the high prevalence of foci of bone marrow 3.0T MRI offers greater sensitivity. The abnormalities of the
edema in asymptomatic athletes and this fact has been the posterior horn of the medial meniscus seen in five knees
motive for several different studies and is a challenge for were related to the peripheral vascular zone commonly seen
interpretation of images, and in terms of choosing the best in children and adolescents.
clinical management.34,35 Using low-field (0.35T) MRI and T1-weighted fat-
The cause of the bone marrow edema that is seen in suppressed (STIR) sequences, Soder et al20 studied 56 knees
asymptomatic athletes has not been elucidated, but there are of adolescent soccer players and controls. Images were
hypotheses based on the suggestion that the biomechanics assessed by two independent radiologists and significant
involved in specific sports may be responsible for these between-group differences were found, particularly in the
abnormalities. Bone is considered a dynamic structure that prevalence of bone marrow edema. Taking Soder et al’s find-
exhibits responses to mechanical stress. This response can be ings as a basis for our study, we used a higher field strength
seen in hypertrophy and trabecular remodeling on radio- (3.0T), but equivalent sequences, in an attempt to improve
graphic examinations. Since trabecular remodeling exhibits detection of injuries in general and, particularly, abnormal-
microfractures and edema when examined histologically, this ities for which low-field MRI has relatively little sensitivity,
abnormality may correspond to the bone marrow edema such as chondral lesions. The present study employed a
seen on MRI images.32,36 This abnormality may be part of larger sample (87 knees) of soccer players with characteris-
a continuous process that would start as a physiological tics approximately similar to those of the Soder et al study.
response to biomechanical stress and end as a stress fracture, The objective of this study was to detect early abnor-
in which case findings in asymptomatic patients would be malities in asymptomatic young soccer players. For this rea-
related to the initial phase of such a continuum.34,37 son, more experienced athletes with greater exposure and
A recent study conducted by Kornaat and Van de training time were excluded from the sample. This limited
Velde examined asymptomatic runners and demonstrated the age group of the participants, who were from 15 to 17
that more than half of the areas of bone marrow edema years old. Despite the extreme importance of detecting inju-
detected on MRI exhibited abnormalities over a 7-month ries in this young age group, ideal for prevention and early
follow-up period, with new lesions appearing and others dis- treatment, these individuals have so far spent few years par-
appearing, while none of these lesions were related to symp- ticipating in competitive sports. This could lead to underes-
toms, suggesting that these findings are most probably timation of some imaging findings, because they had not
related to a physiological bone response to exercise.35 spent enough time training to develop more severe
Another interesting finding of the present study, “overuse” injuries. Another point to consider with relation
although not a statistically significant one, was the presence to populations of athletes is that unrecorded falls and
January 2017 63

impacts could have provoked abnormalities on MRI images. 12. Chang G, Diamond M, Nevsky G, Regatte RR, Weiss DS. Early knee
changes in dancers identified by ultra-high-field 7 T MRI. Scand J
Additionally, many athletes have a high threshold for pain, Med Sci Sports 2014;24:678–682.
which could mask the presence of possible symptoms. The 13. Kaplan LD, Schurhoff MR, Selesnick H, Thorpe M, Uribe JW. Magnetic
significance of these injuries detected in asymptomatic ath- resonance imaging of the knee in asymptomatic professional basket-
ball players. Arthroscopy 2005;21:557–561.
letes could be better established by long-term follow-up in
longitudinal studies that would make it possible to deter- 14. LaPrade RF, Burnett QM 2nd, Veenstra MA, Hodgman CG. The prev-
alence of abnormal magnetic resonance imaging findings in asymp-
mine the relationships between these findings and the emer- tomatic knees. With correlation of magnetic resonance imaging to
gence of symptomatic joint injuries or even chronic joint arthroscopic findings in symptomatic knees. Am J Sports Med 1994;
22:739–745.
problems such as osteoarthritis.
15. Ludman CN, Hough DO, Cooper TG, Gottschalk A. Silent meniscal
In summary, asymptomatic adolescents, both athletes
abnormalities in athletes: magnetic resonance imaging of asymptom-
and nonathletes, had a high prevalence of imaging abnor- atic competitive gymnasts. Br J Sports Med 1999;33:414–416.
malities shown on 3.0T MRI scans, with a higher incidence 16. Major NM, Helms CA. MR imaging of the knee: findings in asymp-
of bone marrow edema among soccer players, which was tomatic collegiate basketball players. AJR Am J Roentgenol 2002;
179:641–644.
statistically significant. Although some abnormalities not
17. Soder RB, Mizerkowski MD, Petkowicz R, Baldisserotto M. MRI of the
described by Soder et al and detected in the present study, knee in asymptomatic adolescent swimmers: a controlled study. Br J
such as cartilage and tendinous lesions, may have been Sports Med 2012;46:268–272.
observed due to the higher sensitivity of high-field MRI, 18. Stahl R, Luke A, Ma CB, et al. Prevalence of pathologic findings
these findings were not statistically significant. Thus, we in asymptomatic knees of marathon runners before and after a
competition in comparison with physically active subjects-a 3.0 T
conclude that these imaging findings are similar to those magnetic resonance imaging study. Skeletal Radiol 2008;37:627–
described with low-field scanners and provide support for 638.
the hypothesis that asymptomatic adolescent athletes may 19. Walczak BE, McCulloch PC, Kang RW, Zelazny A, Tedeschi F, Cole
BJ. Abnormal findings on knee magnetic resonance imaging in
exhibit knee joint abnormalities that can safely be detected
asymptomatic NBA players. J Knee Surg 2008;21:27–33.
using low magnetic field machines. Further studies should
20. Soder RB, Simoes JD, Soder JB, Baldisserotto M. MRI of the knee
be conducted to assess the same population of athletes using joint in asymptomatic adolescent soccer players: a controlled study.
both high- and low-field scanners. AJR Am J Roentgenol 2011;196:W61–65.
21. Aagesen AL, Melek M. Choosing the right diagnostic imaging

References modality in musculoskeletal diagnosis. Prim Care 2013;40:849–
861, viii.
1. Giza E, Micheli LJ. Soccer injuries. Med Sport Sci 2005;49:140–169.
22. Boutsikou K, Kostopoulos S, Glotsos D, et al. Texture analysis of
2. Hagglund M, Zwerver J, Ekstrand J. Epidemiology of patellar tendin- articular cartilage traumatic changes in the knee calculated from
opathy in elite male soccer players. Am J Sports Med 2011;39:1906– morphological 3.0T MR imaging. Eur J Radiol 2013;82:1266–
1911. 1272.
3. Barber Foss KD, Myer GD, Hewett TE. Epidemiology of basketball, 23. Harris JD, Brophy RH, Jia G, et al. Sensitivity of magnetic resonance
soccer, and volleyball injuries in middle-school female athletes. Phys imaging for detection of patellofemoral articular cartilage defects.
Sportsmed 2014;42:146–153. Arthroscopy 2012;28:1728–1737.
4. Junge A, Dvorak J. Soccer injuries: a review on incidence and preven- 24. Krampla W, Roesel M, Svoboda K, Nachbagauer A, Gschwantler
tion. Sports Med 2004;34:929–938. M, Hruby W. MRI of the knee: how do field strength and radiolog-
5. Dvorak J. Osteoarthritis in football: FIFA/F-MARC approach. Br J ist’s experience influence diagnostic accuracy and interobserver
Sports Med 2011;45:673–676. correlation in assessing chondral and meniscal lesions and the
integrity of the anterior cruciate ligament? Eur Radiol 2009;19:
6. Levy AS, Lohnes J, Sculley S, LeCroy M, Garrett W. Chondral delami- 1519–1528.
nation of the knee in soccer players. Am J Sports Med 1996;24:634–
639. 25. Mattila VM, Weckstrom M, Leppanen V, Kiuru M, Pihlajamaki H. Sen-
sitivity of MRI for articular cartilage lesions of the patellae. Scand J
7. Faude O, Rossler R, Junge A. Football injuries in children and adoles- Surg 2012;101:56–61.
cent players: are there clues for prevention? Sports Med 2013;43:
819–837. 26. Sanders TG, Miller MD. A systematic approach to magnetic resonance
imaging interpretation of sports medicine injuries of the knee. Am J
8. Junge A, Rosch D, Peterson L, Graf-Baumann T, Dvorak J. Prevention Sports Med 2005;33:131–148.
of soccer injuries: a prospective intervention study in youth amateur
players. Am J Sports Med 2002;30:652–659. 27. Schoth F, Kraemer N, Niendorf T, Hohl C, Gunther RW, Krombach
GA. Comparison of image quality in magnetic resonance imaging of
9. Wacker F, Konig H, Felsenberg D, Wolf KJ. [MRI of the knee joint of
the knee at 1.5 and 3.0 Tesla using 32-channel receiver coils. Eur
young soccer players. Are there early changes of the internal struc-
Radiol 2008;18:2258–2264.
tures of the knee due to competitive sports?]. Rofo 1994;160:149–
153. 28. Shapiro L, Staroswiecki E, Gold G. Magnetic resonance imaging of
the knee: optimizing 3 Tesla imaging. Semin Roentgenol 2010;45:
10. Walden M, Atroshi I, Magnusson H, Wagner P, Hagglund M. Preven-
238–249.
tion of acute knee injuries in adolescent female football players: clus-
ter randomised controlled trial. BMJ 2012;344:e3042. 29. Wong S, Steinbach L, Zhao J, Stehling C, Ma CB, Link TM. Compara-
tive study of imaging at 3.0 T versus 1.5 T of the knee. Skeletal Radiol
11. Brunner MC, Flower SP, Evancho AM, Allman FL, Apple DF, Fajman
2009;38:761–769.
WA. MRI of the athletic knee. Findings in asymptomatic professional
basketball and collegiate football players. Invest Radiol 1989;24: 30. Fishwick NG, Learmonth DJ, Finlay DB. Knee effusions, radiology and
72–75. acute knee trauma. Br J Radiol 1994;67:934–937.
64 Volume 45, No. 1

31. Saddik D, McNally EG, Richardson M. MRI of Hoffa’s fat pad. Skeletal 36. Schweitzer ME, White LM. Does altered biomechanics cause marrow
Radiol 2004;33:433–444. edema? Radiology 1996;198:851–853.
32. Lazzarini KM, Troiano RN, Smith RC. Can running cause the appear- 37. Navas A, Kassarjian A. Bone marrow changes in stress injuries. Semin
ance of marrow edema on MR images of the foot and ankle? Radiol- Musculoskelet Radiol 2011;15:183–197.
ogy 1997;202:540–542. 38. Hirano A, Fukubayashi T, Ishii T, Ochiai N. Magnetic resonance imag-
ing of Osgood-Schlatter disease: the course of the disease. Skeletal
33. Vanhoenacker FM, Snoeckx A. Bone marrow edema in sports: general
Radiol 2002;31:334–342.
concepts. Eur J Radiol 2007;62:6–15.
39. Roels J, Martens M, Mulier JC, Burssens A. Patellar tendinitis (jumper’s
34. Kornaat PR, de Jonge MC, Maas M. Bone marrow edema-like signal knee). Am J Sports Med 1978;6:362–368.
in the athlete. Eur J Radiol 2008;67:49–53.
40. Cook JL, Khan KM, Kiss ZS, Purdam CR, Griffiths L. Prospective imag-
35. Kornaat PR, Van de Velde SK. Bone marrow edema lesions in the pro- ing study of asymptomatic patellar tendinopathy in elite junior basket-
fessional runner. Am J Sports Med 2014;42:1242–1246. ball players. J Ultrasound Med 2000;19:473–479.
January 2017 65

ORIGINAL RESEARCH
Comparison of Chemical Shift-Encoded

Water–Fat MRI and MR Spectroscopy in
Quantification of Marrow Fat in
Postmenopausal Females
Guanwu Li, MD,1,2* Zheng Xu, MS,3 Hao Gu, BS,3 Xuefeng Li, BS,1 Wei Yuan, MD,4
Shixin Chang, MD,1 Jingzheng Fan, PhD,5 Horea Calimente, PhD,2 and
Jiani Hu, PhD2
Purpose: To validate a chemical shift-encoded (CSE) water–fat imaging for quantifying marrow fat fraction (FF), using
proton magnetic resonance spectroscopy (MRS) as reference.
Materials and Methods: Multiecho T2-corrected MRS and CSE imaging with eight-echo gradient-echo acquisitions at
3T were performed to calculate marrow FF in 83 subjects, including 41 with normal bone mineral density (BMD), 26
with osteopenia, and 16 with osteoporosis (based on DXA). Eight participants were scanned three times with reposition-
ing to assess the repeatability of CSE FF map measurements. Pearson correlation coefficient, Bland–Altman 95% limit of
agreement, and Lin’s concordance correlation coefficient were calculated.
Results: The Pearson correlation coefficient was 0.979 and Lin’s concordance correlation coefficient was 0.962 between
CSE-based FF and MRS-based FF. All data points, calculated using the Bland–Altman method, were within the limits of
agreement. The intra- and interrater agreement for average CSE-based FF was excellent (intrarater, intraclass correla-
tion coefficient [ICC] 5 0.993; interrater, ICC 5 0.976–0.982 for different BMD groups). In the subgroups of varying
BMD, inverse correlations were observed to be very similar between BMD (r 5 –0.560 to –0.710), T-score (r 5 –0.526
to –0.747), and CSE-based FF, and between BMD (r 5 –0.539 to –0.706), T-score (r 5 –0.501 to –0.742), and MRS-
based FF even controlling for age, years since menopause, and body mass index. The repeatability for CSE FF map
measurements expressed as absolute precision error was 1.45%.
Conclusion: CSE imaging is equally accurate in characterizing marrow fat content as MRS. Given its excellent correlation
and concordance with MRS, the CSE sequence could be used as a potential replacement technique for marrow fat
quantification.
Q uantification of marrow fat content is of considerable

interest, especially because of its association with the
pathophysiology of bone loss and cancer therapy-induced
procedure may result in bleeding or infection. Noninvasive
methods for quantification of fatty marrow have been devel-
oped such as micro-computed tomography (CT), dual-
bone marrow damage.1–3 Although marrow biopsy is the energy CT, and magnetic resonance imaging (MRI).1,5–8
reference standard method for quantification of fat content, Among these imaging methods, magnetic resonance spec-
this method is subject to sampling error owing to highly troscopy (MRS) has been found to be safe and accurate and
heterogeneous marrow fat distributions.4 Additionally, this is generally considered as an adequate reference standard
Received Mar 20, 2016, Accepted for publication Jun 6, 2016.
The first two authors contributed equally to this study.
*Address reprint requests to: G.L., Department of Radiology, Yueyang Hospital, Shanghai University of TCM, 110 Ganhe Rd,
Hongkou District, Shanghai 200437, China. E-mail: purenyi@163.com
From the 1Department of Radiology, Yueyang Hospital, Shanghai University of Traditional Chinese Medicine, Shanghai, China; 2Department of
Radiology, Wayne State University, Detroit, Michigan, USA; 3Xin-Zhuang Community Health Center, Shanghai, China; 4Department of Spinal Disease Unit,
Yueyang Hospital, Shanghai University of Traditional Chinese Medicine, Shanghai, China; and 5East Hospital,
Tongji University School of Medicine, Shanghai, China.

V

Li et al.: Water–Fat Imaging Quantifies Marrow Fat
noninvasive technique for fat and metabolite quantification. day during the past month.17 For every participant, age, years since
However, MRS has some disadvantages, such as being menopause (YSM), height, body weight, and body mass index
unable to cover a large area with a single voxel, long acquisi- (BMI) were recorded. The local Ethics Committee Review Board
tion time, and relatively long and complicated approved this study, and written informed consent was obtained
postprocessing. from all participants.
Quantitative chemical shift-encoded (CSE) water–fat
Dual-Energy X-Ray Absorptiometry
imaging is an emerging method for the quantification of fat Areal BMD (g/cm2) of lumbar spine (from L1 to L4) was assessed
content. In order to obtain accurate fat fraction (FF) meas- in all subjects using a Prodigy Lunar scanner (GE Healthcare,
urements, several confounding factors must be considered, Waukesha, WI; v.enCORE.13.40). The scanner was routinely cali-
including T1 bias, noise bias, T2* decay, spectral complexity brated and quality control measures were followed as recom-
of fat, eddy currents, and B0 field inhomogeneity.9–11 Sev- mended by the manufacturer to control possible baseline drift.
eral researchers have presented two- and three-point Dixon Subjects were grouped into three categories according to the T-
MRI to quantify marrow fat content.3,12–14 However, the score: normal bone mass defined as T-scores –1 SD; osteopenia
algorithm was only introduced for a single peak fat as T-score between –1.0 and –2.5 SD; and osteoporosis as T-score
spectrum. –2.5 SD.
Modeling of T2* decay effects is of great importance
MRI Protocol
when quantifying proton density FF either in the marrow in
MRI exams were performed using a 3.0-T unit (Ingenia, Philips
the presence of bone trabecula or in the liver with iron over- Healthcare, Best, the Netherlands) within 2 weeks after dual-
load.1,11 Recent work by multiple groups has shown a good energy x-ray absorptiometry scanning. A surface coil was placed
agreement between CSE MRI and MRS-based marrow FF under the lumbar spine region as the radiofrequency receiver and
measurements when confounders are taken into considera- the body volume coil was used as the radiofrequency transmitter.
tion, eg, single-T2* effects and multipeak fat spectrum The spine imaging protocol included standard clinical sagittal T1-
model.2,8,15 While correcting for a single-T2* value has been and T2-weighted images for anatomical and morphological assess-
demonstrated to improve FF estimation, a single-T2* model ments of the lumbar region.
may not accurately model the effects of the two separate sig- Single-voxel MRS of the third vertebral body was performed
nals of water and fat because the T2* decay rates of water using a stimulated echo acquisition mode sequence without water
and fat are not equal.16 A dual-T2* correction can reduce suppression. After local shimming and gradient adjustments, data
errors in FF estimates from single-T2* reconstruction at were obtained using the following parameters: volume of interest
high FF,10 but has inherently worse noise performance.16 (VOI), 1.5 3 1.5 3 1.5 cm3, TR, 6000 msec (a long TR was
To minimize the aforementioned confounding factors, chosen to minimize T1-bias effects); four TE values, 11/15/20/25
msec (short TE was to reduce J-coupling effects, and multiecho
we aimed to 1) validate CSE imaging with a complex spec-
acquisition allowed the calculation of T2-weighting effects and the
tral model of fat and T20 -correction10 using a priori-known
T2-corrected area of individual spectral peaks, as the different fat
T2 for quantifying vertebral marrow fat content in post-
peaks have different T2 values)1; and data points, 4096; band-
menopausal populations, using multiecho T2-corrected width, 5000 Hz. Outer volume saturation bands were used to
MRS as reference; and 2) to examine relationships between eliminate unwanted signal contamination from outside the voxel.
bone mineral density (BMD), T-score, and marrow fat These saturation bands typically covered the adjacent vertebral discs
assessed by CSE imaging and MRS. and cerebrospinal fluid.18 We set a distance (2–3 mm) between the
VOI and saturation-band to avoid the missing signal from VOI.
Materials and Methods CSE water–fat MRI was performed using an eight-echo 3D
Subjects spoiled gradient echo sequence in the spine sagittal plane. The
In this cross-sectional study, 83 postmenopausal women were imaging parameters were: four echoes per TR, monopolar readouts
recruited from the community between January 2014 and Septem- and flyback gradients; TR, 10 msec; number of TEs, 8 (TEinit,
ber 2015. Inclusion criteria were: women were ambulatory; more 1.48 msec; DTE, 1.2 msec); field of view, 400 3 400 mm; fre-
than 50 years old; were at least 1 year postmenopausal; and able to quency direction, A/P (to minimize breathing artifacts); matrix,
give informed consent. Key exclusion criteria included any disease 256 3 256; slice thickness, 3 mm; flip angle, 38 (given the large
known to affect bone metabolism such as hypothalamic or pitui- T1 difference between water and fat components in the bone mar-
tary disorders, diabetes mellitus, vitamin D deficiency, impaired row,19 a small excitation flip angle was used to minimize T1-bias
renal function, reported malignancy or exposure to radiation; use effects)8,15; bandwidth, 1.3 kHz/pixel; SENSE acceleration, 2; and
of medications known to influence bone metabolism such as estro- number of signals averaged, 1.
gen, bisphosphonates, glucocorticoids, or regular use of aspirin;
and any confounders that had the potential to interfere with the Data Analysis for MRS
interpretation of the findings such as smoking history, drinkers, Theoretically, there are nine resonances in a typical marrow spec-
vertebral hemangiomas, or silent vertebral body fractures. Drinkers trum2,6,20: CH3 methyl protons (0.90 ppm), bulk CH2 methylene
were those who drank an alcoholic beverage more than once per protons (1.30 ppm), CH2 methylene protons a- (2.25 ppm) and
January 2017 67

TABLE 1. Baseline Characteristics of Participants in the Three Groups
Normal BMD group Osteopenia group Osteoporosis group

(n 5 41) (n 5 26) (n 5 16)
Age (years) 61.5 6 7.1 63.0 6 5.1 65.5 6 6.8b

YSM (years) 10.8 6 7.7 13.8 6 5.7 15.3 6 6.9b
Height (cm) 156.3 6 5.8 156.9 6 5.6 155.0 6 5.5
Body weight (kg) 59.8 6 7.4 56.6 6 6.7 53.4 6 5.8b
BMI (kg/m2) 24.4 6 2.5 23.0 6 2.7 22.3 6 2.8b
Vertebral BMD (g/cm2) 1.158 6 0.117 0.910 6 0.031a 0.766 6 0.070a,c
Femoral neck BMD (g/cm2) 0.885 6 0.118 0.782 6 0.095a 0.699 6 0.096a,c
Total hip BMD (g/cm2) 0.954 6 0.120 0.836 6 0.084a 0.726 6 0.080a,c
BMD, bone mineral density; BMI, body mass index; YSM, years since menopause. Data are presented as mean 6 SD.
a
P < 0.001 and bP < 0.05 compared with the normal BMD group, cP < 0.05 compared with the osteopenia group using one-way
ANOVA (multiple comparisons were made with the Bonferroni correction).
b- (1.59 ppm) to the carbonyl, methylene protons a to a double physicist using the Advanced Method for Accurate, Robust and
bond (2.03), diallylic CH2 protons (2.77 ppm), glycerol (4.10– Efficient Spectral fitting of MRS data (AMARES) algorithm with
4.35 ppm), H2O (4.65 ppm), and olefinic protons (5.19–5.31 use of prior knowledge in the jMRUI software (http://www.jmrui.
ppm). T2-corrected MRS spectra acquired from the third vertebra eu).21 The rules applied in the fitting procedure were22: the line-
were exported and processed to estimate marrow FF by an MR width of the water and methylene (1.3 ppm) peaks was
C
O
L
O
R
FIGURE 1: Two examples of marrow fat fraction (FF) results from chemical shift-encoded (CSE) water–fat imaging and MRS. CSE-
based FF and MRS-based FF agree closely in both subjects [a subject with normal bone mineral density (a) and the other with
osteopenia (b)]. Fat peak assignment: (1) –CH5CH– and –CH–O–CO– (5.19–5.31 ppm); (2) H2O (4.65 ppm); (3) –CO–CH2–CH2–
(2.25 ppm) and –CH2–CH5–CH–CH2– (2.03 ppm); (4) –CO–CH2–CH2– (1.59 ppm); (5) –(CH2)n– (1.30 ppm); (6) –(CH2)n–CH3 (0.90
ppm); (7) –CH2–O–CO– (4.20 ppm); (8) –CH5CH–CH2–CH5CH– (2.77 ppm). Peaks 4 and 7 are buried within the methylene (1.3
ppm) and water peaks, respectively. Peak 8 is small and is rarely seen in the human marrow clinically.
68 Volume 45, No. 1

spectrum was modeled using the vertebral marrow fat spectrum

characterized by Karampinos et al10 and Dieckmeyer et al1 instead
of the choice of the liver fat spectrum. A T20 -correction using a
priori-known T2 equal to the average previously measured values
(considering women only, the corresponding mean values were
determined to be T2Water, priori 5 24.6 msec and T2Fat, priori 5
72.6 msec for the vertebral bodies1) was chosen because it can
remove T2* bias without reducing noise performance in vertebral
FF map, as described in detail by Karampinos et al.10 Image recon-
structions yielded coregistered a water-only, fat-only, and quantita-
tive FF map. The FF map accurately reflected the underlying
proton density ratio of the fat signal over the sum of water and fat
signals in a range of 0–100%, when using a multipeak spectral
FIGURE 2: Graph illustrates the mean marrow fat fraction (FF)
measured by MRS and chemical shift-encoded (CSE) imaging model of fat and a small excitation flip angle. The multiecho
sequence for the three groups. Data are presented as mean 6 water–fat algorithms can jointly quantify fat content and T2* per
SD. Group comparisons were performed using ANCOVA fol- voxel.2,8 The FF maps were exported to an OsiriX DICOM viewer
lowed with Bonferroni post-hoc test after adjusting for age,
to manually draw regions of interest (ROIs) by two independent
years since menopause, and body mass (aP < 0.001 compared
with normal bone mass and group; bP < 0.001 compared with raters (X.L. and G.L., with 5 and 13 years of working experience
normal bone mass and osteopenia groups). with MRI, respectively). Three ROI measurements were made
from the three most central slices depicting each vertebra (from L1
unconstrained, all lipid peaks had a linewidth equal to that of the to L4, covering three-fourths of the vertebral height) and were
peak at 1.3 ppm, zero- and first-order phase correction was esti- excluded from the cortical bone and endplates when drawing
mated by AMARES, resonance frequencies were constrained to lie ROIs. The FF values were obtained by averaging the values
within 60.05 ppm of the peaks’ known resonance frequencies, and obtained from the three slices selected.
all the observed or measurable fat peaks were modeled by multiple
Gaussian resonances. The 1.3 ppm lipid peak was modeled by Reproducibility Analysis for CSE Water–Fat
three Gaussian resonances, while the 0.9 ppm and 2.1 ppm lipid Imaging
peaks were each modeled by two Gaussians, and the 2.77 ppm Eight randomly selected participants from the study cohort were
peak as a single Gaussian resonance. The composite water and fat scanned three times with repositioning to assess the repeatability of
peaks in the 4–6 ppm range were modeled by five unconstrained FF measurements at CSE water–fat imaging. The repeatability was
Gaussians (that is the amplitude, linewidth and frequency of the expressed as root mean square absolute precision error in [%]
peaks were all fitted freely). The signal intensity of T2-corrected (absolute units) and as root mean square coefficient of variation
water and fat peaks were calculated from the areas of water and fat (RMS-CV) in [%] (relative units), according to Gluer et al.24
at each TE by a nonlinear least-squares fitting algorithm, using the To assess the interobserver reliability, two observers inde-
data from the different TEs. The diallylic peak at 2.77 ppm was pendently analyzed all the FF maps. In the intraobserver analyses,
too weak to be accurately measured in the bone marrow at clinical two measurements of the primary rater on two separate occasions
field strengths (3T), and the glycerol peak at 4.2 ppm is not 2 months apart were compared. Both the observers were blinded
clearly distinguishable from the 4.65 ppm water peak because of
to the results obtained earlier.
the superimposed water peak. Additionally, several resonances (at
5.31 and 5.19 ppm; 2.25 and 2.03 ppm; 1.59 and 1.30 ppm) are
Statistical Analysis
not resolvable and appear as single peaks. Therefore, we evaluated
Mean 6 standard deviation (SD) values were calculated for each
the FF values using the formula: FF 5 Ilipid/(Ilipid 1 Iwater) 3
variable. Differences between the groups were assessed using one-
100, where Ilipid is the sum of the area amplitudes of the resonan-
way analysis of variance (ANOVA) followed with Bonferroni post-
ces (at 0.9, 1.30, 1.59, 2.03, 2.25, 5.19, and 5.31 ppm) and Iwater
hoc multiple comparisons and analysis of covariance (ANCOVA)
is the area amplitude of H2O resonance.
when controlling for covariates. Correlation analyses were used to
Data Analysis for Water–Fat Imaging describe associations between BMD, T-score, and FF. Intrarater
The gradient echo imaging data were processed offline using in- and interrater reliability were conducted using the intraclass corre-
house-built routines implemented in MatLab (v. 2014-64bit, lation coefficient (ICC). To establish the level of agreement
MathWorks, Natick, MA). To increase the accuracy of the recon- between the fat content with the CSE water–fat separation
struction, a region-growing algorithm was first applied to assess the sequence and that with MRS, linear regression and Bland–Altman
field-map variation similar to that described in Yu et al23 because analysis were performed.25 Finally, Lin’s concordance correlation
the water–fat separation strongly depends on the initial guess of coefficient was calculated to describe the strength of agreement:
the B0 inhomogeneity map. A complex-based water–fat decompo- >0.95 indicates almost perfect agreement; 0.80–0.95, substantial
sition was then performed using a T20 -correction and a precali- agreement; 0.70–0.80, moderate agreement; and <0.70, poor
brated eight-peak spectral model of fat, accounting for the presence agreement.26 Data was analyzed using IBM SPSS v. 23.0 (IBM
of the multiple peaks in the fat spectrum.11 The precalibrated fat SPSS, Armonk, NY). Statistical significance was set to P < 0.05.
January 2017 69

FIGURE 3: (a) Scatterplot and regression line show close correlation of marrow fat fraction (FF) between MRS and chemical shift-
encoded (CSE) imaging in 83 studies (Pearson correlation coefficient, r 5 0.979; P < 0.001). (b) Bland–Altman plot, with dashed
lines representing the 95% CI. All data points were within the limits of agreement (dotted lines), corresponding to 6 1.96 SDs
from mean.
Results FF results from all the subjects. There were significant differ-
Baseline Characteristics ences in the average fat content among the three groups
T1 Table 1 lists the demographic and clinical characteristics of (P < 0.001) after adjustment for age, YSM, and BMI.
the study population. The mean age of the 83 postmeno-
pausal women included in this study was 62.8 6 6.6 years. Repeatability of CSE Imaging Scans
F1 Figure 1 illustrates the FF results from two subjects: one with The RMS absolute precision error of the FF measurements
normal bone mass and the other with osteopenia. Excellent at CSE imaging was 1.45%. RMS-CV indicates the average
fat and water separation was achieved in both subjects. Asso- error of FF measurements. The RMS-CV of FF measure-
F2 ciated MRS plots are also shown. Figure 2 summarizes the ments was 2.50%.
TABLE 2. Correlations Between Bone Mineral Density (BMD), T-score, and Marrow FF Measured by Chemical
Shift-Encoded (CSE) Imaging and MRS
BMD T-score
Parameters Groups r P-value r P-value
MRS-based FFa Normal bone mass –0.555 <0.001 –0.543 <0.001

Osteopenia –0.539 0.005 –0.501 0.009
Osteoporosis –0.706 0.002 –0.742 0.001
a
CSE-based FF Normal bone mass –0.571 <0.001 –0.559 <0.001
Osteopenia –0.560 0.003 –0.526 0.006
Osteoporosis –0.710 0.001 –0.747 0.001
b
MRS-based FF Normal bone mass –0.536 0.001 –0.512 0.001
Osteopenia –0.536 0.005 –0.497 0.016
Osteoporosis –0.610 0.021 –0.665 0.010
b
CSE-based FF Normal bone mass –0.545 <0.001 –0.524 0.001
Osteopenia –0.591 0.003 –0.531 0.009
Osteoporosis –0.604 0.022 –0.664 0.010
a b
P-values were calculated by bivariate correlations and P-values were calculated by partial correlations after controlling for age, years
since menopause, and body mass index.
70 Volume 45, No. 1

Intra- and Interrater Reliability corrected for any T2* decay effects nor modeled multipeak
For FF measurements at CSE imaging, the intrarater reli- fat spectrum. Similarly, Takasu et al14 used the IDEAL algo-
ability for duplicate measurements was substantial agreement rithm to measure FF and discriminate between patients suf-
(ICC 5 0.993, 95% confidence interval [CI] 5 0.990– fering from symptomatic and asymptomatic myeloma but
0.996; P < 0.001). Interobserver agreement for the average without employing T2*-correction and the precalibrated
FF value (L1–L4) at CSE imaging was excellent (ICC 5 multipeak fat spectrum. Our data suggest that vertebral
0.982, 95% CI 5 0.996–0.990 for normal bone mass; ICC marrow fat content could be reproducibly assessed using
5 0.976, 95% CI 5 0.947–0.989 for osteopenia; and ICC CSE water–fat imaging, with an absolute precision error of
5 0.979, 95% CI 5 0.941–0.993 for osteoporosis, 1.45% as well as excellent intra- and interrater reliability.
respectively). Interestingly, the most recent publication by Baum et al8
indicates that whole spine marrow fat could be reproducibly
Agreement between FF at CSE imaging and MRS- (absolute precision error was 1.7% averaged over C3–L5)
measured FF
assessed by using CSE-based water/fat MRI with a T2*-cor-
Mean FF with MRS was 62.1% 6 11.1%, and mean FF
rection and vertebral marrow FF showed anatomical varia-
with CSE imaging was 60.4% 6 10.1% (P > 0.05). Figure
tions with increasing values from C3 to L5. Our results also
F3 3a shows the scatterplot between the marrow fat content
support a previous study15 in which a six-echo gradient-
with MRS and that with the CSE water–fat separation
echo imaging with T2*-correction provided an accurate
sequence, with linear regression. The Pearson correlation
means of determining marrow fat content in the presence of
coefficient was 0.979 (P < 0.001). Lin’s concordance corre-
trabecular bone.
lation coefficient was 0.962. Using the Bland–Altman plot,
Bone marrow is a complex connective tissue, consist-
the CSE image method demonstrated good agreement with
ing of both hematopoietic and fatty marrow. Bone marrow
MRS, where all points were within the 95% limit of agree-
is characterized by a large difference in T1 relaxation times
ment (Fig. 3b).
of fat and water protons, which leads to bias during quanti-
Correlations Between Marrow Fat Content, BMD, fication of fat content.19 The choice of flip angle is critical
and T-score in the CSE-MRI FF quantification. In relation to TR, the
In the subgroups of varying BMD, inverse correlations were flip angle (38 used in our study) should be sufficiently low
found to be very similar between BMD, T-score, and CSE- to minimize T1-bias effects, but sufficiently high to main-
based FF and between BMD, T-score, and MRS-based FF. tain an adequate signal-to-noise ratio.9,29
Adjusting for age, YSM, and BMI did not attenuate the Fat has a number of spectral peaks. The presence of
T2 relationships between BMD, T-score and FF (Table 2). multiple peaks in the fat spectrum complicates the ability to
correct for spatial chemical shift artifacts, particularly the
Discussion spectral peak from olefinic protons at 5.3 ppm (being close
The strength of our work was that initial clinical experience to the water peak).11 Indeed, marrow fat, subcutaneous and
of CSE water–fat imaging using a multipeak spectral model visceral adipose tissue such as liver fat exhibit distinct,
of fat, suitable T2*-corrected method, and a small excitation tissue-specific properties and therefore different metabolic
flip angle applied to marrow fat quantification. Further, a activities.30 In comparison to liver and subcutaneous fat, red
good sample size for the subgroups of varying BMD could marrow has higher levels of unsaturated fatty acid.31 The fat
enhance the statistical power, and the statistical analysis took spectrum varies from organ to organ and a fat spectrum
most of the potential confounding factors into account. Our suitable to liver may not give accurate values for bone mar-
study showed that CSE imaging is equally accurate in char- row. In several previous studies,2,15,32 the fat spectrum was
acterizing marrow fat content as multi-TE T2-corrected modeled using the liver fat spectrum characterized by Ham-
MRS. ilton et al33 and only the three main peaks (0.9, 1.3, and
CSE water–fat MRI offers a quantitative and qualita- 2–2.2 ppm) were considered.32 In our work, the presence of
tive tool to quantify not only adipose tissue content, but multiple peaks in the fat spectrum was considered using a
also an index of the underlying tissue inhomogeneity. There precalibrated fat spectrum characterized in the red bone
are several studies using chemical-shift MRI on vertebrae. marrow.1,10 Excellent fat and water separation was achieved
Ojanen et al27 reported that in-phase and out-of-phase MRI in all the subjects. Since the use of an incorrect fat model
can provide similar vertebral marrow fat estimation as MRS. may result in some errors in MRI estimation of FF, further
Similar results were observed in the study done by Regis- studies are required to elucidate the difference of marrow
Arnaud et al.13 Conversely, Gokalp et al28 reported vertebral proton density FF measurements using various multipeak
marrow FF calculated with a double-echo fast low-angle fat models.
shot sequence is not a reliable parameter for predicting T2* correction is a crucial prestep for quantitative
BMD in female patients. However, these studies neither water–fat imaging, particularly in bone marrow because the
January 2017 71

presence of bone trabeculae creates local heterogeneities in In conclusion, CSE imaging with a complex spectral
the magnetic field that shorten T2*.1,2 Unlike T1 bias effects model of fat and T20 -correction is equally as accurate in
and accurate spectral modeling, T2* correction can’t be characterizing vertebral marrow fat content as MRS. CSE
achieved through choosing appropriate imaging parameters water–fat MRI offers an additional advantage of inhomoge-
(such as low flip angle) and prior knowledge (relative ampli- neous determination of marrow tissue in FF maps over
tudes and position of multiple fat peaks). T2* correction MRS. Given its excellent correlation and concordance with
can be carried out by modifying the signal model to account MRS, the CSE water–fat separation sequence has the poten-
for the signal decay due to T2*. The majority of CSE imag- tial to replace MRS for fat quantification.
ing that correct for T2* decay to improve the accuracy of
marrow fat quantification assume a common (single-T2*) Acknowledgments
for water and fat,2,15,34,35 which is valid for liver applica-
Contract grant sponsor: Natural Science Foundation of
tions at relatively low fat content. However, water and fat
Shanghai Science and Technology Commission; contract
signals have independent T2* that may influence estimation
grant number: 14ZR1442300; Contract grant sponsor:
of fat content, particularly at regions of high fat content Shanghai Municipal Commission of Health and Family
and short T2*, which can be frequently encountered in bone Planning; contract grant number: 201440387; Contract
marrow.10,32 In liver fat quantification, a single-T2* correc- grant sponsor: National Natural Science Foundation of
tion was chosen because of the advantage of a range of China; contract grant number: 81202809, 81373856.
clinically relevant signal-to-noise ratios and water/fat ratios The authors thank for Dongmei Wu, PhD, at MRI Labora-
at low fat concentrations compared to dual-T2* correction.36 tory of East China Normal University for fruitful discus-
A dual-T2* correction can correct for the marrow fat bias at sions during preparation of the article and Yongming Dai,
nominal FF value close to 50%, but shows poor noise per- PhD, for help with pulse sequence optimization.
formance at low FF.10 A recent publication by Karampinos
et al10 indicated that using a priori-known T2, a T20 -correc- Conflict of Interest
tion can remove the FF bias induced by the difference of The authors state no conflicts of interest.
T2* between water and fat components without reducing
noise performance in the FF map of the lumbar spine. Typi-
cally, based on such T20 -correction using a priori-known T2,
References
eight-echo acquisitions used in the present study have pro-
1. Dieckmeyer M, Ruschke S, Cordes C, et al. The need for T(2) correc-
vided an excellent balance between accurate estimation of tion on MRS-based vertebral bone marrow fat quantification: implica-
T2* decay effects and acquisition time. tions for bone marrow fat fraction age dependence. NMR Biomed
2015;28:432–439.
Several studies have documented inverse relationships
2. Karampinos DC, Melkus G, Baum T, Bauer JS, Rummeny EJ, Krug R.
between marrow fat content and bone mass. The correlation Bone marrow fat quantification in the presence of trabecular bone: Ini-
coefficients between marrow fat content and BMD that we tial comparison between water-fat imaging and single-voxel MRS.
Magn Reson Med 2014;71:1158–1165.
observed in the subgroups of varying BMD are higher than
3. Bolan PJ, Arentsen L, Sueblinvong T, et al. Water-fat MRI for assess-
those of prior reports (typical r 5 –0.32 to – ing changes in bone marrow composition due to radiation and chem-
0.558).7,12,37,38 A rational explanation is that multiple con- otherapy in gynecologic cancer patients. J Magn Reson Imaging
founding factors, particularly the presence of multiple peaks 2013;38:1578–1584.
in the fat spectrum and T2* decay effects, were considered 4. Banse X, Devogelaer JP, Munting E, Delloye C, Cornu O, Grynpas M.
Inhomogeneity of human vertebral cancellous bone: systematic den-
in our study. We also documented the inverse association of sity and structure patterns inside the vertebral body. Bone 2001;28:
BMD with CSE-based FF or MRS-based FF was very simi- 563–571.
lar, which further supports that CSE imaging is equally 5. Schafer AL, Li X, Schwartz AV, et al. Changes in vertebral bone mar-
row fat and bone mass after gastric bypass surgery: A pilot study.
accurate in characterizing marrow fat content as MRS. Bone 2015;74:140–145.
We also acknowledge that, because of the necessity of 6. Karampinos DC, Ruschke S, Gordijenko O, et al. Association of MRS-
image registration and establishment of CSE-based FF based vertebral bone marrow fat fraction with bone strength in a
human in vitro model. J Osteoporos 2015;2015:152349.
maps, the whole analysis process was relatively time-
consuming, which is not ideal in real clinical practice. More 7. Griffith JF, Yeung DK, Antonio GE, et al. Vertebral marrow fat content
and diffusion and perfusion indexes in women with varying bone den-
easily handled software or accelerating the method is greatly sity: MR evaluation. Radiology 2006;241:831–838.
needed. Second, a comparison was performed between one 8. Baum T, Yap SP, Dieckmeyer M, et al. Assessment of whole spine ver-
vertebral body with MRS vs. four vertebral levels with CSI tebral bone marrow fat using chemical shift-encoding based water-fat
MRI. J Magn Reson Imaging 2015;42:1018–1023.
imaging, and the results may have interscan variability
9. Yang IY, Cui Y, Wiens CN, Wade TP, Friesen-Waldner LJ, McKenzie
depending on the MR scanner. Finally, our study group CA. Fat fraction bias correction using T1 estimates and flip angle
only consisted of postmenopausal women. mapping. J Magn Reson Imaging 2014;39:217–223.
72 Volume 45, No. 1

10. Karampinos DC, Ruschke S, Dieckmeyer M, et al. Modeling of T2 * 24. Gluer CC, Blake G, Lu Y, Blunt BA, Jergas M, Genant HK. Accurate
decay in vertebral bone marrow fat quantification. NMR Biomed assessment of precision errors: how to measure the reproducibility of
2015;28:1535–1542. bone densitometry techniques. Osteoporos Int 1995;5:262–270.
11. Yu H, Shimakawa A, McKenzie CA, Brodsky E, Brittain JH, Reeder SB. 25. Bland JM, Altman DG. Statistical methods for assessing agreement
Multiecho water-fat separation and simultaneous R2* estimation with between two methods of clinical measurement. Lancet 1986;1:307–
multifrequency fat spectrum modeling. Magn Reson Med 2008;60: 310.
1122–1134.
26. Lin LI. A concordance correlation coefficient to evaluate reproducibil-
12. Li GW, Xu Z, Chen QW, et al. Quantitative evaluation of vertebral ity. Biometrics 1989;45:255–268.
marrow adipose tissue in postmenopausal female using MRI chemical
27. Ojanen X, Borra RJ, Havu M, et al. Comparison of vertebral bone
shift-based water-fat separation. Clin Radiol 2014;69:254–262.
marrow fat assessed by 1H MRS and inphase and out-of-phase MRI
13. Regis-Arnaud A, Guiu B, Walker PM, Krause D, Ricolfi F, Ben Salem among family members. Osteoporos Int 2014;25:653–662.
D. Bone marrow fat quantification of osteoporotic vertebral compres-
28. Gokalp G, Mutlu FS, Yazici Z, Yildirim N. Evaluation of vertebral bone
sion fractures: comparison of multi-voxel proton MR spectroscopy and
marrow fat content by chemical-shift MRI in osteoporosis. Skeletal
chemical-shift gradient-echo MR imaging. Acta Radiol 2011;52:1032– Radiol 2011;40:577–585.
1036.
29. Johnson BL, Schroeder ME, Wolfson T, et al. Effect of flip angle on
14. Takasu M, Kaichi Y, Tani C, et al. Iterative decomposition of water the accuracy and repeatability of hepatic proton density fat fraction
and fat with echo asymmetry and least-squares estimation (IDEAL) estimation by complex data-based, T1-independent, T2*-corrected,
magnetic resonance imaging as a biomarker for symptomatic multiple spectrum-modeled MRI. J Magn Reson Imaging 2014;39:440–447.
myeloma. PLoS One 2015;10:e0116842.
30. Paccou J, Hardouin P, Cotten A, Penel G, Cortet B. The role of bone
15. Gee CS, Nguyen JT, Marquez CJ, et al. Validation of bone marrow marrow fat in skeletal health: usefulness and perspectives for clini-
fat quantification in the presence of trabecular bone using MRI. cians. J Clin Endocrinol Metab 2015;100:3613–3621.
31. Ren J, Dimitrov I, Sherry AD, Malloy CR. Composition of adipose tis-
16. Horng DE, Hernando D, Hines CD, Reeder SB. Comparison of R2* sue and marrow fat in humans by 1H NMR at 7 Tesla. J Lipid Res
correction methods for accurate fat quantification in fatty liver. 2008;49:2055–2062.
32. Le Ster C, Gambarota G, Lasbleiz J, Guillin R, Decaux O, Saint-Jalmes
17. Li GW, Chang SX, Xu Z, Chen Y, Bao H, Shi X. Prediction of hip H. Breath-hold MR measurements of fat fraction, T1, and T2* of water
osteoporotic fractures from composite indices of femoral neck and fat in vertebral bone marrow. J Magn Reson Imaging 2016; doi:
strength. Skeletal Radiol 2013;42:195–201. 10.1002/jmri.25205 [Epub ahead of print].
18. Liney GP, Bernard CP, Manton DJ, Turnbull LW, Langton CM. Age, 33. Hamilton G, Yokoo T, Bydder M, et al. In vivo characterization of the
gender, and skeletal variation in bone marrow composition: a prelimi- liver fat (1)H MR spectrum. NMR Biomed 2011;24:784–790.
nary study at 3.0 Tesla. J Magn Reson Imaging 2007;26:787–793.
34. Arentsen L, Yagi M, Takahashi Y, et al. Validation of marrow fat
19. Schick F, Bongers H, Jung WI, et al. Proton relaxation times in human assessment using noninvasive imaging with histologic examination of
red bone marrow by volume selective magnetic resonance spectros- human bone samples. Bone 2015;72:118–122.
copy. Appl Magn Reson 1992;3:947–963.
35. MacEwan IJ, Glembotski NE, D’Lima D, et al. Proton density water
20. Di Pietro G, Capuani S, Manenti G, et al. Bone marrow lipid profiles fraction as a biomarker of bone marrow cellularity: validation in ex
from peripheral skeleton as potential biomarkers for osteoporosis: A vivo spine specimens. Magn Reson Imaging 2014;32:1097–1101.
(1)H-MR spectroscopy study. Acad Radiol 2016;23:273–283.
36. Chebrolu VV, Hines CD, Yu H, et al. Independent estimation of T*2
21. Naressi A, Couturier C, Devos JM, et al. Java-based graphical user for water and fat for improved accuracy of fat quantification. Magn
interface for the MRUI quantitation package. MAGMA 2001;12:141– Reson Med 2010;63:849–857.
152.
37. Yeung DK, Griffith JF, Antonio GE, Lee FK, Woo J, Leung PC. Osteo-
22. Griffith JF, Yeung DK, Chow SK, Leung JC, Leung PC. Reproducibility porosis is associated with increased marrow fat content and
of MR perfusion and (1)H spectroscopy of bone marrow. J Magn decreased marrow fat unsaturation: a proton MR spectroscopy study.
Reson Imaging 2009;29:1438–1442. J Magn Reson Imaging 2005;22:279–285.
23. Yu H, Reeder SB, Shimakawa A, Brittain JH, Pelc NJ. Field map esti- 38. Griffith JF, Yeung DK, Ma HT, Leung JC, Kwok TC, Leung PC. Bone
mation with a region growing scheme for iterative 3-point water-fat marrow fat content in the elderly: a reversal of sex difference seen in
decomposition. Magn Reson Med 2005;54:1032–1039. younger subjects. J Magn Reson Imaging 2012;36:225–230.
January 2017 73

ORIGINAL RESEARCH
Background Parenchymal Enhancement

Over Exam Time in Patients With and
Without Breast Cancer
Amy Melsaether MD,1* Akshat C. Pujara MD,1 Kristin Elias MD,1
Kristine Pysarenko MD,1 Anjali Gudi MD,1 Katerina Dodelzon MD,1
James S. Babb PhD,1 Yiming Gao MD,1 and Linda Moy MD1,2
Purpose: To compare background parenchymal enhancement (BPE) over time in patients with and without breast
cancer.
Materials and Methods: This retrospective Institutional Review Board (IRB)-approved, Health Insurance Portability and
Accountability Act (HIPAA)-compliant study included 116 women (25–84 years, mean 54 years) with breast cancer who
underwent breast magnetic resonance imaging at 3T between 1/2/2009 and 12/29/2009 and 116 age and date-of-
exam-matched women without breast cancer (23–84 years, mean 51 years). Two independent, blinded readers (R1, R2)
recorded BPE (minimal, mild, moderate, marked) at three times (100, 210, and 320 seconds postcontrast). Subsequent
cancers were diagnosed in 9/96 control patients with follow up (12.6–93.0 months, mean 63.6 months). Exact Mann–
Whitney, Fisher’s exact, and McNemar tests were performed.
Results: Mean BPE was not found to be different between patients with and without breast cancer at any time (P 5 0.36–
0.64). At time 2 as compared with time 1, there were significantly more patients, both with and without breast cancer,
with BPE >minimal (R1: 90 vs. 41 [P < 0.001] and 81 vs. 36 [P < 0.001]; R2: 84 vs. 52 [P < 0.001] and 79 vs. 43 [P <
0.001]) and BPE >mild (R1: 59 vs. 10 [P < 0.001] and 47 vs. 13 [P < 0.001]; R2: 49 vs. 12 [P < 0.001] and 41 vs. 18 [P <
0.001]). BPE changes between times 2 and 3 were not significant (P 5 0.083–1.0). Odds ratios for control patients develop-
ing breast cancer were significant only for R2 and ranged up to 7.67 (1.49, 39.5; P < 0.01) for BPE >mild at time 2.
Conclusion: BPE changes between the first and second postcontrast scans and stabilizes thereafter in most patients.
Further investigation into the most clinically relevant timepoint for BPE assessment is warranted.
setting of neoadjuvant chemotherapy.13 It follows, then, that

B ackground parenchymal enhancement (BPE) is the vol-
ume and intensity of normal fibroglandular tissue
(FGT) enhancement on breast magnetic resonance imaging
higher levels of BPE are associated with higher estrogenic
states, and possibly therefore with an increased risk of devel-
(MRI).1 Initial reports assessed the impact of BPE on the oping breast cancer. Indeed, associations between BPE >mild
diagnostic performance of breast MRI2,3 and focused on the and the presence of a breast cancer,14,15 as well as BPE
first postcontrast scan, since this is when cancer detection is >minimal and the development of a breast cancer,16 have
performed.4 been suggested. Dontchos et al16 and King et al14 both
The degree of BPE is directly related to blood flow, assessed BPE on the first postcontrast image set; however, the
inversely related to menopausal status,5 and varies along with first postcontrast image set varied between 90 and 110 sec-
menstrual cycle phases.1,6 BPE decreases with estrogen- onds for Dontchos et al16 and was not specified for King
reducing therapies including tamoxifen,7,8 aromatase inhibi- et al.14 In contrast, Telegrafo et al15 assessed BPE on the third
tors,8–10 bilateral salpingo-oophorectomy,11,12 and in the postcontrast image set, 3 minutes postcontrast injection.
*Address reprint requests to: A.M., Breast Imaging, 160 E 34th St, 3rd Fl., New York, NY 10016. E-mail: Amy.melsaether@nyumc.org
From the Department of Radiology, NYU Perlmutter Cancer Center, New York University School of Medicine, New York, New York, USA; 2Department of
1
Radiology, Center for Advanced Imaging Innovation and Research (CAI(2)R), New York University School of Medicine, New York, New York, USA; and
Current affiliation for Katerina Dodelzon: Department of Radiology, Weill Cornell Imaging at New York-Presbyterian, New York, NY

V

Melsaether et al.: BPE Over Exam Time
TABLE 1. Breast MRI Acquisition Parameters
Parameter Sagittal T2-weighted Sagittal T1-weighted

gradient recalled echo
Time to repetition (msec) 4470 3.57

Time to echo (msec) 84 1.08
Flip angle (degrees) 130 12
Field of view (mm) 270 270
Matrix 163 3 320 235 3 384
Slice thickness (mm) 3.0 1.0
Voxel size (mm) 1.2 3 0.8 3 3.0 1.0 3 0.7 3 1.0
The growing importance of BPE is reflected in the cancer were reviewed to assess for development of subsequent
updated 2013 Breast Imaging Reporting and Data System (BI- cancers.
RADS) lexicon,17 which, in accordance with timing for cancer
Image Acquisition
detection, includes a recommendation for BPE assessment at
MR images were acquired using a dedicated breast coil on a 3T
90 seconds postcontrast.4 Although BPE is known to con- magnet with the patient in the prone position; acquisition parame-
tinue to increase over time,4,18 the behavior of BPE over the ters are reported in Table 1. T1
course of an examination and its relationship to breast cancer Both breasts were imaged with a fat-suppressed sagittal T2-
risk at various timepoints is not yet characterized. weighted sequence, a sagittal T1-weighted nonfat-suppressed 3D
Thus, the purpose of this study was to evaluate BPE volumetric scan, and a sagittal T1-weighted fat suppressed 3D vol-
throughout the MRI exam in patients with and without a umetric scan that was obtained before and at three timepoints after
known cancer, and to assess long-term follow-up in patients intravenous injection of 0.1 mmol/L gadolinium chelate/kg body
without a known cancer based on initial degree of BPE. weight. These three sagittal postcontrast image sets were completed
at 100, 210, and 320 seconds postinjection. Sagittal postcontrast
Materials and Methods subtraction images were generated.
Written informed consent was waived for all patients included in
Image Interpretation
this Institutional Review Board-approved, Health Insurance Port-
All 232 cases were randomized and independently interpreted for
ability and Accountability Act-compliant, retrospective study.
BPE at each timepoint by two breast imagers (R1 and R2, both 6
Between 1/1/2009 and 12/31/2009, 825 unique patients under-
years of experience), who were blinded to clinical history and exam
went breast MRI (ages 23–85 years, mean 51 years). Of these 825
indication, in accordance with the BI-RADS lexicon (minimal up to
patients, 128 without a history of breast radiation were imaged to
25%; mild 26–50%; moderate 51–75%; marked >75%).5 Readers
evaluate the extent of disease in the setting of newly diagnosed
also assessed each case for FGT in accordance with the BI-RADS
breast cancer. Twelve of these examinations did not have three
lexicon (almost entirely fat up to 25%; scattered fibroglandular tissue
retrievable postcontrast image sets in our picture archiving and
26–50%; heterogeneous fibroglandular tissue 51–75%; extreme
communication system (PACS) and were excluded, yielding 116 fibroglandular tissue >75%).5 Each breast was assessed independ-
included subjects (all female, ages 25–84 years, mean 54 years). ently, and in cases of asymmetry the higher level of BPE or FGT
These 116 subjects were age (within 5 years) and date-of-exam was assigned to the case. Readers could use precontrast, postcontrast
(within 1 year)-matched with 116 patients without a breast cancer subtracted and unsubtracted, and T2-weighted images in their assess-
or history of breast radiation who underwent breast MRI (ages 23– ment of BPE and FGT. Postcontrast subtracted image sets were typi-
84 years, mean 51 years). Matching was based on closest age in cally used to assess BPE, with the unsubtracted images available for
years and months within the provided timeframe and all matched problem solving in cases of motion, and precontrast unsubtracted T1
subjects were included prior to image review. Indications for MRI and T2 images were typically used to assess FGT.
and other patient characteristics were collected by chart review.
Mammographic density was collected from reports of mammo- Statistical Analysis
grams performed within 6 months of the MRI. For the 10 patients Interreader agreement was assessed using the linear weighted kappa
with breast cancer for whom menopausal status was not available, coefficient. Kappa (K) was interpreted as an indication of poor
age was used as a proxy, with 8 women 50 and under (ages 38–48 agreement when less than zero, as slight agreement when 0 K
years, mean 44 years) considered premenopausal and 2 women 0.2, as fair agreement when 0.20 K 0.4, as moderate agreement
aged 54 and 69 years considered postmenopausal.19 Follow-up when 0.40 K 0.6 and as substantial agreement when K > 0.6.
imaging and biopsy results in the group of patients without breast The BPE scores are summarized in terms of the mean and standard
January 2017 75

TABLE 2. Patient Characteristics
Patient group With breast cancer Without

breast cancer
Patients (no.) 116 116

Age (years, mean 1 SD) 54 1 11 51 1 11
Risk factors and indications for breast MRI [no. (%)]
Extent of disease 116 (100)
Family history of breast cancer 12 (10.3) 55 (47.4)
BRCA mutation 11 (9.5) 24 (20.7)
BRCA1 3 (2.6) 2 (1.7)
BRCA2 5 (4.3) 8 (6.9)
BRCA, not specified 3 (2.6) 14 (12.1)
History of atypia 22 (19.0) 12 (10.3)
Problem solving 12 (10.3)
History of ovarian cancer 5 (4.3)
Dense breasts 3 (2.6)
Augmentation injections 2 (1.7)
History of papilloma 2 (1.7)
History of phylloides 1 (0.9)
Menopausal status [no. ( %)]
Post 54/116 (46.6)b 50/116 (43.1)
b
Pre 62/116 (53.4) 66/116 (56.9)
Week 1 5/62 (0.8) 0/66 (0.0)
Week 2 29/62 (46.8) 62/66 (93.9)
Week 3 13/62 (20.1) 1/66 (1.5)
Week 4 2/62 (3.2) 0/66 (0.0)
b
Unknown/>4 weeks 13/62 (20.1) 3/66 (4.5)
a
Mammographic density [no. ( %)]
Predominantly fatty (no.) 4/83 (4.8) 2/93 (2.2)
Scattered fibroglandular (no.) 15/83 (18.1) 24/93 (25.8)
Heterogeneously dense (no.) 41/83 (49.4) 44/93 (47.3)
Extremely dense (no.) 23/83 (27.7) 23/93 (24.7)
a
Refers to mammogram performed within 6 months of MRI, which was not available in 33 patients in the newly diagnosed breast
cancer group and 23 patients in the non-breast cancer group.
b
Menopausal status assigned by age proxy in 10 women. Two were categorized as postmenopausal (ages 54 years and 69 years) and 8
were categorized as premenopausal, week unknown (ages 38-48 years, mean 44 years).
deviation. An exact Mann–Whitney test was used to compare timepoints in terms of the percentage of subjects in each group
patient groups and subgroups in terms of the ordinal BPE assess- with BPE >minimal or BPE >mild as determined by each reader.
ments from each reader at each time and the change in BPE The Spearman rank correlation was used to characterize the associ-
between consecutive timepoints. Fisher’s exact test was used to ation of age, mammographic density, and FGT with BPE from
compare patient groups and subgroups in terms of the percentage each reader at each time. All statistical tests were conducted at the
of subjects with BPE >minimal or BPE >mild as determined by two-sided 5% significance level using SAS 9.3 (SAS Institute, Cary,
each reader at each time. A McNemar test was used to compare NC).
76 Volume 45, No. 1

TABLE 3. Correlations With Background Parenchymal Enhancement at Each of 3 Timepoints
Age Fibroglandular tissue Mammographic density

a
Time
Reader 1 Reader 2 Reader 1 Reader 2 Reader 1 Reader 2
r P r P r P r P r P r P
1 20.20, 0.002 20.17, 0.008 0.18, 0.006 0.17, 0.01 0.14, 0.07 0.14, 0.07
2 20.33, <0.001 20.27, <0.001 0.23, <0.001 0.22, 0.001 0.11, 0.14 0.15, 0.05
3 20.31, <0.001 20.29, <0.001 0.21, 0.001 0.22, 0.001 0.09, 0.23 0.14, 0.07
a
Times 1, 2, and 3 concluded 100, 210, and 320 seconds after contrast administration, respectively.
Results For reader 1, rates of asymmetry were 4/116, 19/116, and

Patient Characteristics/BPE Correlations 20/116 at times 1, 2, and 3 for breast cancer patients and
The breast cancer group included 76 women with invasive 5/116, 20/116, and 20/116 at times 1, 2, and 3 for control
ductal carcinoma (IDC), 35 women with ductal carcinoma patients. For reader 2, rates of asymmetry were 5/116, 9/
in situ (DCIS), three women with invasive lobular carci- 116, and 9/116 at times 1, 2, and 3 for breast cancer
noma (ILC), and two women with metaplastic cancer. The patients and 6/116, 6/116, and 5/116 at times 1, 2, and 3
index lesions from the extent of disease examinations for control patients. For FGT, asymmetry was rare (R1:
included 82 masses (0.6–10.0 cm, mean 2.3 cm), 24 non- breast cancer 0/116 and control 0/116; R2: breast cancer 2/
mass enhancement lesions (0.5–6.8 cm, mean 2.5 cm), one 116 and control 2/116).
0.2 cm focus and nine biopsy changes without MRI findings There was no significant difference in mean BPE
suspicious for cancer. Additional patient characteristics and between patients with and without breast cancer for either
T2 indications for MRI are listed in Table 2. reader at any time (Table 4). More patients with breast can- T4
For both readers, significant inverse correlations cer, as compared to without, fell into BPE >minimal at all
between BPE and age (r 5 –0.33 to –0.17, P 5 < 0.01– timepoints and BPE >mild at all but the first timepoint,
0.01) and significant positive correlations between BPE and but these differences were not significant (Table 4).
with FGT (r 5 0.17–0.23, P 5 < 0.01–0.01) were seen.
The strength of correlations between age, mammographic BPE Over Time
density and FGT with BPE at all three times for both read- The majority of patients both with and without breast can-
T3 ers is reported in Table 3. cer had minimal BPE at time 1 and greater than minimal
BPE thereafter (Table 4, Fig. 1). Mean BPE was signifi- F1
BPE Comparison Between Groups cantly greater at time 2 as compared with time 1 for both
As noted in Materials and Methods, in cases of asymmetry patient groups and readers (P < 0.001 for all data points)
the higher level of BPE was used to characterize the exam. and at time 3 as compared with time 2 for only the group
TABLE 4. Comparison of BPE in Patients With and Without Breast Cancer at 3 Timepoints
Mean BPE (std) BPE > minimal (n) of 116 BPE > mild (n) of 116
a
Reader Time With Without P With Without P With Without P
BC BC value BC BC value BC BC value
1 1 1.5 (0.7) 1.4 (0.7) 0.64 35.3% (41) 31.0% (36) 0.58 8.6% (10) 11.2% (13) 0.66
1 2 2.4 (0.9) 2.3 (1.1) 0.38 77.6% (90) 69.8% (81) 0.23 50.9% (59) 40.5% (47) 0.15
1 3 2.4 (0.9) 2.3 (1.1) 0.44 80.2% (93) 71.6% (83) 0.17 50.9% (59) 43.1% (50) 0.29
2 1 1.6 (0.7) 1.5 (0.8) 0.48 44.8% (52) 37.1% (43) 0.29 10.3% (12) 15.5% (18) 0.33
2 2 2.2 (1.0) 2.1 (1.0) 0.36 72.4% (84) 68.1% (79) 0.57 42.2% (49) 35.3% (41) 0.35
2 3 2.3 (1.0) 2.2 (1.0) 0.46 73.3% (85) 69.0% (80) 0.56 44.0% (51) 37.1% (43) 0.35
a
January 2017 77

C
O
L
O
R
FIGURE 1: Distribution of BPE over time in patients with breast cancer (BC) and without breast cancer (w/o BC) for each of two
readers.
78 Volume 45, No. 1

FIGURE 2: BPE at three timepoints in two patients. A–C: BPE in a 42-year-old BRCA2 1 woman is minimal on the first (A), second
(B), and third (C) postcontrast sagittal T1-weighted subtracted images. D–F: BPE in a 37-year-old woman with a family history of
breast cancer in her mother, maternal aunt, and sister, increased from mild on the first (D) to marked on the second (E) and third
(F) postcontrast sagittal T1-weighted subtracted images.
without breast cancer (R1 P 5 0.04, R2 P 5 0.02 without any data point (Table 5). There was also no significant dif- T5
breast cancer and R1 P 5 0.06 and R2 P 5 0.1 with breast ference between the percentage of DCIS patients and inva-
cancer). There was a significant increase in the number of sive cancer patients with BPE >minimal at any data point,
patients with BPE >minimal and BPE >mild at time 2 as or BPE >mild at any data point, apart from reader 1 at
compared with time 1 [BPE >minimal R1: 90 vs. 41 in time 1, which showed more DCIS than invasive cancer
patients with breast cancer (P < 0.001) and 81 vs. 36 in patients with BPE >mild (P 5 0.01) (Table 5).
patients without breast cancer (P < 0.001) and R2: 84 vs.
52 in patients with breast cancer (P < 0.001) and 79 vs. 43 Follow-up in Patients Without Breast Cancer
in patients without breast cancer(P < 0.001), and BPE > Follow-up imaging and/or biopsy results were available in
mild R1: 59 vs. 10 in patients with breast cancer (P < 92 of the 116 patients without a known breast cancer at
0.001) and 47 vs. 13 in patients without breast cancer (P < 12–93.0 months (mean 63.6 months) after the MRI
0.001); R2: 49 vs. 12 in patients with breast cancer (P < included in this study. Subsequent cancers were diagnosed
0.001) and 41 vs. 18 in patients without breast cancer (P < in nine (10%) of these 92 patients, including DCIS in four,
F2 0.01)] for both patient groups and readers (Table 4, Figs. (1 ILC in three and IDC in two. Follow-up duration among
and 2)). Compared with time 2, at time 3 the increase in these patients was 12.0–81.0 months (mean 42.1 months).
the number of patients with BPE >minimal and BPE The odds ratio for the development of breast cancer
>mild ranged from 0–3 patients (0–2.6%) and not was for BPE >mild was significant for R2 at times 2 and 3 (7.7
statistically significant for either patient group or reader 1.5, 39.5 P 5 0.01 and 6.9 1.3, 35.3, P 5 0.03), while the
(Table 4, Fig. 1). odds ratio (OR) for BPE >minimal was significant only for
R2 and time 1 (OR 7.3 1.4, 27.3, P 5 0.02) (Table 6). T6
Breast Cancer Subgroup
Subgroup analysis of the 116 breast cancer patients did not Interreader Agreement
show significant differences in mean BPE between 35 In both subsets of patients, interreader agreement was sub-
patients with DCIS and 81 patients with invasive cancers at stantial at all three timepoints (K 5 0.79, 0.72, and 0.74 in
January 2017 79

80
TABLE 5. Comparison of BPE in Patients With Invasive Breast Cancer and DCIS at 3 Timepoints
Mean BPE (std) BPE > minimal BPE > mild

a
Reader Time Invasive DCIS P value Invasive DCIS P value Invasive DCIS P value
1 1 1.4 (0.6) 1.7 (0.9) 0.05 30.9% (25/81) 45.7% (16/35) 0.14 3.7% (3/81) 20.0% (7/35) 0.01
1 2 2.3 (0.9) 2.5 (1.1) 0.36 79.0% (64/81) 74.3% (26/35) 0.63 48.1% (39/81) 57.1% (20/35) 0.42
1 3 2.4 (0.9) 2.5 (1.1) 0.32 81.5% (66/81) 77.1% (27/35) 0.62 48.1% (39/81) 57.1% (20/35) 0.42
2 1 1.5 (0.6) 1.7 (0.8) 0.11 40.7% (33/81) 54.3% (19/35) 0.22 7.4% (6/81) 17.1% (6/35) 0.18
2 2 2.2 (0.9) 2.4 (1.13) 0.38 75.3% (61/81) 65.7% (23/35) 0.37 37.0% (30/81) 54.3% (19/35) 0.10
2 3 2.2 (0.9) 2.4 (1.1) 0.30 75.3% (61/81) 68.6% (24/35) 0.5 39.5% (32/81) 54.3% (19/35) 0.16
a
TABLE 6. Comparison of BPE in Control Patients Who Did and Did Not Develop a Breast Cancer With Odds Ratios for the Development of Breast Cancer as
Conferred by BPE > Minimal and BPE > Mild at 3 Timepoints
BPE > minimal BPE > mild

a
Reader Time Cancer No Cancer OR 95% CI P Cancer No Cancer OR 95% CI P
1 1 44.4% (4/9) 30.1% (25/83) 1.9 0.5–7.5 0.46 11.1% (1/9) 10.8% (9/83) 1.0 0.1–9.2 1.00
1 2 77.8% (7/9) 68.7% (57/83) 1.6 0.3–8.2 0.72 44.4% (4/9) 38.6% (32/83) 1.3 0.3–5.1 0.73
1 3 77.8% (7/9) 69.9% (58/83) 1.5 0.3–7.8 1.00 66.7% (6/9) 39.8% (33/83) 3.0 0.7–13.0 0.16
J_ID: JMRI Customer A_ID: JMRI25338 Cadmus Art: JMRI25338 Ed. Ref. No.: 16-0391.R1 Date: 2-December-16
2 1 77.8% (7/9) 32.5% (27/83) 7.3 1.4–37.3 0.01 33.3% (3/9) 13.3% (11/83) 3.3 0.7–15.0 0.14
2 2 77.8% (7/9) 68.7% (57/83) 1.6 0.3–8.2 0.72 77.8% (7/9) 31.3% (26/83) 7.7 1.5–39.5 <0.01
Stage:
2 3 77.8% (7/9) 69.9% (58/83) 1.5 0.3–7.8 1.00 77.8% (7/9) 33.7% (28/83) 6.9 1.3–35.3 0.02
a

Page: 80
Volume 45, No. 1

breast cancer patients at times 1, 2, and 3, and K 5 0.82, Differences in BPE between patients with invasive and
0.75, and 0.75 in control patients at times 1, 2, and 3). in-situ cancers have, to our knowledge, not been reported.
Interreader agreement for FGT was substantial in both sub- Here, apart from reader 1 at time 1, who showed more
sets of patients (K 5 0.78 each). DCIS than invasive cancer patients with BPE >mild
(P 5 0.01), there were no significant differences in BPE
Discussion between subgroups. Because greater BPE was not seen in
In this study we show that BPE changes in most but, the DCIS subgroup at any other timepoint by either reader
importantly not in all patients between the first and second and because of the small number of patients at this data
postcontrast image acquisitions. On the first postcontrast point, the authors feel this significant result is spurious.
image set, more than half of patients in both the breast can- Specifically regarding risk, higher BPE has been associ-
cer and without breast cancer groups had minimal BPE, but ated with breast cancer risk in three recent studies.14–16
on the second and third postcontrast image sets, the distri- King et al14 reported an increased likelihood of the presence
bution of BPE was more varied, with minimal, mild, and of a breast cancer in patients with BPE >mild (OR R1 3.3,
moderate BPE accounting for around 30% of patients each, 95% confidence interval [CI] 1.3–8.3 and R2 10.1, 95%
whether or not they had a cancer, and marked BPE CI 2.9–35.3) in 39 screening patients, and Telegrafo et al15
accounting for 9–17% of patients. This more varied distri- reported that 73–78% of patients with breast cancer and 0–
bution of BPE after the first postcontrast image set suggests 4% of patients without breast cancer had BPE >mild in a
that differences in BPE between patients/patient groups may mixed population. In contrast, the comparison portion of
be better demonstrated later in the MRI exam. our study showed no significant difference in mean BPE or
These results point in a different direction than the in the distribution of BPE between breast cancer and con-
current American College of Radiology MR BI-RADS trol patients. At times 2 and 3, the percentages of breast
guidelines,4 which recommend that BPE be assessed on cancer patients with BPE >minimal and BPE >mild were
early postcontrast imaging (90 sec). These guidelines focus higher than those without, suggesting that with a larger
on the first postcontrast image set, when the enhancement sample size statistical significance may have been reached.
difference between FGT and invasive cancers is greatest, However, King et al14 showed statistical significance with
because of concern that BPE could affect the diagnostic per- fewer than half of the patients in this study, which may sug-
formance of MRI. To that end, Hambly et al2 and Demar- gest different implications for BPE in a high-risk screening
tini et al3 showed similar sensitivities in low and high BPE population as compared with patients who may not be at
groups, while two additional studies, which may be limited, high risk, but who do develop a breast cancer. The high
as MRI occurred after lesion biopsy, suggest the sensitivity percentages of cancer patients with BPE >mild in the study
for known breast cancers is decreased when BPE is by Telegrafo et al15 may be in part explained by its later
>mild.20,21 Whether BPE affects sensitivity is not definitely assessment of BPE; however, the 0–4% of patients without
clear, but, as almost all cancers are seen on the first postcon- breast cancer who had BPE >mild at 3 minutes is fewer
trast image set,22–25 BPE’s effect on exam interpretation is than the 17%, 19%, and 11–16% reported by Dontchos
likely best assessed on the first postcontrast image set. et al,16, King et al,14 and this study, respectively, on the ear-
However, some recent literature suggests additional roles lier first postcontrast image set. Patient populations and
for BPE as an imaging biomarker for breast cancer risk,14–16 reader variability may therefore explain the differences.
prognosis,26,27 and predicting response to chemotherapy.28 The follow–up portion of our study was structured
While some of these studies used early postcontrast images like the work by Dontchos et al,16 which reported increased
for BPE assessment,14,16,27 others assessed BPE or mean BPE risk for breast cancer in patients with BPE >minimal (R
at 3 minutes or later.15,26,28 Since we show BPE changes in 9.0, 95% CI 1.1–71) on the initial screening MRI in 23
most patients between the first and second postcontrast image patients who developed breast cancer (in six of these
sets and stabilizes thereafter, standardizing when to assess patients the breast cancer was found on the initial screening
BPE is important, and assessment of BPE after the postcon- MRI). Our ORs for the development of breast cancer, also
trast dataset may be more reliable when assessing risk, prog- based on the initial MRI, were similar to this study and to
nosis, or posttherapy risk reduction. those published by King et al,14 but significant at time 1
In that vein, our correlations between BPE and age for BPE >minimal (7.3, 95% CI 1.4–37.3) and at times 2
and BPE and FGT, while significant for all timepoints, were and 3 for BPE >mild (7.7 95% CI 1.5–39.5 and 6.9, CI
stronger and more significant at times 2 and 3 as compared 1.3–35.3, respectively) only for reader 2, perhaps due to the
with time 1. These correlations are in agreement with other smaller number of cancers included (n 5 9).
studies that noted decreased BPE among postmenopausal This study has several limitations. The first limitation
women,2,5 and found a positive correlation between BPE is the retrospective nature; specifically, menopausal status
and FGT.5 and mammographic density at the time of MRI were not
January 2017 81

available for all cases, and follow-up was not available for all enhanced breast MR imaging: a longitudinal intraindividual cohort
study. Radiology 2014;271:45–55.
patients without breast cancer. Our study included patients
9. King V, Goldfarb SB, Brooks JD, et al. Effect of aromatase inhibitors
undergoing examination for extent of disease; these patients on background parenchymal enhancement and amount of fibrogland-
are not always scanned during week 2 and, in this popula- ular tissue at breast MR imaging. Radiology 2012;264:670–678.
tion, 50% of our premenopausal patients were scanned 10. Mousa NA, Elada R, Crystal P, Nayot D, Casper RF. The effect of
during week 2, as compared with 93% of our premeno- acute aromatase inhibition on breast parenchymal enhancement in
magnetic resonance imaging: a prospective pilot clinical trial. Meno-
pausal control patients. This discrepancy could bias the pause 2012;19:420–425.
breast cancer group toward higher BPE,1 although in our 11. DeLeo MJ 3rd, Domchek SM, Kontos D, Conant E, Chen J, Weinstein
case does not contribute to a significant difference between S. Breast MRI fibroglandular volume and parenchymal enhancement
in BRCA1 and BRCA2 mutation carriers before and immediately after
the groups. Another potential limitation is that all three risk-reducing salpingo-oophorectomy. AJR Am J Roentgenol 2015;
postcontrast timepoints were read for BPE during a single 204:669–673.
readout. This was intended to simulate clinical interpreta- 12. Price ER, Brooks JD, Watson EJ, Brennan SB, Comen EA, Morris EA.
tion of BPE and facilitate assessment of change in BPE over The impact of bilateral salpingo-oophorectomy on breast MRI back-
ground parenchymal enhancement and fibroglandular tissue. Eur
time. However, in borderline cases where two consecutive Radiol 2014;24:162–168.
timepoints may have spanned the spectrum of the same 13. Chen JH, Yu H, Lin M, Mehta RS, Su MY. Background parenchymal
BPE category, this approach may have favored placing the enhancement in the contralateral normal breast of patients under-
going neoadjuvant chemotherapy measured by DCE-MRI. Magn
sequences into separate BPE categories. In addition, there Reson Imaging 2013;31:1465–1471.
were a small number of cancers in the follow-up population.
14. King V, Brooks JD, Bernstein JL, Reiner AS, Pike MC, Morris EA. Back-
Finally, the results reflect the qualitative interpretations of ground parenchymal enhancement at breast MR imaging and breast
two readers at a single institution using a single breast MRI cancer risk. Radiology 2011;260:50–60.
protocol, and additional studies are needed to assess 15. Telegrafo M, Rella L, Stabile Ianora AA, Angelelli G, Moschetta M.
Breast MRI background parenchymal enhancement (BPE) correlates
generalizability. with the risk of breast cancer. Magn Reson Imaging 2016;34:173–176.
In conclusion, BPE changes between the first and sec- 16. Dontchos BN, Rahbar H, Partridge SC, et al. Are qualitative assess-
ond postcontrast scans and stabilizes thereafter in most ments of background parenchymal enhancement, amount of fibro-
patients. If BPE is going to be considered in risk stratifica- glandular tissue on MR images, and mammographic density
associated with breast cancer risk? Radiology 2015;276:371–380.
tion, prognosis, and in predicting response to therapy, fur-
17. D’Orsi CJ, Sickles EA, Mendelson EB, et al. ACR BIRADSVR Atlas,
ther investigation into a standardized interpretation Breast Imaging Reporting and Data System. Reston, VA: American
timepoint, perhaps after the first postcontrast image set, and College of Radiology; 2013.
perhaps separate from the BPE assessment for exam per- 18. Giess CS, Yeh ED, Raza S, Birdwell RL. Background parenchymal
enhancement at breast MR imaging: normal patterns, diagnostic chal-
formance, is necessary.
lenges, and potential for false-positive and false-negative interpreta-
tion. Radiographics 2014;34:234–247.
References 19. https://www.nia.nih.gov/health/publication/menopause U.S. Depart-
1. Kuhl CK, Bieling HB, Gieseke J, et al. Healthy pre-menopausal breast ment of Health and Human Services, National Institute on Aging. Last
parenchyma in dynamic contrast-enhanced MR imaging of the breast: updated 7/20/2015. Accessed 5/20/2016.
normal contrast medium enhancement and cyclical-phase depend-
20. Telegrafo M, Rella L, Stabile Ianora AA, Angelelli G, Moschetta M.
ency. Radiology 1997;203:137–144.
Effect of background parenchymal enhancement on breast cancer
2. Hambly NM, Liberman L, Dershaw DD, Brennan S, Morris EA. Back- detection with magnetic resonance imaging. Diagn Interv Imaging
ground parenchymal enhancement on baseline screening breast MRI: 2016;97:315–320.
impact on biopsy rate and short-interval follow-up. AJR 2011;196:
21. Uematsu T, Kasami M, Watanabe J. Does the degree of background
218–224.
enhancement in breast MRI affect the detection and staging of breast
3. DeMartini WB, Liu F, Peacock S, Eby PR, Gutierrez RL, Lehman CD. cancer? Eur Radiol 2011;21:2261–2267.
Background parenchymal enhancement on breast MRI: impact on
22. Kuhl CK, Schrading S, Strobel K, Schild HH, Hilgers RD, Bieling HB.
diagnostic performance. AJR Am J Roentgenol, 2012;198:W373–380.
Abbreviated breast magnetic resonance imaging (MRI): first postcon-
4. Morris EA, Comstock CE, Lee CH, et al. ACR BI-RADSVR Magnetic Res- trast subtracted images and maximum-intensity projection—a novel
onance Imaging. In: ACR BI-RADSV R Atlas, Breast Imaging Reporting approach to breast cancer screening with MRI. J Clin Oncol 2014;32:
and Data System. Reston, VA: American College of Radiology; 2013. 2304–2310.
5. King V, Gu Y, Kaplan JB, Brooks JD, Pike MC, Morris EA. Impact of 23. Mango VL, Morris EA, David Dershaw D, et al. Abbreviated protocol
menopausal status on background parenchymal enhancement and for breast MRI: are multiple sequences needed for cancer detection?
fibroglandular tissue on breast MRI. Eur Radiol, 2012;22:2641–2637. Eur J Radiol 2015;84:65–70.
6. Amarosa AR, McKellop J, Klautau Leite AP, et al. Evaluation of the 24. Grimm LJ, Soo MS, Yoon S, Kim C, Ghate SV, Johnson KS. Abbrevi-
kinetic properties of background parenchymal enhancement through- ated screening protocol for breast MRI: a feasibility study. Acad
out the phases of the menstrual cycle. Radiology 2013;268:356–365. Radiol 2015;22:1157–1162.
7. King V, Kaplan J, Pike MC, et al. Impact of tamoxifen on amount of 25. Heacock L, Melsaether AN, Heller SL et al. Evaluation of a known
fibroglandular tissue, background parenchymal enhancement, and breast cancer using an abbreviated breast MRI protocol: Correlation
cysts on breast magnetic resonance imaging. Breast J 2012;18:527– of imaging characteristics and pathology with lesion detection and
534. conspicuity. Eur J Radiol 2016;85:815–823.
8. Schrading S, Schild H, K€
uhr M, Kuhl C. Effects of tamoxifen and aro- 26. van der Velden BH, Dmitriev I, Loo CE, Pijnappel RM, Gilhuijs KG.
matase inhibitors on breast tissue enhancement in dynamic contrast- Association between parenchymal enhancement of the contralateral
82 Volume 45, No. 1

breast in dynamic contrast-enhanced MR imaging and outcome of patients treated with neoadjuvant chemotherapy. Medicine 2016;
patients with unilateral invasive breast cancer. Radiology 2015;276: 95:e3000.
675–685.
28. Chen JH, Yu HJ, Hsu C, Mehta RS, Carpenter PM, Su MY. Back-
27. Choi JS, Ko ES, Ko EY, Han BK, Nam SJ. Background parenchy- ground parenchymal enhancement of the contralateral normal breast:
mal enhancement on preoperative magnetic resonance association with tumor response in breast cancer patients receiving
imaging: association with recurrence-free survival in breast cancer neoadjuvant chemotherapy. Transl Oncol 2015;8:204–209.
January 2017 83

ORIGINAL RESEARCH
Stimulated Echo Diffusion Tensor Imaging

(STEAM-DTI) with Varying Diffusion Times
as a Probe of Breast Tissue
Jose R. Teruel PhD,1,2* Gene Y. Cho PhD,3,4 Melanie Moccaldi RT,5
Pål E. Goa, PhD,6 Tone F. Bathen PhD,1 Thorsten Feiweier PhD,7
Sungheon G. Kim PhD,3,4 Linda Moy MD,3,4 and Eric E. Sigmund PhD3,4
Purpose: To explore the application of diffusion tensor imaging (DTI) for breast tissue and breast pathologies using a
stimulated-echo acquisition mode (STEAM) with variable diffusion times.
Materials and Methods: In this Health Insurance Portability and Accountability Act-compliant study, approved by the
local institutional review board, eight patients and six healthy volunteers underwent an MRI examination at 3 Tesla
including STEAM-DTI with several diffusion times ranging from 68.5 to 902.5 ms. A DTI model was fitted to the data for
each diffusion time, and parametric maps of mean diffusivity, fractional anisotropy, axial diffusivity, and radial diffusivity
were computed for healthy fibroglandular tissue (FGT) and lesions. The median value of radial diffusivity for FGT was fit-
ted to a linear decay to obtain an estimation of the surface-to-volume ratio, from which the radial diameter was
calculated.
Results: For healthy FGT, radial diffusivity presented a linear decay with the square root of the diffusion time resulting
in a range of estimated radial diameters from 202 to 496 mm, while axial diffusivity presented a nearly time-independent
diffusion. Residual fat signal was reduced at longer diffusion times due to the shorter T1 of fat. Residual fat signal to
the overall signal in the healthy volunteers’ FGT was found to range from 2.39% to 2.55% (shortest mixing time), and
from 0.40% to 0.51% (longest mixing time) for the b500 images.
Conclusion: The use of variable diffusion times may provide an in vivo noninvasive tool to probe diffusion lengths in
breast tissue and breast pathology, and might aid by improving fat suppression at longer diffusion times.
D iffusion-weighted MR imaging (DWI) is an imaging

tool that provides microstructural information from
tissue in vivo, noninvasively, and without the need of con-
Of particular interest with relation to this work is the diffu-
sion tensor imaging (DTI) model, that it is aimed to pro-
vide additional microstructural information linked to
trast agents.1 In the breast, DWI has been studied for differ- diffusion directionality to characterize breast tissue, that
ent applications such as tissue characterization, might be used to increase differentiation between malignant
differentiation of lesions, and response to treatment.2–7 Fur- and benign lesions, and healthy fibroglandular tissue
thermore, different DWI sequences can provide a variety of (FGT).8–11
information, including standard isotropic diffusion using the The b-value defines the diffusion weighting, and it is
apparent diffusion coefficient (ADC),2–7 tensor models,8–11 described as b5c2 G 2 d2 tD with tD 5ðD2 d3Þ for the Stejskal-
perfusion information based on the intravoxel incoherent Tanner pulse sequence,17 where tD is the diffusion time, G
motion effect in the microvasculature,12–15 and kurtosis.16 is the diffusion gradient amplitude, D is the time between
Received Apr 11, 2016, Accepted for publication Jun 21, 2016.
*Address reprint requests to: J.R.T., Department of Radiology, University of California San Diego (UCSD), 8950 Villa La Jolla Drive (Suite-C101). 92037 La
Jolla, CA. E-mail: jteruelantolin@ucsd.edu
From the 1Department of Circulation and Medical Imaging, Norwegian University of Science and Technology (NTNU), Trondheim, Norway; 2Department of
Radiology, University of California San Diego (UCSD), La Jolla, California, USA; 3Center for Advanced Imaging Innovation and Research (CAI2R), New York
University School of Medicine, New York, New York, USA; 4Bernard and Irene Schwartz Center for Biomedical Imaging, Department of Radiology, New
York University School of Medicine, New York, New York, USA; 5Cancer Institute, New York University School of Medicine, New York, New York, USA;
6
Department of Physics, Norwegian University of Science and Technology (NTNU), Trondheim, Norway; and 7Siemens Healthcare, Erlangen, Germany
84 V

Teruel et al.: STEAM-DTI Applied to Breast Tissue
diffusion gradients, and d is the diffusion gradient duration. directions (TE/repetition time [TR]: 45/11500 ms; in-plane resolu-
For breast studies with a conventional spin echo diffusion tion: 2.1 3 2.1 mm2; image matrix: 156 3 192; slice thickness;
pulse sequence, the short T2 values of breast tissue18 imposes 5 mm; 10 slices; number of excitations [NEX]: 3; GRAPPA parallel
the use of short diffusion times to avoid a complete loss of sig- imaging factor 2), with SPAIR fat suppression, axial orientation
nal due to transverse magnetization decay. The diffusion length and anterior–posterior (AP) phase encoding. This acquisition was
repeated six times with identical parameters but varying the TM
and, therefore, the microstructural length scale to which a
(TM 5 46, 80, 150, 300, 600, 880 ms), leading to tD 5 68.5,
DWI protocol is sensitive, is proportional to the square root
102.5, 172.5, 322.5, 622.5, and 902.5 ms. One additional b 5 0
of the diffusion time.19 Consequently, the diffusion time is the
s/mm2 image was acquired just before each acquisition with
key parameter to explore different tissue length scales. reversed phase encoding posterior–anterior (PA). In addition, in-
Stimulated-echo acquisition mode (STEAM) allows phase and out-of-phase T1 VIBE series with and without SPAIR
exploring different diffusion times beyond the T2 value of fat suppression in axial orientation were acquired (TR 5 6.45 ms;
the tissue of interest while maintaining a constant echo time TE 5 2.46 ms [in-phase], 3.69 ms [out-of-phase]; in-plane resolu-
(TE) by storing the transverse magnetization in the longitu- tion: 1.25 3 1.25 mm2; image matrix: 260 3 320; slice thick-
dinal plane and recovering it after an additional time, ness 5 2.5 mm; number of slices: 88) for the last four cases. The
known as mixing time (TM).20,21 During TM, the signal full protocol was acquired in 40 min. An identical STEAM-DTI
decay will be governed by a much longer T1 relaxation; protocol was acquired for an isotropic cylindrical water phantom.
therefore, STEAM can be used to achieve longer diffusion Patients were scanned using a 3T MAGNETOM Trio, a
Tim System (Siemens Healthcare, Erlangen, Germany) with a
times that are not feasible with conventional spin echo pulse
seven-channel bilateral breast coil. Due to the time constraints in
sequences that are limited by T2 relaxation.
the clinical environment, patients were scanned with an abbreviated
The time-dependent diffusion mechanism is a complex version of this protocol that included STEAM-DTI with two b-
phenomenon dependent on the microstructural environment values (0, 500 s/mm2) in six directions (TE/TR: 52/3500 ms [52/
(presence of restricting structures, permeability of barriers, 4500 ms for one case, and 74/3500 for another case]; in-plane res-
and length scale of barriers compared with diffusion length olution: 2.1 3 2.1 mm2; image matrix: 156 3 192; slice thickness;
range).22–26 In this study focusing on mammary tissue, we 5 mm; 3 slices; NEX: 3; GRAPPA parallel imaging factor 2) with
have, in accordance with our observations and the expected SPAIR fat suppression, axial orientation, and AP phase encoding.
large confinement scale (several hundred microns) of mam- This acquisition was repeated four times with identical parameters
mary ducts, adopted the Mitra short time limit.22 In the but varying the TM 5 43, 147, 440, 877 ms, leading to tD 5 69,
short time limit, an increase of diffusion time would result in 173, 466, and 903 ms (tD 5 80, 184, 477, and 914 ms for the
case with TE 5 74 ms). One additional b 5 0 s/mm2 image was
a linear decrease of the diffusivity versus diffusion length, as
acquired just before each complete acquisition with reversed phase
more water molecules approach the structural barriers. Time-
encoding (PA). In addition, in-phase and out-of-phase T1 VIBE
dependent diffusion MRI experiments have already provided series were acquired with and without fat suppression (TR 5 5.81
promising results for other body applications. For instance, ms; TE 5 2.45 ms [in-phase], 3.68 ms [out-of-phase]; in-plane res-
the use of time-dependent diffusion has proven its ability to olution: 2.1 3 2.1 mm2; image matrix: 156 3 192; slice thick-
estimate transverse muscle fiber diameter and has shown ness 5 2.5 mm; number of slices: 28). The abbreviated protocol
promising results in detecting smaller changes in fiber diame- acquisition time was less than 10 min for all patients.
ters after exercise in patients with chronic exertional compart-
ment syndrome compared with healthy controls and in Image Preprocessing, Segmentation, and
Signal-to-Noise Ratio Calculation
modulating diffusion kurtosis in skeletal muscle.27–29
Nonaffine eddy current correction was performed automatically on
The purpose of this work was to study the feasibility
the scanner by a prototype processing included by the vendor.
of time-dependent diffusion in breast tissue using a Before tissue segmentation, the DWI data were preprocessed as fol-
STEAM-DTI approach with variable diffusion times. lows. The three averages were rigidly registered in two-dimensional
(2D) using a cross-correlation metric in Matlab (The MathWorks
Materials and Methods Inc, Natick, MA) to improve alignment of corresponding diffusion
In this Health Insurance Portability and Accountability Act directions before averaging them. Afterward, the acquired b0
(HIPPA)-compliant study, approved by the local institutional images with inverse phase-encoding polarity were used for geomet-
review board, eight patients (females, age range: 37–67), and six ric distortion correction of the inhomogeneous static magnetic field
healthy volunteers (females, age range: 25–34 years) were included using the algorithm presented in Holland et al,30, and proved for
and provided written informed consent to undergo an MRI exami- breast applications in Teruel et al.31 Subsequently, the multiple
nation including STEAM-DTI with several diffusion times. STEAM-DTI series acquired for each subject were rigidly regis-
Healthy volunteers were scanned using a 3 Tesla (T) Biog- tered to the shortest diffusion time series using Matlab.
raph mMR (Siemens Healthcare, Erlangen, Germany) with a four- After these steps, healthy fibroglandular tissue (FGT) or
channel bilateral breast coil. The protocol collected a prototype lesions were segmented. Lesions were manually segmented using
STEAM-DTI sequence with two b-values (0, 500 s/mm2) in six b500 images by covering the full extent of the lesion included in
January 2017 85

the field of view (three slices maximum). For FGT, two slices with Tensor eigenvalues ðk1 ; k2 ; k3 Þ were obtained by matrix diag-
the highest FGT content were selected from the b0 images. The onalization. Parametric maps of axial diffusivity ðADÞ½AD5k1 ,
FGT segmentation was thresholded to avoid low signal areas. The radial diffusivity ðRDÞ½RD5ðk2 1k3 Þ=2 and fractional anisotropy
ffi
qffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
signal-to-noise ratio (SNR) of the final segmentation was calculated 1 ðk1 2k2 Þp
2
1ðk2 2k3 Þ2 1ðk3 2k1 Þ2
ðFAÞ FA5 2 ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi were obtained voxel-
using the three acquired averages as follows. Three subtraction 2 2
k1 1k2 1k3 2
images were obtained by subtracting the three averages of each wise at each diffusion time. The same DTI fitting was carried out
image pairwise. The average of the standard deviation of the signal for the isotropic water phantom to validate the fitting approach
of these three subtraction images within the ROI was used as the and values were sampled in a central ROI.
noise, while the signal was calculated as the mean of the averaged
image. This approach is an extension to three-averages acquisition Surface-to-Volume Ratio and Diameter Estimation
from the difference approach presented in Dietrich et al.31 Considering the range of diffusion times explored in this work, the
Parallel to this procedure, in-phase and out-of-phase VIBE median value of the RD for each case was fitted to a linear decay
images with SPAIR fat suppression were used to correct for the to obtain an estimation of the surface-to-volume ratio (SVR), from
influence of unsuppressed fat signal in the diffusion measurements. which the diameter (1) can be calculated.
Despite using SPAIR fat suppression for diffusion, it is generally pffiffiffiffiffiffiffiffiffiffi

impossible to obtain completely fat-free breast images, as adipose 4 RD0 t 4
RDðtÞ5RD0 12SVR pffiffiffi ; 15 (3)
tissue is the main component of breast tissues for most cases. Fur- 6 p SVR
thermore, the influence of the unsuppressed fat will vary with the
with RD0 being the zero intercept of the linear fitting of the radial
diffusion time because of the shorter T1 (367 ms) of fat at 3T.18
diffusivity with the square root of the diffusion time.
First, water and “residual fat” images were obtained by direct addi-
tion and subtraction (divided by 2) of the in- and out-of-phase fat-
T1 Mapping
S0f
suppressed images. Second, a fat-over-water image S0w was As an additional output from the acquired data, the signal decay
obtained dividing the residual fat image by the water image voxel- with TM without diffusion weighting was fitted to obtain T1 maps
wise. In a third step, the evolution of the residual fat-over-water of breast tissue. Specifically, the signal intensity of the virtual b0
signal with TM was obtained using T1 values of water (T1w) and images at different TMs was fitted to a free monoexponential decay
fat (T1f ) from Rakow-Penner et al18: voxel-wise to estimate T1 values for each ROI. T1 fitting was per-
formed after image preprocessing including fat signal correction.
STMf S0f 2TM ðT 11f 2T 11w Þ
5 e (1)
STMw S0w Results
From the eight patients included for the clinical protocol,
The fat-over-water images were registered to the DWI image stack
five patients presented a lesion large enough for clear visuali-
with the same procedure given above. Finally, the calculated pro-
portion of fat signal at each TM was subtracted for the diffusion zation in the diffusion images (>1 cm) covered by the field
images voxel-wise. The same absolute contribution from residual of view of the clinical protocol. The five lesions corresponded
fat was subtracted from b0 and b500 images assuming a null diffu- to three malignant lesions (two invasive ductal carcinomas
sion coefficient in adipose tissue.33 and one mucinous invasive ductal carcinoma), and two
benign lesions (one fibroadenoma and one benign cyst).
DTI Lesion sizes on MR were 11 mm and 85 mm for invasive
Using the segmented ROIs, a voxel-wise DTI fitting was per-
ductal carcinomas, 63 mm for mucinous invasive ductal carci-
formed. Due to the nature of the STEAM sequence, imaging and
noma, 14 mm for fibroadenoma, and 25 mm for the benign
spoiler gradients introduced an additional diffusion weighting, in
cyst. All lesions were histologically verified with the exception
particular at long diffusion times. For this reason, the DTI fitting
was performed using the complete b-matrix calculated by the ven-
of the cyst that was categorized by MR imaging features (BI-
dor including all gradients’ contributions. Additionally, the b0 RADS 2) and previous ultrasound and mammogram. For the
image was also included as a diffusion-weighted image in the fit- other three patients, the lesion was not covered by the field
ting, as the b-matrix elements were non-negligible even in the of view of the clinical protocol; however, they presented dense
absence of diffusion gradients. The fitting was performed by breasts, and therefore, were included for FGT analysis.
" ! #
X Fat Signal Decay With TM
S5exp 2 bij Dij 1C with bij 5bji and Dij 5Dji The mean contribution of the residual fat signal to the over-
i;j5x;y;z
all signal intensity for each ROI in the healthy volunteers’
(2) FGT was found to range from 2.39% to 2.55% for the
Therefore, the deterministic fitting with the seven inputs (b0 image b500 images with the shortest TM, and from 0.40% to
and six b500 directions) provided seven outputs; the six unique 0.51% for the b500 images with the longest TM. Addition-
tensor coefficients and the virtual b0 image ½S0 5e C , C being the ally, in Figure 1 it can be observed how residual fat signal is F1
constant fitted in the model above. evident at the shortest diffusion time, particularly in the
86 Volume 45, No. 1

C
O
L
O
R
FIGURE 1: Diffusion images of a healthy volunteer (age: 32 years old) showing the decrease of unsuppressed fat signal with the
longest TM. Red arrows point to residual fat signal in the b500 image at the shortest diffusion time.
FIGURE 2: Cancer patient with invasive ductal carcinoma in the left breast (age: 37 years old). A: Diffusion image with b500 and
TM 5 43 ms. B: Diffusion image with b500 and TM 5 877 ms. Lower limit in signal intensity is kept equal to 0 in both images to
illustrate how the unsuppressed fat signal has completely decayed at the longer TM. Furthermore, residual ghosting from adipose
(short T1), is also removed in the longest TM image.
b500 image, and how this residual fat signal disappears at Water Phantom
the longest diffusion time adding conspicuity to FGT and The results for all parameters under study for the 6 acquired
reducing the fat signal contamination within FGT. Likewise, diffusion times are reported in Table 1 using an in-plane T1
F2 in Figure 2 it can be observed how lesion conspicuity is circular ROI of 2.25 cm2 on an isotropic water phantom.
increased at longer diffusion time when fat signal has The T1 relaxation time median (interquartile range [IQR])
completely decayed. for the selected ROI was 2399 (490) ms. The SNR
TABLE 1. Results of STEAM-DTI for an Isotropic Water Phantom Using Variable Diffusion Times*
tD\Parameter MD AD RD FA
tD 5 68.5 2.09 [0.07] 2.26 [0.18] 1.99 [0.05] 0.08 [0.03]

tD 5 102.5 2.13 [0.07] 2.32 [0.13] 2.03 [0.06] 0.08 [0.03]
tD 5 172.5 2.09 [0.09] 2.27 [0.10] 2.00 [0.09] 0.08 [0.03]
tD 5 322.5 2.12 [0.09] 2.28 [0.11] 2.05 [0.11] 0.08 [0.03]
tD 5 622.5 2.06 [0.17] 2.27 [0.22] 1.98 [0.15] 0.10 [0.04]
tD 5 902.5 2.10 [0.23] 2.28 [0.34] 1.99 [0.19] 0.11 [0.04]
23 2
Values in median [IQR] within the ROI. tD 5 diffusion time (ms); MD 5 mean diffusivity (310 mm /s); AD 5 axial diffusivity
(31023 mm2/s); RD 5 radial diffusivity (31023 mm2/s); FA 5 fractional anisotropy (dimensionless).
January 2017 87

C
O
L
O
R
FIGURE 3: Axial and radial diffusivity parametric maps of a healthy FGT (healthy volunteer, age: 28 years old; case 6 in Table 2) at
the shortest and the longest diffusion times (tD). Parametric maps are shown by direct overlay on corresponding slices of T1-
weighted VIBE anatomical images for one slice.
measured for the b0 images ranged between 84 (shortest tD) the b500 images ranged between 9 and 15 at the shortest
and 34 (longest tD). The averaged SNR for the six b500 tD, and between 6 and 9 at the longest tD.
images ranged between 26 (shortest tD) and 15 (longest tD). For illustration, the parametric map of the radial
diameter for a healthy volunteer is shown in Figure 5, F5
including the projection of the median value of the radial
Healthy Volunteers diameter along the x and y axes.
AD and RD parametric maps at the shortest and longest
diffusion times for one healthy volunteer are shown in Patients
For patients where FGT was analyzed (n 5 3), MD, FA,
F3 Figure 3. In Figure 4, the median values of the AD and RD
AD, RD are reported in Table 3. The SNR values of the b0 T3
F4 within the ROI shown in Figure 3 are plotted against the
images for these three cases were between 15 and 22 at the
square root of the diffusion time (which is proportional to
shortest tD, and were each 10 at the longest tD. Similarly,
diffusion length), including a constant fit for AD, and a lin-
the SNR of the b500 images ranged between 8 and 11 at
ear fit for RD. The MD, AD, RD, and FA values at the
the shortest tD, and were each 6 at the longest tD.
T2 shortest and longest diffusion times are shown in Table 2
In Table 4, the results for the breast lesions analyzed T4
for the six healthy volunteers. In addition, the estimated
are presented at the shortest and longest tD. In Figure 6, the F6
radial diameters for each healthy volunteer are also reported
median AD and RD of one cyst and one malignant tumor
in Table 2. The SNR values of the b0 images for these six
(invasive ductal carcinoma) are plotted versus the square
cases were between 21 and 38 at the shortest tD, and were
root of tD. In addition to these two cases, the plot of AD
between 10 and 15 at the longest tD. Similarly, the SNR of
and RD for one healthy FGT case is presented to illustrate
the three regimes of structural restriction observed in this
study. The SNR values of the b0 images for the different
lesions (n 5 5) ranged from 13 to 20 at the shortest tD, and
from 11 to 16 at the longest tD. Similarly, the SNR of the
b500 images ranged from 9 to 14 at the shortest tD, and
from 6 to 20 at the longest tD.
T1 Mapping
The median T1 values for each FGT case under study
ranged between 1074 and 1605 ms with a mean of 1206
C
O ms. The T1 map for the full ROI of one slice of healthy
L FGT is shown in Figure 7. For the lesions analyzed, the fol- F7
O
R lowing median values (IQR) were found for each case: 1254
FIGURE 4: Evolution of axial (AD) and radial diffusivity (RD) (429) and 1156 (923) ms for the invasive ductal carcino-
with the square root of the diffusion time (tD). Straight lines mas, 747 (425) ms for the mucinous carcinoma, 1013
present constant fit for AD and linear fit for RD. Values shown
are median values from healthy FGT ROI (only the slice shown (409) ms for the fibroadenoma, and 1674 (577) ms for the
in Fig. 3). benign cyst.
88 Volume 45, No. 1

TABLE 2. Parameters Obtained from the Healthy FGT from Volunteers Using the Full Protocola
Healthy
FGT tD 5 68.5 ms tD 5 902.5 ms
MD FA AD RD MD FA AD RD Ø (lm)
Case 1b 1.54 0.17 1.78 1.42 1.44 0.22 1.81 1.28 370
[0.10] [0.51] [0.38] [0.37] [0.13] [0.42] [0.38]
[0.40]
b
Case 2 1.92 0.10 2.10 1.82 1.82 0.17 2.12 1.65 496
[0.18] [0.05] [0.19] [0.19] [0.25] [0.13] [0.38] [0.25]
Case 3 2.34 0.11 2.59 2.22 2.12 0.18 2.44 1.96 434
[0.73] [0.07] [0.85] [0.69] [0.57] [0.11] [0.60] [0.54]
Case 4 1.91 0.16 2.27 1.75 1.54 0.28 2.01 1.30 202
[0.70] [0.10] [0.94] [0.61] [0.58] [0.23] [0.59] [0.60]
Case 5 2.03 0.11 2.26 1.91 1.90 0.20 2.28 1.71 429
[0.30] [0.06] [0.32] [0.31] [0.39] [0.15] [0.50] [0.43]
Case 6 2.06 0.10 2.28 1.95 1.80 0.32 2.38 1.52 232
[0.36] [0.06] [0.33] [0.39] [0.60] [0.28] [0.51] [0.74]
a
Results reported for the shortest and the longest diffusion times. Values in median [IQR] from a two-slices ROI.
tD 5 diffusion time (ms); MD 5 mean diffusivity (31023 mm2/s); AD 5 axial diffusivity (31023 mm2/s); RD 5 radial diffusivity
(31023 mm2/s); FA 5 fractional anisotropy (dimensionless); Ø 5 diameter (lm).
b
Fat correction was not performed for these two cases because VIBE images were not acquired.
C
O
L
O
R
FIGURE 5: Radial diameter (ø) parametric map for one breast of a healthy volunteer (age: 28 years old). Distribution of the median
values of the radial diameter along with the X and Y positions independently.
January 2017 89

TABLE 3. Parameters Obtained from the Healthy FGT from Patientsa
Healthy FGT tD 5 69 ms tD 5 903 ms

MD FA AD RD MD FA AD RD Ø
Case 1 2.16 0.15 2.48 1.99 2.07 0.31 2.68 1.76 490
[0.19] [0.11] [0.27] [0.23] [0.74] [0.24] [0.11] [0.68]
Case 2 1.89 0.16 2.17 1.73 1.71 0.29 2.13 1.51 370
[0.34] [0.09] [0.42] [0.35] [0.74] [0.20] [0.86] [0.73]
Case 3 1.96 0.19 2.36 1.76 1.55 0.34 2.02 1.32 210
[0.28] [0.11] [0.40] [0.28] [0.66] [0.22] [0.69] [0.62]
a
Results reported for the shortest and the longest diffusion times. Values in median [IQR] from a two-slices ROI.
(31023 mm2/s); FA 5 fractional anisotropy (dimensionless); Ø 5 diameter [mm].
Discussion STEAM-DTI time-dependent diffusion has already proven

Our results present a preliminary estimation of the radial its utility to probe different tissue structures in the body,
diameter of breast glandular tissue in vivo and noninvasively as for instance for estimation of muscle fiber diameters and
using MRI by means of time-dependent diffusion with a their change through exercise in healthy controls and
STEAM-DTI sequence. In addition, estimation of different patients27,28 as well as for bovine skeletal muscles.34 DTI
length scale regimens in breast tissue have been attempted of the breast (as well as nondirectional DWI), is in general
including: cyst (no restriction), healthy FGT (diffusion performed with a fixed diffusion time for all b-values
length approaching restricting structures), and cancer selected. While the radial diameter of lactiferous ducts is
(restriction structures smaller than the observed diffusion highly variable throughout the breast, with reported values
length range). Furthermore, we have demonstrated addition- between 70 microns in the terminal units35 and up to
al benefits of using longer diffusion times with STEAM for around 500 microns in some areas of the nipple for the
breast applications, including complete fat suppression at nonlactating breast,36,37 it is clear how probing healthy
longer diffusion times and the possibility of performing FGT would be aided by extended diffusion times. Our
accurate T1 mapping using the T1 decay at different TMs. reported tD range between 69 and 903 ms (between
While this is among the first attempts of measuring approximately 25 and 91 microns in 2D diffusion length ½
pffiffiffiffiffiffiffiffiffiffiffi
quantitative time-dependent diffusion in the breast, LD 5 4DtD taking D 5 2.3 mm2/ms) provides a diffusion
TABLE 4. Parameters Obtained from the Five Lesions Included from the Patients Scanned with the Clinical Proto-
col at the Shortest and Longest Diffusion Timesa
Tissue type tD 5 69 ms tD 5 903 ms

MD FA AD RD MD FA AD RD
Simple cyst 1.84 0.10 1.99 1.78 2.14 0.12 2.30 2.07
[0.18] [0.07] [0.09] [0.25] [0.48] [0.06] [0.06] [0.43]
Invasive ductal carcinoma 1.11 0.16 1.31 1.00 1.07 0.27 1.36 0.93
[0.49] [0.10] [0.50] [0.47] [0.56] [0.18] [0.64] [0.58]
Fibroadenoma 1.78 0.14 2.10 1.61 1.34 0.36 1.75 1.10
[0.07] [0.12] [0.21] [0.09] [0.49] [0.38] [0.34] [0.57]
Mucinous carcinoma 2.47 0.13 2.76 2.33 2.40 0.19 2.79 2.21
[0.31] [0.07] [0.38] [0.29] [0.72] [0.14] [0.79] [0.73]
Invasive ductal carcinomab 1.03 0.19 1.21 0.95 0.93 0.27 1.28 0.79
[0.18] [0.10] [0.27] [0.15] [0.23] [0.18] [0.54] [0.21]
a
Values in median [IQR] within the ROI. Only one lesion for each patient.
(31023 mm2/s); FA 5 fractional anisotropy (dimensionless).
b
For this patient, short tD 5 80 ms and long tD 5 914 ms.
90 Volume 45, No. 1

C
O
L
O
R
FIGURE 6: Different length scales observed in our study. Cyst (unrestricted diffusion) (age: 47 years old); healthy FGT with ROI
overlaid in red (size of restricting structures approached by diffusion length) (age: 28 years old); invasive ductal carcinoma
(restricted diffusion) (age: 37 years old). LD: Diffusion length; tD: Diffusion time.
length range that allows sensitivity to the radial length many cases. In standard DWI pulse sequences, the signal
scale of healthy FGT. decay of fat and FGT is governed by a nearly equal T2
The values of parametric DTI maps at the shortest dif- relaxation time.18 However, using STEAM, the transverse
fusion time in our study are within the ranges previously magnetization is stored temporarily in the longitudinal plane
reported in DTI studies of the breast.8–11 As presented in during the TM. During this period, the signal decay is gov-
our results, the mean diffusivity in healthy FGT decreased erned by the T1 relaxation time, which is short in fat
with the diffusion time as expected. This decrease was large- (367 ms at 3T) compared with a much longer T1 of the
ly due to a decrease in the radial diffusivity, while axial dif- FGT (1445 ms at 3T).18 Therefore, at appropriate TMs,
fusivity remained at similar values of the shortest diffusion a decay of the fat signal can be achieved while conserving
time (less than 5% variation). A nearly constant AD with the FGT signal. This might open further possibilities for
the diffusion time and a linear decay of the radial diffusivity using STEAM-DWI as a tool for cancer detection, as the
resulted in an increase of FA with the diffusion time. This contrast of lesions and tissue clearly increases when the
increase in FA is consistent with a higher sensitivity to the residual fat contamination is not present. In particular,
microstructure at higher diffusion times. For lesions, higher conspicuity of the margins of the lesions might aid
changes in AD and RD with the diffusion time were not in the differentiation of malignant and benign lesions.
clearly linear, and the slight fluctuations observed may Our study is affected by several limitations due to its
require further models beyond the short time limit regime, exploratory nature that include known drawbacks of DWI
consistent with the smaller restriction scale in cancerous
lesions, as probed in a recent oscillating gradient preclinical
study.38 Also, it is important to notice that even for AD of
healthy tissue, there were small fluctuations with the diffu-
sion time, including unphysical increases. These small
changes, in particular when a slight increase in AD is
accompanied by a slight decrease in RD, may be due to
eigenvalue repulsion effects when moving to images with
lower SNR at higher diffusion times.39
An additional important finding in our study is the
C
complete loss of signal from adipose tissue obtained at long O
diffusion times. Independently of the fat suppression used, L
O
there is an unavoidable fat signal contamination in diffusion R
images of the breast, due to its fat content and the difficulty FIGURE 7: T1 relaxation time map for a healthy volunteer (age:
this imposes in obtaining accurate water shimming for 32 years old) overlaid on T1-weighted VIBE image.
January 2017 91

experiments,40 and additional limitations due to the use of Furthermore, as it has been explained before, the diffusion-
a STEAM sequence. Nevertheless, our processing routines weighting patterns differ for each diffusion time. In addi-
have been chosen to overcome the majority of drawbacks. tion, even though SNR was measured and controlled for, it
Rigid registration has been used in this study with the becomes lower as the TM increases. Thus, diffusion images
intention of obtaining the most accurate fixed alignment of with longer diffusion times were noisier in our study than
all corresponding images without deformation of the origi- the corresponding ones at shorter diffusion times.
nal images. Correction of eddy-current–induced distortion Another limitation concerns the recruitment process
was performed immediately upon acquisition by vendor for patients. Due to the consecutive recruitment of patients
software. Inhomogeneous static field geometric correction without regard of age or breast density, many patients pre-
was implemented for all diffusion times independently. sented only heterogeneously or scattered FGT that was not
Finally, the variable influence at different diffusion times of adequate for FGT evaluation, preventing their inclusion in
the residual fat signal was minimized by correcting for it our study. Furthermore, the low slice coverage (three slices
with separate fat-suppressed fat/water imaging data. with slice thickness 5 mm) mandated by scan time limits is
The use of STEAM involves some inherent drawbacks. problematic for full breast evaluation. Additionally, the
In particular, the diffusion weighting and diffusion direc- study is limited by the low number of volunteers and
tions are affected by the unbalanced slice selection gradients patients and the fact that volunteers were scanned using a
and spoiler gradients, and this effect accentuates with longer different scanner than the one dedicated for clinical exams.
diffusion times. We have aimed to minimize this effect by Therefore, the results presented in this work will require
using the full b-matrix computed by the vendor software further validation in a larger and more heterogeneous sam-
including all gradient contributions. Furthermore, we added ple of volunteers and patients and, optimally, the use of the
b0 (nominally unweighted images) as diffusion-weighted same platform.
images for our fitting as the b-matrices were not null. Nev- In summary, our study presents promising preliminary
ertheless, even when accounting for these effects, the effec- results concerning the use of a STEAM-DTI sequence for
tive gradient direction was found to be different for the breast applications. The use of longer diffusion times and
different diffusion times explored, and this might introduce time-dependent diffusion may provide valuable tools to
a bias in our fitting. Strategies to correct for this problem, optimize diffusion-weighted MRI examinations and to char-
such as to obtain equivalent diffusion directions at different acterize underlying breast structures with different length
diffusion times by altering the scanner gradient directions to scales. Furthermore, we have demonstrated that images
compensate for the bias introduced by the “butterfly” gra- completely free of fat signal can be obtained at longer diffu-
dients, have been reported and are under consideration for sion times when using the STEAM sequence.
further studies.41 As it was presented in the results, the
STEAM sequence was validated using an isotropic water
Acknowledgments
phantom to prove diffusion time independence in an isotro-
pic environment. However, it is important to note that Contract grant sponsor: Central Norway Regional Health
noise effects, the influence of “butterfly” gradients, and gra- Authority (RHA); Contract grant sponsor: Liaison Com-
mittee between the RHA and the Norwegian University of
dient nonlinearity due to off-isocenter position in the mag-
Science and Technology (NTNU)
net bore when using a breast coil might influence the
We thank Ms. Joon Lee for assistance in recruitment, and
observed anisotropy.
Dr. Dmitry Novikov and Mr. Jason Ostenson for valuable
Finally, in this study, we reported the estimated radial
discussions on this study.
diameter (ø) as the median value of the parametric map of
the radial diameter computed by a voxel-wise fitting of the
short time limit model. We have to bear in mind that the
radial diameters obtained in our study are also affected by References
the presence of water diffusion in FGT outside the duct 1. Le Bihan, D, Breton D. Imagerie de diffusion in-vivo par r
esonance. C
lumen, such as in lobules and fibrous stroma. It is impor- R Acad Sci (Paris) 1985;301:1109–1112.
tant to acknowledge as well that performing this fitting 2. Guo Y, Cai YQ, Cai ZL, et al. Differentiation of clinically benign and
malignant breast lesions using diffusion-weighted imaging. J Magn
voxel-wise poses challenges. Despite having carefully Reson Imaging 2002;16:172–178.
designed registration between corresponding volumes at dif-
3. Marini C, Iacconi C, Giannelli M, Cilotti A, Moretti M, Bartolozzi C.
ferent TMs, it is impossible to obtain perfect alignment Quantitative diffusion-weighted MR imaging in the differential diagno-
through all images contributing to the two-step analysis (six sis of breast lesion. Eur Radiol 2007;17:2646–2655.
or four DTI fittings are performed for each diffusion time 4. Partridge SC, Mullins CD, Kurland BF, et al. Apparent diffusion coeffi-
cient values for discriminating benign and malignant breast MRI
independently, and time-dependent diffusion fitting is sub- lesions: effects of lesion type and size. AJR Am J Roentgenol 2010;
sequently applied using the parametric maps computed). 194:1664–1673.
92 Volume 45, No. 1

5. Pinker K, Baltzer P, Bogner W, et al. Multiparametric MR imaging with 23. Mitra PP, Sen PN, Schwartz LM, et al. Diffusion propagator as a probe
high resolution dynamic contrast-enhanced and diffusion-weighted of the structure of porous media. Phys Rev Lett 1992;24:3555.
imaging at 7 T improves the assessment of breast tumors: a feasibility
24. Fieremans E, Novikov DS, Jensen JH, et al. Monte Carlo study of a
study. Radiology 2015;276:360–370.
two-compartment exchange model of diffusion. NMR Biomed 2010;
6. Park SH, Moon WK, Cho N, et al. Diffusion-weighted MR imaging: 23:711–724.
pretreatment prediction of response to neoadjuvant chemotherapy in
25. Novikov DS, Fieremans E, Jensen JH, et al. Random walk with bar-
patients with breast cancer. Radiology 2010;257:56–63.
riers. Nat Phys 2011;7:508–514.
7. McLaughlin RL, Newitt DC, Wilmes LJ, et al. High resolution in vivo
characterization of apparent diffusion coefficient at the tumor-stromal 26. Novikov DS, Jensen JH, Helpern JA, et al. Proc Natl Acad Sci U S A
boundary of breast carcinomas: a pilot study to assess treatment 2014;111:5088–5093.
response using proximity-dependent diffusion-weighted imaging. 27. Sigmund EE, Sui D, Ukpebor O, et al. Stimulated echo diffusion ten-
J Magn Reson Imaging 2014;39:1308–1313. sor imaging and SPAIR T2 -weighted imaging in chronic exertional
8. Partridge SC, Ziadloo A, Murthy R, et al. Diffusion tensor MRI: prelimi- compartment syndrome of the lower leg muscles. J Magn Reson
nary anisotropy measures and mapping of breast tumors. J Magn Imaging 2013;38:1073–1082.
Reson Imaging 2010;31:339–347. 28. Sigmund EE, Novikov DS, Sui D, et al. Time-dependent diffusion in
9. Baltzer PA, Sch€
afer A, Dietzel M, et al. Diffusion tensor magnetic reso- skeletal muscle with the random permeable barrier model (RPBM):
nance imaging of the breast: a pilot study. Eur Radiol 2011;21:1–10. application to normal controls and chronic exertional compartment
syndrome patients. NMR Biomed 2014;27:519–528.
10. Eyal E, Shapiro-Feinberg M, Furman-Haran E, et al. Parametric diffu-
sion tensor imaging of the breast. Invest Radiol 2012;47:284–291. 29. Marschar AM, Kuder TA, Stieltjes B, et al. In vivo imaging of the time-
dependent apparent diffusional kurtosis in the human calf muscle.
11. Teruel JR, Goa PE, Sjøbakk TE, et al. Diffusion weighted imaging for J Magn Reson Imaging 2015;41:1581–1590.
the differentiation of breast tumors: from apparent diffusion coeffi-
cient to high order diffusion tensor imaging. J Magn Reson Imaging 30. Holland D, Kuperman JM, Dale AM. Efficient correction of inhomoge-
2016;43:1111–1121. neous static magnetic field-induced distortion in echo planar imaging.
Neuroimage 2010;50:175–183.
12. Sigmund EE, Cho GY, Kim S, et al. Intravoxel incoherent motion
imaging of tumor microenvironment in locally advanced breast can- 31. Teruel JR, Fjøsne HE, Østlie A, et al. Inhomogeneous static magnetic
cer. Magn Reson Med 2010;65:1437–1447. field-induced distortion correction applied to diffusion weighted MRI
of the breast at 3T. Magn Reson Med 2015;74:1138–1144.
13. Cho GY, Moy L, Zhang JL, et al. Comparison of fitting methods and
b-value sampling strategies for intravoxel incoherent motion in breast 32. Dietrich O, Raya JG, Reeder SB, Reiser MF, Schoenberg SO. Mea-
cancer. Magn Reson Med 2015;74:1077–1085. surement of signal-to-noise ratios in MR images: influence of multi-
channel coils, parallel imaging, and reconstruction filters. J Magn
14. Suo S, Lin N, Wang H, et al. Intravoxel incoherent motion diffusion- Reson Imaging 2007;26:375–385.
weighted MR imaging of breast cancer at 3.0 tesla: comparison of dif-
ferent curve-fitting methods. J Magn Reson Imaging 2015;42:362–270. 33. Baron P, Dorrius MD, Kappert P, et al. Diffusion-weighted imaging of
normal fibroglandular breast tissue: influence of microperfusion and
15. Teruel JR, Goa PE, Sjøbakk TE, Østlie A, Fjøsne HE, Bathen TF. A fat suppression technique on the apparent diffusion coefficient. NMR
simplified approach to measure the effect of the microvasculature in Biomed 2010;23:399–405.
diffusion-weighted MR imaging applied to breast tumors: preliminary
results. Radiology 2016 [epub ahead of print]. doi: http://dx.doi.org/ 34. Kim S, Chi-Fishman G, Barnett AS, et al. Dependence on diffusion
10.1148/radiol.2016151630. time of apparent diffusion tensor of ex vivo calf tongue and heart.
Magn Reson Med 2005;54:1387–1396.
16. Iima M, Yano K, Kataoka M, et al. Quantitative non-Gaussian diffusion
and intravoxel incoherent motion magnetic resonance imaging: differ- 35. Mayr NA, Staples JJ, Robinson RA, et al. Morphometric studies in
entiation of malignant and benign breast lesions. Invest Radiol 2015; intraductal breast carcinoma using computerized image analysis. Can-
50:205–211. cer 1991;67:2805–2812.
17. Stejskal EO, Tanner JE. Spin diffusion measurements: spin echoes in 36. Rusby JE, Brachtel EF, Michaelson JS, et al. Breast duct anatomy in
the presence of a time-dependent field gradient. J Chem Phys 1965; the human nipple: three-dimensional patterns and clinical implica-
42:288–292. tions. Breast Cancer Res Treat 2007:106;171–179.
18. Rakow-Penner R, Daniel B, Yu H, Sawyer-Glover A, Glover GH. Relax- 37. Taneri F, Kurukahvecioglu O, Akyurek N, et al. Microanatomy of milk
ation times of breast tissue at 1.5T and 3T measured using IDEAL. ducts in the nipple. Eur Surg Res 2006;38:545–549.
38. Reynaud O, Winters KV, Hoang DM, et al. Surface-to-volume ratio
19. €
Einstein A. Uber die von der molekularkinetischen Theorie der W€
arme mapping of tumor microstructure using oscillating gradient diffusion
geforderte Bewegung von in ruhenden Fl€ ussigkeiten suspendierten weighted imaging. Magn Reson Med 2016;76:237–247.
Teilchen. Annalen der Physik 1905;322:549–560.
39. Anderson AW. Theoretical analysis of the effects of noise on diffusion
20. Merboldt KD, H€
anicke W, Frahm J. Diffusion imaging using stimulated tensor imaging. Magn Reson Med 2001;46:1174–1188.
echoes. Magn Reson Med 1991;19:233–239.
40. Le Bihan D, Poupon C, Amadon A, et al. Artifacts and pitfalls in diffu-
21. Tanner JE. Use of the stimulated echo in NMR diffusion studies. sion MRI. J Magn Reson Imaging 2006;24:478–488.
J Chem Phys 1970;52:2523.
41. Lundell H, Alexander DC, Dyrby TB. High angular resolution diffusion
22. Sen PN. Time-dependent diffusion coefficient as a probe of geome- imaging with stimulated echoes: compensation and correction in
try. Concepts Magn Reson Part A 2004;23A:1–21. experiment design and analysis. NMR Biomed 2014;27:918–925.
January 2017 93

ORIGINAL RESEARCH
Dynamic Contrast-Enhanced and

Diffusion-Weighted MRI of Estrogen
Receptor-Positive Invasive Breast Cancers:
Associations Between Quantitative MR
Parameters and Ki-67 Proliferation Status
Jong Ki Shin, MD1 and Jin You Kim, MD1,2*
Purpose: To explore the association between quantitative parameters derived from dynamic contrast-enhanced magnetic
resonance imaging (DCE-MRI) and diffusion-weighted imaging (DWI) and Ki-67 proliferation status in patients with estrogen
receptor (ER)-positive invasive breast cancer.
Materials and Methods: We retrospectively reviewed the records of 88 patients with ER-positive invasive breast cancer
who underwent preoperative DCE-MRI and DWI on a 3T scanner. Perfusion parameters (Ktrans, Kep, and Ve) and apparent
diffusion coefficients (ADCs) were recorded, and we correlated these data with the Ki-67 status. The Ki-67 proliferation
index was categorized as high (14%) or low (<14%).
Results: In the high-Ki-67 group, the mean Ktrans was significantly higher (P < 0.001) than that of the low-Ki-67 group, and
the mean ADC significantly lower (P < 0.001). However, the mean Kep and Ve values did not differ between the two groups
(P 5 0.248 and P 5 0.055, respectively). Univariate analysis showed that a higher Ktrans (>0.274), a lower ADC (0.893 3 1023
mm2/s), a larger tumor size (>2 cm), a higher histological grade (grade 3), the presence of axillary metastasis, and positive
P53 status were significantly associated with high-Ki-67 status (all P values < 0.05). Of these variables, a higher Ktrans (>0.274;
adjusted odds ratio [OR] 5 9.027, 95% confidence interval [CI] 5 1.929–42.245; P 5 0.005) and a higher histological grade
(grade 3; adjusted OR 5 7.510, 95% CI 5 1.305–43.205; P 5 0.024) independently predicted a high Ki-67 status.
Conclusion: Ktrans derived from DCE-MRI is associated independently with the Ki-67 proliferation status in patients with
ER-positive invasive breast cancer.
aggressive breast cancer.4 It has been suggested as a clinically

B reast cancer is a heterogeneous disease. Histologically
similar tumor phenotypes may exhibit clinically different
behaviors, probably attributed to molecular differences.1,2
valuable marker to distinguish luminal A and luminal B
breast cancers. Luminal breast cancers with Ki-67 levels of
Currently, an immunohistochemical (IHC) classification at least 14% were assigned to the luminal B subtype and
system based on the expression of the estrogen receptor (ER), had a worse clinical outcome compared with tumors with
progesterone receptor (PR), human epidermal growth factor Ki-67 levels of less than 14% (luminal A).3 Luminal A sub-
receptor 2 (HER2), and Ki-67 is widely used. This classifica- types were less responsive to chemotherapy, whereas luminal
tion is thought to reflect the intrinsic breast cancer subtypes, B subtypes were responsive not only to chemotherapy but
which include basal-like, HER2-enriched, and luminal A and also endocrine treatment or molecular-targeted therapy.3,5
B subtypes.3 Accordingly, the Ki-67 status could be a potential prognostic
The Ki-67 proliferation index reflects the extent of marker for determining management of patients with
proliferative activity and is a reliable identifier of more ER-positive breast cancer. Currently, although the National
*Address reprint request to: J.Y.K., Department of Radiology, Pusan National University Hospital, Pusan National University School of Medicine and Medical
Research Institute 1-10, Ami-Dong, Seo-gu, Busan, 602-739, Korea. E-mail: youdosa@naver.com
From the 1Medical Research Institute, Pusan National University School of Medicine, Busan, Republic of Korea; and 2Department of Radiology, Pusan
National University Hospital, Busan, Republic of Korea.

V

Shin and Kim: Associations Between MRI Parameters and Ki-67
Comprehensive Cancer Network (NCCN) guidelines do Of these, 104 who were histologically diagnosed with ER-positive
not recommend Ki-67-based assessment, increasing evidence tumors were included in our retrospective analysis. Of these women,
shows that the Ki-67 proliferation index is associated with seven who had undergone vacuum-assisted or excisional biopsies for
treatment efficacy and survival.6–8 A recent meta-analysis of diagnosis, five who had received neoadjuvant chemotherapy, and four
data from over 6000 patients showed that a high Ki-67 with inadequate DWI data were excluded. For one woman with bilat-
eral breast cancer, only the left-side tumor was included. For three
index is associated with a >50% risk of death and a 64%
women with multifocal or multicentric breast cancer, only the largest
greater risk of recurrence in patients with early-stage breast
tumors were included. Finally, 88 breast tumors (mean invasive size,
cancer, after adjustment for stage, histological grade, and
1.99 6 1.67 cm; range, 0.5–11.0 cm) in 88 women (mean age, 51.53
ER status. Therefore, routine assessment of the Ki-67 is years; range, 28–82 years) were included. The tumors comprised
recommended.8 invasive ductal carcinoma (n 5 81 [92.0%]), invasive lobular carci-
Dynamic contrast-enhanced magnetic resonance imag- noma (n 5 4 [4.5%]), mucinous carcinoma (n 5 2 [2.3%]), and
ing (DCE-MRI)-derived parameters including Ktrans, Kep, metaplastic carcinoma (n 5 1 [1.1%]).
and Ve are quantitative measures of perfusion,9 thus reflect-
ing the extent of tumor angiogenesis; neoangiogenesis MRI Techniques
increases the vascular permeability of, and blood flow All breast MRIs were performed using a 3T system (Trio Tim, Sie-
within, a malignant tumor. In addition, diffusion-weighted mens Medical Systems, Erlangen, Germany) fitted with a dedicated
four-channel breast array coil (Siemens Medical Systems) with the
imaging (DWI) yields a quantitative parameter, the apparent
patient in the prone position. After a bilateral transverse localizer
diffusion coefficient (ADC), which allows the biological
image was obtained, sagittal fat-suppressed T2-weighted turbo spin-
characteristics of tumors to be assessed. The ADC is affected
echo images were obtained (TR/TE, 6760/111; matrix, 320 3 234;
by cell density and tissue microstructure.10 Using such data, field-of-view [FOV], 220 3 220 mm; slice thickness, 3.0 mm; num-
recent studies have focused on identifying MRI biomarkers ber of excitations [NEX], 3.0; and no gap). Prior to DCE-MRI
that predict breast cancer prognosis. Earlier reports have acquisition, DWI was performed using a single-shot echo-planar
shown that the ADC is associated with well-established imaging technique with fat suppression in the axial plane (TR/TE,
prognostic factors, including tumor size, histological grade, 6,600/91; matrix, 160 3 84; FOV, 380 3 200 mm; slice thickness,
axillary lymph node status, and hormone receptor sta- 5.0 mm; NEX, 7.0; acquisition time, 203 sec; and b values of 0
tus.11,12 Furthermore, Mori et al13 recently found a signifi- and 1000 s/mm2). Diffusion gradients were applied in three
cant correlation between the ADC and Ki-67 status, which orthogonal directions.
also predicts outcomes, in patients with luminal-type breast For T1 mapping, three precontrast T1-weighted fat-saturated
cancer. However, the results of previous studies on the volume-interpolated breath-hold examination (VIBE) sequences with
relationships between perfusion parameters derived from different flip angles (28, 88, and 158) were obtained before contrast
medium was administered; we used the following parameters: TR/TE,
DCE-MRI and prognostic factors (especially the Ki-67
4.3/1.5; matrix, 192 3 134; FOV, 320 3 320 mm; and slice thickness,
index) have been inconsistent. Kim et al14 reported that
2.0 mm. An intravenous bolus injection of 0.1 mmol/kg gadobutrol
tumors with high proliferation indices had significantly
(Gadovist; Bayer Schering Pharma, Berlin, Germany) was adminis-
higher mean values of Ktrans and Kep, but other studies tered using an MR-compatible power injector at a rate of 2 mL/s,
found no such correlations.15–17 followed immediately by a 20-mL saline flush. Forty-three phases of
On the basis of the findings in these previous studies, dynamic contrast-enhanced T1-weighted images were obtained, with a
we questioned whether the Ki-67 status could be predicted temporal resolution of 11 sec, using the following parameters: TR/TE,
preoperatively by using the quantitative MR parameters. If 3.5/3.1; matrix, 192 3 134; FOV, 320 3 320 mm; slice thickness,
associations exist between these parameters and Ki-67, it 2.0 mm; NEX, 1.0; flip angle, 108; and acquisition time, 7 min 53 sec.
might be helpful for predicting outcome and determining
treatment in patients with breast cancer, especially with Imaging Analysis
To analyze breast tumor pharmacokinetics, enhancement kinetics
luminal subtype. Therefore, the purpose of our present
were calculated using the two-compartment pharmacokinetic
study was to explore the association between quantitative
model of Tofts et al.9,18 DCE-MR images were postprocessed using
parameters derived from DCE-MRI and DWI and Ki-67 commercially available software (Tissue 4D; Siemens Healthcare);
proliferation status in patients with ER-positive invasive this yielded the following perfusion parameters: forward volume
breast cancer. transfer constant (Ktrans, min21), reverse volume transfer constant
(Kep, min21), and extravascular extracellular space volume per unit
Materials and Methods volume of tissue (Ve). Motion correction of DCE-MRI data was
Subjects achieved using the nonrigid registration technique of the Tissue
After Institutional Review Board approval, a retrospective analysis 4D program.19 Precontrast T1 values were generated using the vari-
of prospectively collected data from 132 women diagnosed with able flip angle method,20 and the T1 map was aligned to the
invasive breast cancer who underwent preoperative DCE-MRI and motion-corrected images. Signal intensity was converted to gado-
DWI between October 2014 and February 2015 was performed. linium concentration, and time-versus-signal intensity curves were
January 2017 95

to the location and size of the ROI; the radiologists were informed
that the patients had invasive breast cancer but were blinded to the
clinicopathological data. The radiologists identified lesions with
signal intensities higher than those in normal breast tissue on DWI
and then manually drew oval or round ROIs as large as possible
within the breast tumors evident on ADC maps (Fig. 1b). The
ROIs were carefully placed to avoid cystic or necrotic tumor com-
ponents on the basis of T2-weighted imaging and/or DWI.
Histopathological Analysis
All patients underwent surgical resection for breast cancer with sen-
tinel lymph node biopsy and/or axillary lymph node dissection.
Tumor size, histological grade, lymphovascular invasion status, and
axillary lymph node status were determined by reference to the sur-
gically excised specimens. Histological tumor grading was per-
formed using the Nottingham combined histological grading
system.22 The expression statuses of ER and PR, HER2, p53, and
Ki-67 in surgical specimens were determined using an avidin-
biotin IHC technique. ER and PR statuses were assessed using the
Allred scoring system, which is the sum of the proportion score
and intensity score of positively stained tumor cells.23 Tumors with
Allred scores 3 were regarded as positive. The HER2 staining
intensity was scored as 0, 11, 21, or 31.24 Tumors with scores of
3 1 were considered HER2-positive, and tumors with scores of 0
or 1 1 were considered HER2-negative. Tumors with scores of
C
O 2 1 were sent for fluorescence in-situ hybridization (FISH) to
L determine the HER2 status. If the ratio of the HER2 gene signal
O
R to the chromosome 17 probe signal was >2.2, the tumor was con-
FIGURE 1: A 50-year-old woman with breast cancer in the right sidered HER2-positive. P53 status was considered positive if
breast. In the Ktrans-based perfusion map (a), an ROI was drawn 10% of nuclei were stained positively, regardless of the staining
at the portion of the tumor with the highest Ktrans value. The intensity.25 In terms of Ki-67 proliferation status, we divided all
mean Ktrans value was 0.457. In the ADC map (b), the largest
patients into two groups using a Ki-67 cutoff of 14% according to
possible circular ROI was placed within the tumor (arrow). The
mean ADC value was 0.818 3 1023 mm2/s. Surgical histopathol- St. Gallen consensus5; tumors with Ki-67 14% were considered
ogy revealed a 1.6-cm invasive ductal carcinoma of histologic high-Ki-67 tumors; other tumors were of low-Ki-67 tumors.
grade 2, ER-positive, PR-positive, and HER2-negative. The
Ki-67 index was 60%.
Statistical analysis for categorical variables was performed using the
chi-square test or Fisher’s exact test when the expected value in any
calculated. In the present study, for arterial input function (AIF)
cell was less than five. For continuous variables, the Shapiro–Wilk
extraction, population-averaged AIF options within the Tissue 4D
test for normality and Levene’s F test for equal variance were per-
software was used.21 Goodness of fit (chi2) of the three different
formed. If the data were normally distributed and had equal var-
AIFs (fast, intermediate, or slow) was assessed for each tumor, and
lowest chi2, representing highest goodness of fit, was chosen. To iance, Student’s t-test was used. Otherwise, the Mann–Whitney U-
estimate perfusion parameters, a volume-of-interest (VOI) and a test was used.
region-of-interest (ROI) were drawn manually on all images by a Receiver operating characteristic (ROC) curve analysis was
single MRI technician (L.S.P., with 7 years of experience in performed to determine the optimal cutoff values for quantitative
research imaging) who was supervised by an experienced radiologist MR parameters for predicting high-Ki-67 status, using the maxi-
(J.Y.K.). To standardize the image analysis as much as possible, the mum Youden index (ie, sensitivity 1 specificity –1). Multivariate
VOI was placed on a representative slice that shows the largest logistic regression analysis was performed to explore confounder-
dimension of the tumor and included the entire tumor and normal adjusted associations between quantitative MR parameters and
breast tissue adjacent to the tumor. Three to five ROIs were placed high-Ki-67 status; we included all factors that were significant
in the high perfusion area on the automatically generated Ktrans- (P < 0.05) upon univariate analysis. The goodness of fit of the
based perfusion map and the ROI with the highest Ktrans value was models was assessed by the Hosmer–Lemeshow test statistic.26
F1 selected for the analysis (Fig. 1a). All statistical analyses were performed using SPSS (v. 18.0;
To estimate ADC values, all DWI data were retrospectively SPSS, Chicago, IL) and MedCalc (v. 10.3.0.0, MedCalc Software,
reviewed by two radiologists (J.Y.K. and H.J.K., with 5 and 2 years Mariakerke, Belgium) software. P < 0.05 was considered to reflect
of experience in breast MRI, respectively) in consensus with regard statistical significance.
96 Volume 45, No. 1

TABLE 1. Clinicopathological Characteristics by Ki-67 Proliferation Status
Variables Total (n 5 88) Low-Ki-67 (n 5 39) High-Ki-67 (n 5 49) P valuea
Mean age, yearsb 51.53 6 11.52 52.18 6 10.98 51.02 6 12.02 0.643
b
Mean tumor size, cm 1.99 6 1.67 1.52 6 0.69 2.36 6 2.11 0.011
c
Tumor size 0.044
2 cm 57 (64.8) 30 (76.9) 27 (55.1)
>2 cm 31 (35.2) 9 (23.1) 22 (44.9)
c
Histologic grade < 0.001
Grade 1/2 62 (70.5) 37 (94.9) 25 (51.0)
Grade 3 26 (29.5) 2 (5.1) 24 (49.0)
d
Histologic type 0.278
Ductal 80 (90.9) 34 (87.2) 46 (93.9)
Other 8 (9.1) 5 (12.8) 3 (6.1)
c
Lymphovascular invasion 0.063
Negative 71 (80.7) 35 (89.7) 36 (73.5)
Positive 17 (19.3) 4 (10.3) 13 (26.5)
c
Axillary node metastasis 0.005
Negative 72 (81.8) 37 (94.9) 35 (71.4)
Positive 16 (18.2) 2 (5.1) 14 (28.6)
d
PR status 0.236
Negative 9 (10.2) 3 (7.7) 6 (12.2)
Positive 79 (89.8) 36 (92.3) 43 (87.8)
c
HER2 status 0.236
Negative 64 (72.7) 31 (79.5) 33 (67.3)
Positive 24 (27.3) 8 (20.5) 16 (32.7)
d
P53 status 0.012
Negative 77 (87.5) 38 (97.4) 39 (79.6)
Positive 11 (12.5) 1 (2.6) 10 (20.4)
a
P value for low-Ki-67 versus high-Ki-67 comparison.
b
Data are mean values 6 standard deviations. Data were tested using the Student’s t-test.
c
Data were tested using the chi-square test.
d
Data were tested using the Fisher’s exact test.
Numbers in parentheses are percentages. PR, progesterone receptor; HER2, human epidermal growth factor receptor 2.
Results positive axillary lymph nodes (28.6% vs. 5.1%, P 5 0.005),

The Ki-67 proliferation indices of surgically excised specimens and positive P53 expression (20.4% vs. 2.6%, P 5 0.012)
ranged from 0 to 90. Using the 14% cutoff, 49 tumors when compared to the low-Ki-67 group (Table 1).
(55.7%) were categorized as high-Ki-67 and the remaining 39 Table 2 shows the relationships between quantitative MRI T2
(44.3%) as low-Ki-67. Patient clinicopathological characteris- parameters and histopathological variables. The mean Ktrans
tics and their associations with the Ki-67 proliferation status value in the high-Ki-67 group was significantly higher than that
T1 are shown in Table 1. The mean tumor size differed signifi- in the low-Ki-67 group (0.279 6 0.145 vs. 0.177 6 0.117,
cantly between the high- and low-Ki-67 groups (2.36 6 2.11 P < 0.001) (Fig. 2a). In addition, the mean ADC of the high- F2
vs. 1.52 6 0.69 cm, P 5 0.011). The high-Ki-67 group had Ki-67 group was significantly lower than that of the low-Ki-67
more large size (>2 cm; 44.9% vs. 23.1%, P 5 0.044), higher group (0.913 3 1023 vs. 1.029 3 1023 mm2/s, P < 0.001)
histologic grade (grade 3; 49.0% vs. 5.1%, P < 0.001), (Fig. 2b). However, the mean Kep and Ve values did not differ
January 2017 97

TABLE 2. Relationships Between Quantitative MRI Parameters and Histopathological Variables
Variables Krans Kep Ve ADC (3 1023 mm2/s)
Ki-67 index
Low-Ki-67 ( < 14%) 0.177 6 0.117 0.478 6 0.195 0.387 6 0.160 1.029 6 0.177
High-Ki-67 ( 14%) 0.279 6 0.145 0.556 6 0.257 0.462 6 0.182 0.913 6 0.165
P value < 0.001 0.248 0.055 < 0.001
Tumor size
2 cm 0.195 6 0.104 0.480 6 0.193 0.418 6 0.174 0.983 6 0.160
>2 cm 0.305 6 0.173 0.598 6 0.282 0.448 6 0.180 0.931 6 0.208
P value 0.004 0.060 0.429 0.044
Histologic grade
Grade 1/2 0.206 6 0.139 0.478 6 0.182 0.412 6 0.170 1.009 6 0.181
Grade 3 0.299 6 0.128 0.624 6 0.306 0.467 6 0.186 0.858 6 0.123
P value 0.001 0.026 0.276 < 0.001
Histologic type
Ductal 0.228 6 0.138 0.505 6 0.221 0.428 6 0.177 0.956 6 0.157
Other 0.299 6 0.172 0.687 6 0.314 0.437 6 0.178 1.056 6 0.335
P value 0.184 0.071 0.217 0.647
Lymphovascular invasion
Negative 0.225 6 0.140 0.514 6 0.210 0.426 6 0.176 0.957 6 0.157
Positive 0.268 6 0.147 0.551 6 0.321 0.440 6 0.179 0.995 6 0.258
P value 0.230 1.000 0.684 0.568
Axillary node metastasis
Negative 0.226 6 0.141 0.493 6 0.208 0.436 6 0.176 0.976 6 0.142
Positive 0.270 6 0.146 0.650 6 0.301 0.395 6 0.177 0.914 6 0.296
P value 0.211 0.035 0.384 0.017
HER2 status
Negative 0.212 6 0.121 0.495 6 0.217 0.423 6 0.165 0.984 6 0.189
Positive 0.292 6 0.176 0.592 6 0.265 0.444 6 0.205 0.911 6 0.139
P value 0.049 0.082 0.736 0.072
P53 status
Negative 0.235 6 0.148 0.518 6 0.243 0.435 6 0.179 0.973 6 0.184
Positive 0.228 6 0.092 0.549 6 0.157 0.384 6 0.153 0.905 6 0.134
P value 0.587 0.145 0.995 0.228
Data are mean values 6 standard deviations. All data were tested using the Mann-Whitney U-test. HER2, human epidermal growth
factor receptor 2.
significantly between the two groups (P 5 0.248 and P 5 (grade 3) (P < 0.001), and an association with axillary lymph
0.055, respectively). The mean Ktrans value was significantly node metastasis (P 5 0.017). However, no associations were
higher for tumors with a larger size (>2 cm) (P 5 0.004), higher found between quantitative MRI parameters and histologic type,
histological grade (grade 3) (P 5 0.001), and HER2-positivity lymphovascular invasion, or P53 status (Table 2).
(P 5 0.049). The mean ADC was significantly lower for tumors ROC curve analysis showed that the mean Ktrans and
with a larger size (>2 cm) (P 5 0.044), higher histological grade ADC values significantly discriminated the high-Ki-67 from
98 Volume 45, No. 1

FIGURE 2: Boxplot graphs revealing statistically significant differences in the mean Ktrans (a) (P < 0.001) and ADC (b) (P < 0.001) val-
ues between the high- and low-Ki-67 groups. 䊊 5 outlier, D 5 extreme value.
the low-Ki-67 group. The optimal cutoff values for Ktrans status (Table 3). These variables were included in the multi- T3
and ADC in terms of predicting high-Ki-67 were 0.274 variate analysis. In the multivariate analysis, a higher Ktrans
(sensitivity 5 53.1%, specificity 5 92.3%, area under the value (>0.274; adjusted odds ratio [OR]: 9.027, 95% CI:
curve 5 0.728, standard error 5 0.055, 95% confidence 1.929–42.245; P 5 0.005) and a higher histological grade
interval [CI] 5 0.623–0.817, P < 0.001) and 0.893 3 1023 (grade 3; adjusted OR: 7.510, 95% CI: 1.305–43.205;
mm2/s (sensitivity 5 46.9%, specificity 5 87.2%, area under P 5 0.024) remained independently associated with a high-
the curve 5 0.722, standard error 5 0.054, 95% Ki-67 status after adjustment for tumor size, axillary nodal
F3 CI 5 0.616–0.812, P < 0.001), respectively (Fig. 3). We status, P53 status, and ADC value (Table 4). The Hosmer– T4
performed binary regression analysis using these cutoff Lemeshow statistic showed a good fit for the final model
values. (chi-square 5 1.037, df 5 5, P 5 0.960).
On univariate analysis, a higher Ktrans value (>0.274)
(P < 0.001) and a lower ADC (0.893 3 1023 mm2/s) Discussion
(P 5 0.001), large tumor size (>2 cm) (P 5 0.036), a higher We evaluated the associations between MRI perfusion and
histological grade (grade 3) (P < 0.001), the presence of diffusion parameters and the Ki-67 proliferation status in
axillary node metastasis (P 5 0.011), and positive P53 status patients with ER-positive invasive breast cancer. First, we
(P 5 0.034) were significantly associated with high-Ki-67 found a significant association between Ktrans and the Ki-67
FIGURE 3: Receiver operating characteristic curve analysis for differentiating high- (14%) from low- (<14%) Ki-67 tumors (a);
Ktrans: optimal cutoff, 0.274; area under the curve, 0.728; standard error, 0.055; 95% confidence interval, 0.623–0.817. (b) ADC;
optimal cutoff, 0.893 3 1023 mm2/s; area under the curve, 0.722; standard error, 0.054; 95% confidence interval, 0.616–0.812.
January 2017 99

TABLE 3. Univariate Logistic Regression Analysis of Variables Associated With High-Ki-67 Status.
Variables Odds ratio 95% confidence interval P value
Tumor size ( > 2 cm vs. 2 cm ) 2.716 1.068, 6.909 0.036

Histologic grade (grade 3 vs. grade1/2) 17.760 3.849, 81.948 < 0.001
Histologic type (other vs. ductal) 2.255 0.504, 10.090 0.288
Lymphovascular invasion (presence vs. absence) 3.160 0.939, 10.632 0.063
Axillary node metastasis (positive vs. negative) 7.400 1.567, 34.935 0.011
HER2 status (positive vs. negative) 1.879 0.705, 5.006 0.207
P53 status (positive vs. negative) 9.744 1.189, 79.856 0.034
trans
K ( > 0.274 vs. 0.274) 13.565 3.680, 50.004 < 0.001
23 2
ADC ( 0.893 vs. > 0.893) (x 10 mm /s) 6.015 2.015, 17.957 0.001
proliferation status. Tumors with higher Ki-67 indices had standard methodology and accepted cutoff points for Ki-67
higher Ktrans values. Furthermore, multivariate analysis index and validate these findings.
revealed that a higher Ktrans was the independent factor Several reports on the relationships between perfusion-
associated with a high-Ki-67 status, followed by a higher derived DCE-MRI parameters and breast cancer prognosis
histologic grade. Since the positive correlation between Ki- have appeared, but their conclusions vary.14–17,29 Koo et al
67 status and histological grade has been well accepted, the found that higher Ktrans, Kep, or lower Ve was associated
strongest independent relationship found in the present with high histological grade, high nuclear grade, or ER neg-
study between Ktrans and Ki-67 status is very meaningful. ativity.16 However, in one recent study neither Ktrans, Kep,
In terms of diffusion parameters, we noted that a nor Ve was significantly associated with the histological or
lower ADC was associated with high-Ki-67 status, consistent nuclear grade or hormone receptor status; significant corre-
with the results of recent studies.13,17,27,28 In our present lations were evident between the Ki-67 index, on the one
work, however, the ADC was not an independent predictor hand, and both Ktrans and Kep.14 Such inconsistent results
of outcomes on multivariate analysis. This may be attributed may be attributed to differences among study groups and
to the high correlation between the mean ADC and histopa- the use of diverse methods to quantify perfusion parameters.
thologically prognostic factors, including tumor size, histo- In the present study, we found that higher Ktrans and Kep
logical grade, and the presence/absence of axillary lymph values were associated with a higher histological grade, and
node metastasis. As these factors are highly correlated, a sep- a higher Ktrans was associated with a high-Ki-67 status. It
arate, independent relationship for these variables would be remains unclear why Ktrans is associated with the Ki-67 sta-
weaker. Furthermore, previous studies were based on differ- tus; it may be due to the vascular characteristics of tumors
ent cutoff values for Ki-67 expression, which makes compar- with high Ki-67 indices. A significant correlation between
isons difficult. Further studies are required to establish the Ki-67 status and vascular endothelial growth factor
TABLE 4. Multivariate Logistic Regression Analysis of Variables Associated With High-Ki-67 Status
Variables Adjusted 95% confidence P value

odds ratio interval
Tumor size ( > 2 cm vs. 2 cm) 0.912 0.242, 3.437 0.892

Histologic grade (grade 3 vs. grade1/2) 7.510 1.305, 43.205 0.024
Axillary node metastasis (positive vs. negative) 5.238 0.713, 38.462 0.104
P53 status (positive vs. negative) 8.356 0.800, 87.275 0.076
trans
K ( > 0.274 vs. 0.274) 9.027 1.929, 42.245 0.005
23 2
ADC ( 0.893 vs. > 0.893) (x 10 mm /s) 3.752 0.938, 15.000 0.061
100 Volume 45, No. 1

(VEGF) receptor 2 was found in breast cancer.30 VEGF and biology and increases the clinical utilities of DCE-MRI and
its receptor play an important role in angiogenesis and DWI in the field of breast oncology, especially when indi-
induce increased vascular permeability.31 We assume that vidualized treatment plans are required. In addition, if this
tumors with high Ki-67 indices exhibit higher-level angioge- information could be incorporated into the routine diagnos-
netic activity and higher vascular permeability. Of the vari- tic work-up of patients with breast cancer, the value would
ous perfusion parameters, Ktrans reflects a combination of be enhanced. Further studies should focus on the use of
blood flow, vessel permeability, and vessel density, while Kep these functional parameters in clinical practice and their
reflects only vessel permeability.9 Thus, we assume that potential application to breast cancer management.
Ktrans might better represent the actual vascular features of a Our study had several limitations. First, the work was
tumor than might Kep. This would facilitate the identifica- retrospective in nature, and our patient numbers were small;
tion of highly proliferative tumors, as indicated by the Ki- some selection bias was inevitable. Second, we adopted 14%
67 index. as the Ki-67 cutoff (values 14% indicated high-level pro-
Clinically, the Ki-67 proliferation index is generally liferation). However, the optimal Ki-67 cutoff in clinical
determined immunohistochemically.32 Ki-67 status has been practice is still being debated vigorously. Third, we only
widely investigated in breast cancer patients; a high-Ki67 focused on the perfusion and diffusion characteristics of
status is associated with a poor prognosis and poor response breast cancer and did not evaluate the morphology of the
to treatment. A higher Ki-67 index after neoadjuvant chem- lesions on MRI. Incorporation of morphological assessment
otherapy, indicating that a residual tumor burden exists, is of lesions by using the Breast Imaging Reporting and Data
associated with a poor outcome.6,7 Viale et al33 compared System (BI-RADS) MR lexicon may be of value in clinical
the effects of letrozole and tamoxifen by the Ki-67 index; practice. Fourth, when quantifying MRI-derived parameters,
letrozole afforded more benefits in patients with higher indi- we manually drew ROIs within the areas of tumors that had
ces. In addition, Ki-67 is a candidate biomarker for luminal the highest Ktrans values on perfusion maps or had the larg-
B tumors in ER-positive cases. Some luminal B cancers can est diameter on ADC maps. These may not accurately
be identified by their expression of HER2, but HER2 status reflect the characteristics of the entire tumor. Furthermore,
alone is not sensitive enough to identify most luminal B we did not assess the reproducibility in the placement of the
cancers and the major biological distinction between lumi- ROIs and its effect on the results. Histogram analysis of per-
nal A and B is the proliferation signature.3 Therefore, the fusion and diffusion MR data might more accurately reflect
Ki-67 index, in conjunction with more traditional prognos- the characteristics of the entire tumor and could improve
tic factors, may serve as both a prognostic and predictive reproducibility for measurement of MRI parameters. Finally,
marker, allowing identification of patients likely to experi- we did not explore possible correlations between quantita-
ence poor outcomes, thus permitting clinicians to select tive MRI data and long-term clinical outcomes. A high Ki-
appropriate individual treatments early. 67 value is associated with poor outcomes.32 Further pro-
Recently, several studies have sought to identify DCE- spective investigations are needed to validate the clinical
MRI and DWI imaging biomarkers that can predict prog- utility of our work.
nosis and the response to therapy. A reduction in Ktrans or In conclusion, MRI-derived quantitative parameters
Kep and an increase in the ADC after neoadjuvant chemo- such as Ktrans and the ADC were associated significantly
therapy are known to be early predictors of the response to with Ki-67 proliferation status in patients with ER-positive
such chemotherapy.34–36 Furthermore, Li et al37 showed in invasive breast cancer. On multivariate analysis, a higher
a prospective study that a higher posttreatment Ktrans inde- Ktrans value was the strongest independent predictor of a
pendently predicted overall survival. After adjusting for high-Ki-67 status. Further work is needed to confirm that
other prognostic factors, every increment of 0.01 in the quantitative MRI parameters facilitate noninvasive assess-
posttreatment Ktrans was accompanied by a 4% increase in ment of prognosis in breast cancer patients.
the risk of death.37 In the present study we explored the
relationships between quantitative MRI parameters, such as
Ktrans and ADC, and Ki-67 proliferation status in patients Acknowledgment
with ER-positive invasive breast cancer. We found that these We thank Li Sa Park for assistance in measurement of
parameters usefully and preoperatively identified tumors perfusion parameters.
with higher Ki-67 indices (thus, more aggressive luminal B
subtype). Our results suggested that preoperative MRI
parameters could be used to identify a poor prognosis sub-
group that would likely benefit from additional systemic
References
1. Desmedt C, Haibe-Kains B, Wirapati P, et al. Biological processes asso-
therapy among patients with luminal subtypes. We believe ciated with breast cancer clinical outcome depend on the molecular
that this knowledge improves our understanding of tumor subtypes. Clin Cancer Res 2008;14:5158–5165.
January 2017 101

2. Rouzier R, Perou CM, Symmans WF, et al. Breast cancer molecular 21. Orton MR, d’Arcy JA, Walker-Samuel S, et al. Computationally effi-
subtypes respond differently to preoperative chemotherapy. Clin Can- cient vascular input function models for quantitative kinetic modelling
cer Res 2005;11:5678–5685. using DCE-MRI. Phys Med Biol 2008;53:1225–1239.
3. Cheang MC, Chia SK, Voduc D, et al. Ki67 index, HER2 status, and 22. Elston CW, Ellis IO. Pathological prognostic factors in breast cancer. I.
prognosis of patients with luminal B breast cancer. J Natl Cancer Inst The value of histological grade in breast cancer: Experience from
2009;101:736–750. a large study with long-term follow-up. Histopathology 1991;19:
403–410.
4. Wintzer H, Zipfel I, Schulte-M€
onting J, Hellerich U, von Kleist S. Ki-67
immunostaining in human breast tumors and its relationship to prog- 23. Allred DC, Harvey JM, Berardo M, Clark GM. Prognostic and predic-
nosis. Cancer 1991;67:421–428. tive factors in breast cancer by immunohistochemical analysis. Mod
Pathol 1998;11:155–168.
5. Goldhirsch A, Wood WC, Coates AS, et al. Strategies for subtypes—
dealing with the diversity of breast cancer: Highlights of the St. Gallen 24. Moeder CB, Giltnane JM, Harigopal M, et al. Quantitative justification
international expert consensus on the primary therapy of early breast of the change from 10% to 30% for human epidermal growth factor
cancer 2011. Ann Oncol 2011;22:1736–1747. receptor 2 scoring in the American Society of Clinical Oncology/Col-
lege of American Pathologists guidelines: Tumor heterogeneity in
6. Jones RL, Salter J, A’Hern R, et al. The prognostic significance of Ki67
breast cancer and its implications for tissue microarray based assess-
before and after neoadjuvant chemotherapy in breast cancer. Breast
ment of outcome. J Clin Oncol 2007;25:5418–5425.
Cancer Res Treat 2009;116:53–68.
25. Yamashita H, Toyama T, Nishio M, et al. p53 protein accumulation
7. Dowsett M, Smith IE, Ebbs SR, et al. Prognostic value of Ki67 expres-
predicts resistance to endocrine therapy and decreased post-relapse
sion after short-term presurgical endocrine therapy for primary breast
survival in metastatic breast cancer. Breast Cancer Res 2006;8:R48.
cancer. J Natl Cancer Inst 2007;99:167–170.
26. Hosmer DW, Lemeshow S. Applied logistic regression. New York:
8. Petrelli F, Viale G, Cabiddu M, Barni S. Prognostic value of different cut-
John Wiley & Sons; 1989.
off levels of Ki-67 in breast cancer: a systematic review and meta-
analysis of 64,196 patients. Breast Cancer Res Treat 2015;153:477–491. 27. Kim EJ, Kim SH, Park GE, et al. Histogram analysis of apparent diffu-
sion coefficient at 3.0T: Correlation with prognostic factors and sub-
9. Tofts PS, Brix G, Buckley DL, et al. Estimating kinetic parameters from
types of invasive ductal carcinoma. J Magn Reson Imaging 2015;42:
dynamic contrast-enhanced T 1-weighted MRI of a diffusable tracer:
1666–1678.
Standardized quantities and symbols. J Magn Reson Imaging 1999;
10:223–232. 28. Molinari C, Clauser P, Girometti R, et al. MR mammography using
diffusion-weighted imaging in evaluating breast cancer: a correlation
10. Partridge SC. Future applications and innovations of clinical breast mag-
with proliferation index. Radiol Med 2015;120:911–918.
netic resonance imaging. Top Magn Reson Imaging 2008;19:171–176.
29. Radjenovic A, Dall BJ, Ridgway JP, Smith MA. Measurement of phar-
11. Kim JY, Seo HB, Park S, et al. Early-stage invasive ductal carcinoma:
macokinetic parameters in histologically graded invasive breast
Association of tumor apparent diffusion coefficient values with axillary
tumours using dynamic contrast-enhanced MRI. Br J Radiol 2008;81:
lymph node metastasis. Eur J Radiol 2015;84:2137–2143.
120–128.
12. Razek AA, Gaballa G, Denewer A, Nada N. Invasive ductal carcinoma:
30. Nakopoulou L, Stefanaki K, Panayotopoulou E, et al. Expression of
Correlation of apparent diffusion coefficient value with pathological
the vascular endothelial growth factor receptor-2/Flk-1 in breast carci-
prognostic factors. NMR Biomed 2010;23:619–623.
nomas: correlation with proliferation. Hum Pathol 2002;33:863–870.
13. Mori N, Ota H, Mugikura S, et al. Luminal-type breast cancer: Correla-
31. Gavard J, Gutkind JS. VEGF controls endothelial-cell permeability by
tion of apparent diffusion coefficients with the Ki-67 labeling index.
promoting the beta-arrestin-dependent endocytosis of VE-cadherin.
Radiology 2014;274:66–73.
Nat Cell Biol 2006;8:1223–1234.
14. Kim JY, Kim SH, Kim YJ, et al. Enhancement parameters on dynamic
32. Urruticoechea A, Smith IE, Dowsett M. Proliferation marker Ki-67 in
contrast enhanced breast MRI: Do they correlate with prognostic fac-
early breast cancer. J Clin Oncol 2005;23:7212–7220.
tors and subtypes of breast cancers? Magn Reson Imaging 2015;33:
72–80. 33. Viale G, Giobbie-Hurder A, Regan MM, et al. Prognostic and predic-
tive value of centrally reviewed Ki-67 labeling index in postmeno-
15. Shin HJ, Kim HH, Shin KC, et al. Prediction of low-risk breast cancer
pausal women with endocrine-responsive breast cancer: Results from
using perfusion parameters and apparent diffusion coefficient. Magn
breast international group trial 1-98 comparing adjuvant tamoxifen
with letrozole. J Clin Oncol 2008;26:5569–5575.
16. Koo HR, Cho N, Song IC, et al. Correlation of perfusion parameters
34. Thukral A, Thomasson DM, Chow CK, et al. Inflammatory breast can-
on dynamic contrast-enhanced MRI with prognostic factors and sub-
cer: Dynamic contrast-enhanced MR in patients receiving bevacizu-
types of breast cancers. J Magn Reson Imaging 2012;36:145–151.
mab—initial experience. Radiology 2007;244:727–735.
17. Li L, Wang K, Sun X, et al. Parameters of dynamic contrast-enhanced
35. Ah-See ML, Makris A, Taylor NJ, et al. Early changes in functional
MRI as imaging markers for angiogenesis and proliferation in human
dynamic magnetic resonance imaging predict for pathologic response
breast cancer. Med Sci Monit 2015;21:376–382.
to neoadjuvant chemotherapy in primary breast cancer. Clin Cancer
18. Tofts PS, Kermode AG. Measurement of the blood? brain barrier perme- Res 2008;14:6580–6589.
ability and leakage space using dynamic MR imaging. 1. Fundamental
36. Park SH, Moon WK, Cho N, et al. Diffusion-weighted MR imaging:
concepts. Magn Reson Med 1991;17:357–367.
Pretreatment prediction of response to neoadjuvant chemotherapy in
19. Crum WR, Hartkens T, Hill DL. Non-rigid image registration: theory patients with breast cancer. Radiology 2010;257:56–63.
and practice. Br J Radiol 2004;77:S140–153.
37. Li SP, Makris A, Beresford MJ, et al. Use of dynamic contrast-
20. Fram EK, Herfkens RJ, Johnson GA, et al. Rapid calculation of T1 enhanced MR imaging to predict survival in patients with primary
using variable flip angle gradient refocused imaging. Magn Reson breast cancer undergoing neoadjuvant chemotherapy. Radiology
Imaging 1987;5:201–208. 2011;260:68–78.

ORIGINAL RESEARCH
Haralick Textural Features on T2-Weighted

MRI Are Associated With Biochemical
Recurrence Following Radiotherapy for
Peripheral Zone Prostate Cancer
Khemara Gnep, MD,1,2,3* Aur eline Fargeas, PhD,1,2
errez-Carvajal, PhD,1,2 Fr
Ricardo E. Guti eric Commandeur, PhD,1,2
ed
Romain Mathieu, MD,1,2,4 Juan D. Ospina, PhD,1,2 Yan Rolland, MD,6
Tanguy Rohou, MD,5,6 S
ebastien Vincendeau, MD,4 Mathieu Hatt, PhD,7
Oscar Acosta, PhD,1,2 and Renaud de Crevoisier, MD, PhD1,2,3
Purpose: To explore the association between magnetic resonance imaging (MRI), including Haralick textural features,
and biochemical recurrence following prostate cancer radiotherapy.
Materials and Methods: In all, 74 patients with peripheral zone localized prostate adenocarcinoma underwent pretreat-
ment 3.0T MRI before external beam radiotherapy. Median follow-up of 47 months revealed 11 patients with biochemi-
cal recurrence. Prostate tumors were segmented on T2-weighted sequences (T2-w) and contours were propagated onto
the coregistered apparent diffusion coefficient (ADC) images. We extracted 140 image features from normalized T2-w
and ADC images corresponding to first-order (n 5 6), gradient-based (n 5 4), and second-order Haralick textural features
(n 5 130). Four geometrical features (tumor diameter, perimeter, area, and volume) were also computed. Correlations
between Gleason score and MRI features were assessed. Cox regression analysis and random survival forests (RSF) were
performed to assess the association between MRI features and biochemical recurrence.
Results: Three T2-w and one ADC Haralick textural features were significantly correlated with Gleason score (P < 0.05).
Twenty-eight T2-w Haralick features and all four geometrical features were significantly associated with biochemical
recurrence (P < 0.05). The most relevant features were Haralick features T2-w contrast, T2-w difference variance, ADC
median, along with tumor volume and tumor area (C-index from 0.76 to 0.82; P < 0.05). By combining these most
powerful features in an RSF model, the obtained C-index was 0.90.
Conclusion: T2-w Haralick features appear to be strongly associated with biochemical recurrence following prostate can-
cer radiotherapy.
antigen (PSA) rate, and Gleason score.2 Patients with inter-

E xternal beam radiotherapy (RT) is considered the
standard treatment for localized prostate cancer. How-
ever, associated 5-year biochemical recurrence rates are
mediate- or high-risk prostate cancer can be treated with
either RT, radical prostatectomy, or brachytherapy, with or
reported at 0–10%, 10–20%, and 30–40% for each of the without androgen deprivation therapy for the most aggres-
three D’Amico risk classifications (low, intermediate, and sive tumors. These risk groups, however, are heterogeneous
high, respectively).1 These are defined based on the clinical in terms of radiosensitivity, rendering it important to iden-
T-stage (on digital rectal examination), prostate-specific tify other predictors of recurrence to intensify local or

*Address reprint requests to: K.G., Avenue de la Bataille Flandres-Dunkerque, cs 44229, F-35042 Rennes, France. E-mail: khemara.gnep@gmail.com
From the 1INSERM, U1099, Rennes, France; 2Universite de Rennes 1, LTSI, Rennes, France; 3Department of Radiotherapy, Centre Eugène Marquis, Rennes,
France; 4Department of Urology, Centre Hospitalier Universitaire Pontchaillou, Rennes, France; 5Department of Radiology, Centre Hospitalier Universitaire
Pontchaillou, Rennes, France; 6Department of Radiology, Centre Eugène Marquis, Rennes, France; and 7LaTIM, INSERM UMR 1101, University of Brest,
France

V 103

tures, derived from the gray-level co-occurrence matrix

(GLCM), quantify the spatial relationships between neigh-
boring voxels using second-order statistics. Numerous quan-
titative parameters can then be calculated within these
matrices, corresponding to different characterizations of
intensity patterns within the image (see Table 3).11 Haralick
parameters extracted from MRI prior to initiating treatment
were associated with tumor response and/or survival in the
field of cancer (but never in the case of prostate can-
cer).12–17 Haralick textural features have already been used
to discriminate prostate cancer from healthy tissue18,19 or
for Gleason score assessment.20,21
Thus, this study sought to assess the association
between features extracted from T2-weighted (T2-w) and
apparent diffusion coefficient map (ADC) MRI images,
including Haralick textural features, and biochemical
TABLE 1. Characteristics of Patients and Tumors
Patient characteristics
Patients 74
FIGURE 1: Flow chart of the study. RT, radiotherapy; ADT,
androgen deprivation therapy; T1-w, T1-weighted. Median age (y) 68 (range 50-82)
Tumor characteristics
systemic treatments for tumors with the highest recurrence Median pretreatment PSA (ng/mL) 9.8 (range 3.4-47.0)
risk. PSA 10 46 (57%)
Over the last few years, imaging biomarkers, popularly 10 < PSA 20 23 (31%)
called “radiomics,” have been recognized as potentially
PSA > 20 9 (12%)
powerful tools to aid in identifying patients with tumor sub-
groups with high recurrence risk.3 Multiparametric prostate Clinical stage (AJCC 1992)
magnetic resonance imaging (MRI) is primarily performed T1c 10 (14%)
in prostate cancer cases for staging, as well as assessing T2a 5 (6%)
tumor location, local extent (extracapsular extension and/or
T2b 26 (35%)
seminal vesicle invasion), and lymph node involvement. The
recently developed PI-RADS (Prostate Imaging Reporting T2c 10 (14%)
and Data System) is a structured reporting scheme designed T3 23 (31%)
to evaluate the prostate,4 able to detect tumors with a sensi- Extra-capsular extension (MRI) 42 (57%)
tivity ranging from 0.70 to 0.84 and specificity from 0.68
Seminal vesical invasion (MRI) 15 (20%)
to 0.86, as shown in a recent meta-analysis.5
While qualitative analysis of prostate MRI plays a cru- Gleason score
cial role at the pretreatment staging step, quantitative analy- 6 19 (26%)
sis of MRI images is, in contrast, rarely used in clinical 7 36 (48%)
practice and especially for being associated with recurrence. 8 and 9 19 (26%)
The only established parameters associated with survival are
Group at risk (according to D’Amico)
therefore MRI-derived T-stage, tumor volume, extracapsular
extension, seminal vesicle invasion, and volume of malignant Intermediate 32 (43%)
spectroscopic MRI metabolism.6–9 High 42 (57%)
Textural features, characterizing the spatial variations Nodal involvement 7 (9%)
of signal intensity in an image, may be particularly useful in
predicting recurrence, as they could quantify underlying AJCC, American joint commission on cancer; PSA, prostate-
specific antigen. Data are numbers (%) unless otherwise
pathophysiological processes associated with tumor aggres- specified.
siveness or resilience to treatment.10 Haralick textural fea-

Gnep et al.: Impact of MRI in Prostate Cancer
TABLE 2. MRI Sequence Parameters By MRI Scanner
Sequence parameter Siemens Verio Philips Achieva

MRI (n 5 47) MRI (n 5 27)
T2-weighted
Matrix (pixels) 448 3 448 720 3 720
Field of view (mm) 239 3 239 300 3 300
TE (msec) 100-159 120
TR (msec) 4000-7291 2700-4352
Slice thickness (mm) 3 3
ADC map
Matrix (pixels) 102 3 160 256 3 256
Field of view (mm) 221 3 260 240 3 240
TE (msec) 95 95
TR (msec) 4400 1286-2319
Section thickness (mm) 6 6
Diffusion gradient B0-100-400-800 B0-150-400-600-1000
TR, repetition time; TE, echo time; ADC, apparent diffusion coefficient.
recurrence following radiotherapy for peripheral zone pros- intensity-modulated RT (IMRT) and image-guided RT (IGRT),
tate cancer. Due to the large number of tested features and delivering total doses of 74–80 Gy to the prostate. High-risk
relatively limited number of events, machine-learning meth- tumor patients (57%) also received adjuvant ADT for 36 months.
ods were combined with more standard statistical tests in Patients were followed up by means of clinical examination and
order to identify the most powerful group of features to be PSA level analysis every 6 months after RT for 5 years. After a
median follow-up of 47 months (range: 19–65 months), 11
associated with biochemical recurrence.
patients exhibited biochemical recurrence, defined according to the
Materials and Methods Phoenix criteria (2 ng/ml increase above the PSA nadir after RT).22
The 4-year freedom-from-biochemical-recurrence rate was 81%
Patient, Tumor, and Treatment Characteristics, as (95% confidence interval [CI]: 70–92%).
Well as Biochemical Control
Our Institutional Review Board approved this retrospective study MRI Technique
and waived the need to obtain informed patient consent. From MRI scanning was performed using two different 3.0 T scanners
January 2009 to February 2012, a total of 349 consecutive patients (Achieva TX, Philips Healthcare, Best, The Netherlands; Magne-
were identified from our electronic medical database as having tom Verio, Siemens Healthcare, Malvern, PA), both with 6-
undergone RT for localized prostate cancer at our institution. The channel cardiac phased-array surface coils, with the patients
following inclusion criteria applied: histologically confirmed pros- scanned in supine position. MRI sequences included axial turbo
tate adenocarcinoma on needle biopsy; intermediate- and high-risk spin echo T2-w and axial diffusion sequences using multiple b val-
localized nonmetastatic prostate cancer; multiparametric 3.0T ues. ADC maps were calculated by the manufacturer’s software.
MRI, according to European Society of Urogenital Radiology rec- The sequences’ parameters are detailed in Table 2. T2
ommendations, where the tumor had to be apparent and in the
peripheral zone (PZ). The exclusion criteria were: T4 stage; pres- Preprocessing of T2-w and ADC Images
ence of postbiopsy hemorrhage on MRI; radical prostatectomy Only T2-w and ADC images were considered. Figure 2 depicts the F2
before RT; androgen deprivation therapy (ADT) before MRI; total study workflow, divided into three steps: manual segmentations,
RT dose <74 Gy, and follow-up period <18 months. Finally, region of interest (ROI) propagation, and feature extraction.
according to those criteria, 74 patients were considered in our All manual segmentations of the tumor, prostate, PZ, and
F1 study. A flow chart of the study is presented in Fig. 1. transition zone were performed by a radiation oncologist (with 5
Patient, tumor, and treatment characteristics are presented in years of experience: K.G.) and each segmentation was validated by
T1 Table 1. All patients underwent a total of 12 transrectal biopsies a senior radiologist with 30 years of experience, mainly in prostate
that can be divided into six biopsies in each lobe. In addition, all cancer (Y.R.). They were delineated on the axial T2-w images on
biopsies were performed before MRI. All patients underwent each slice with the help of available MRI images (including T1-w,
January 2017 105

C
O
L
O
R
FIGURE 2: Overview workflow of the method. RT, radiotherapy; T1-w, T1-weighted; T2-w, T2-weighted; ADC, apparent diffusion
coefficient; DCE, dynamic contrast-enhanced; ROIs, regions of interest; TZ, transition zone; PZ, peripheral zone; GLCM, gray-level
co-occurrence matrix.
T2-w, diffusion, ADC, and dynamic contrast-enhanced [DCE] perfu- tration and B-spline-based resampling were applied between the
sion), MRI radiology reports, and pathology biopsy reports. Tumor ADC and T2-w images when necessary. Each registration was visually
segmentations had to be highly representative of PC on all sequences. validated by the same expert. As T2-w and ADC intensity values can
This meant no postbiopsy hemorrhage on T1-w, a focal hypointense vary over several parameters, a normalization per voxel for each
signal on T2-w, a diffusion restriction on the diffusion (hyperintense sequence was performed, which also enabled comparison between
signal) and ADC (hypointense signal) images, and an increased early patient data. The normalized signal intensity was the ratio of the raw
enhancement on DCE perfusion. We used ITK-SNAP software for signal intensity to the maximal signal intensity of the sequence,
segmentation (open source software; www.itk-snap.org). Rigid regis- which corresponded to the urine signal inside the bladder.23

C
O
L
O
R
FIGURE 3: Gray-level co-occurrence matrix (GLCM). The number of gray-levels of the gray-scale image is voluntarily reduced to 5 in
a 5 3 5 pixels matrix and is reproduced in an analogous numeric gray-levels grid. The GLCM is constructed by considering the rela-
tionships between pixels on a particular angle (08, 458, 908, 1358). For example, in the left-right direction (or right-left, at 08) a “3”
has one “5” neighbor and six “3” neighbors still at 08. Then the normalized GLCM is obtained by taking account of the probability
and not the count of the co-occurrence. Haralick’s textural features are obtained by applying different formulas to the GLCM.
January 2017 107

TABLE 3. Feature Characteristics
First order features

Geometrical features Tumor volume; maximal tumor diameter; maximal tumor perimeter; maxi-
mal tumor area
X
Mean 1
uðyÞ
jvðxÞj
y2vðxÞ
Median m value that ensures that the pixel number that belongs to 5(x) of greater
or equal than m is equal to the pixel number of lesser or equal than m
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2ffi
Standard deviation X
1
jvðxÞj uðyÞ2uðxÞ
y2vðxÞ
Gradient-based features
2 3 2 3
21 10 11 11 12 11
6 7 6 7
6 7 6 7
G156
6 22 10 12 7
7; G256
6 10 10 10 7
7
4 5 4 5
21 10 11 21 22 21
Sobel gradient filter 2 3 2 3
20 11 12 22 11 10
6 7 6 7
6 7 6 7
G356
6 21 10 11 7
7; G456
6 21 10 11 7
7;
4 5 4 5
22 21 10 10 11 12
2 3 2 3
15 23 23 15 15 23
6 7 6 7
6 7 6 7
6 7 6 7
G16
6 15 10 23 7
7; G26
6 15 10 23 7
7;
6 7 6 7
4 5 4 5
15 23 23 23 23 23
2 3 2 3
15 15 15 23 15 15
6 7 6 7
6 7 6 7
6 7 6 7
G36
6 23 10 7;
23 7 G46
6 23 10 15 7
7;
6 7 6 7
4 5 4 5
23 23 23 23 23 23
Kirsch gradient filter 2 3 2 3
23 23 15 23 23 23
6 7 6 7
6 7 6 7
6 7 6 7
G56
6 23 10 15 7
7; G66
6 23 10 15 7
7;
6 7 6 7
4 5 4 5
23 23 15 23 15 15
2 3 2 3
23 23 23 23 23 23
6 7 6 7
6 7 6 7
6 7 6 7
G76
6 23 10 23 7
7; G86
6 15 10 23 7
7:
6 7 6 7
4 5 4 5
15 15 15 15 15 23

Haralick’s textural features (0 8, 45 8, 90 8, 135 8, mean)a

X
Contrast
i;j
ji2jj2 pði; jÞ
X
Correlation ijpði; jÞ2lx ly
i;j
rx ry
XN 21
Difference entropy 2 k50 pi2j ðkÞlog pi2j ðkÞ1e
XN 21
Difference variance k 2 pi2j ðkÞ
k50
X
Energy
i;j
pði; jÞ2
X
Entropy 2 i;j pði; jÞlog pði; jÞ1e 5HXY
Information measure of correlation 1 HXY 2HXY 1

maxðHX ;HY Þ
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Information measure of correlation 2 12e 22 ðHXY 2 2HXY Þ
Inverse difference moment X pði; jÞ
i;j 11ði2jÞ2
X2N
Sum average kpi1j ðkÞ
k52
X2N
Sum entropy 2 k52 pi1j ðkÞlog pi1j ðkÞ1e 5HX 1Y
X
Sum of squares
i;j
ð2i2lÞ2 pði; jÞ
X2N
Sum variance ðk2HX 1Y Þpi1j ðkÞ
k52
vðxÞ, a 939 pixels window centered on the pixel of interest x; uðyÞ value of the pixel y of the image u; pixel pði; jÞ; ði; jÞ th entry in a
normalized gray-tone spatial-dependence matrix, 5 P(i, j)/R; px, ith entry in the marginal-probability matrix obtained by summing the rows
XN
of pði; jÞ; 5 j51 pði; jÞ; Sobel and Kirsch images Gi are obtained by filtering of the initial image u by the convolution masks given
above; Gi 5Gi u where * is the convolution operator; N, number of distinct gray levels in the quantized image; lx ; ly ; rx ; ry means et
standard deviations of the partial probability densities px et py ; pi1j (and pi2j , respectively) is the sum of the probabilities pði; jÞ of which
the sum (and the difference, respectively) of the coordinates
is i1 j (and i 2 j, respectively); H XY and HY design
entropies of the partial
X X
probability densities px et py; HXY 1 5 i;j pði; jÞlog px ðiÞpy ðjÞ and HXY 1 5 i;j px ðiÞpx ðjÞlog px ðiÞpy ðjÞ are two estimations of joint
entropy. e > 0 is the infinitesimal value to guaranty the strict positivity of the log argument.
a
Haralick’s textural features has been calculated in the four directions corresponding to the four angles (0 8, 45 8, 90 8, 135 8). The
average value of the four angle values has been also calculated.
MRI Feature Extraction entropy, energy, entropy, difference variance, difference entropy,
Four kinds of features (first-order statistics, gradient-based features, and two different information measures of correlation (Table 3). T3
second-order Haralick textures, and geometrical measurements) for We also calculated the average value of the four angle values (h).
a total of 144 features were computed in each T2-w and ADC Haralick textural features characterize homogeneity, gray level tran-
slides (Fig. 2). The first-order features, over a local 9 3 9 voxel sitions, and the presence of organized structures. The number of
window centered on the voxel of interest, included the local mean, gray-levels in the GLCM was set to eight, in line with a previous
median, and standard deviation signal intensity. Gradient-based study.13 A fourth set of features consisted of four geometric tumor
features were based on the gradient magnitude of the Sobel and characteristics on T2-w, namely, 2D parameters among the trans-
Kirsch filters, ie, an index of how different the voxel was from its verse, sagittal, and coronal plans (maximal tumor diameter, perime-
neighbors. The second-order of Haralick textural features consisted ter, and area), and the tumor volume. We applied a spherical
F3 of a voxel’s relationship with its neighbors (Fig. 3). Contrary to the approach to compute the 2D parameters, since a tumor’s shape can
first-order metrics, these consider the relationship between pairs of be compared to a sphere.24 All feature extraction methods were
neighboring voxels in the image. These features were derived from implemented using MatLab (MathWorks, Natick, MA), with the
the gray level co-occurrence matrix (GLCM), with a spatial rela- results presented in Table 3.
tionship defined as the relative direction h (limited to 08, 458, 908,
and 1358, in this study). A total of 130 Haralick features11 were Statistical Analysis
then extracted, namely: contrast, correlation, sum of squares, Correlations between extracted features and Gleason scores were
inverse difference moment, sum average, sum variance, sum investigated using the Spearman’s rank correlation coefficient.
January 2017 109

TABLE 4. Correlation Between MRI Features and Gleason Score (Significant Features Only)
MRI modalities Features rho coefficient P-valuea
T2-weighted Difference Entropy 90 8 20.236 0.04

Difference Variance 45 8 20.250 0.03
Difference Variance 90 8 20.246 0.03
ADC map Contrast 45 8 20.251 0.04
ADC, apparent diffusion coefficient.
a
Spearman test.
The association between MRI features and the biochemical sular extension, Gleason score, nodal status, D’Amico risk groups,
recurrence risk was assessed using the Cox proportional hazards androgen deprivation therapy, dose of radiotherapy, type of MRI
model in univariate analysis on the entire cohort, then in the high- scanner, and age) were tested.
risk tumor subgroup. In order to derive comprehensible hazard Harrell’s C-index was also used to quantify the association of
ratios (HRs) and CI, values of textural features were standardized each feature to establish a ranking between them29 and enable
as follows: Y 5 X 2l
r , where X is the initial value of the feature, l comparison of all the investigated methods.
the mean, and r the standard deviation of the feature among all Kaplan–Meier biochemical recurrence estimates for three of
patients. We were unable to perform a Cox proportional hazards the most biochemical recurrence-associated parameters are pre-
model in multivariate analysis due to the low number of patients sented in graphs. For each parameter, the median value was chosen
and recurrence events; hence, our choice to also exploit random in order to divide the entire cohort into two parts. Statistical sig-
survival forests (RSF) based on approximate log-rank splitting nificance between curves was assessed using the log-rank test.
rules.25–27 RSF is particularly adapted when the number of varia- All calculations were performed using R software (R Founda-
bles is bigger than the number of events. The RSF model is based tion for Statistical Computing, Vienna, Austria) and the Statistical
on prediction trees for analyzing right-censored survival data. In Package for the Social Sciences, v. 20.0 (SPSS, Chicago, IL).
RSF, randomization is introduced in two forms: 1) a randomly
selected bootstrap sample for growing the tree; 2) splitting the root Results
into two daughter nodes using a splitting rule (log-rank, in this Correlation Between MRI Features and Gleason
study) on a randomly selected covariate. The split is most effective
Score
when the survival difference between the daughter nodes is maxi- Table 4 presents Spearman’s correlation coefficient for the T4
mized. Thus, by grouping hazard estimates from terminal nodes, a
three T2-w features and the only ADC feature found to be
cumulative hazard function for the tree can be calculated. In this
significantly (P < 0.05) associated with Gleason score.
study, 2000 trees were grown in a preceding trial to generate the
ensemble cumulative hazard. The quality of a predictive survival There was no significant correlation between geometrical,
model can be judged by its prediction error. Harrell’s concordance first-order T2-w, gradient-based T2-w, or ADC features.
index (C-index) can be used to determine the model’s quality. The
Association With Biochemical Recurrence
C-index ranges from 0.5 (noninformative feature) to 1.0 (perfect
Table 5 summarizes the significant parameters (P < 0.05) T5
association discrimination). The relevance of each model covariate
was determined by applying it to the data not used in its construc- involving biochemical recurrence risk resulting from the
tion, namely, the “out-of-bag” data (OOB data). High positive Cox proportional hazards model in univariate analysis. All
relevance values indicate informative covariates, whereas small posi- geometrical characteristics, nodal status, Gleason score, sem-
tive or negative relevance values indicate noninformative covariates. inal vesical invasion, Kirsch operators, and first-order char-
This OOB data thus enabled us to automatically range and select acteristics on both T2-w and ADC (mean, median, and
the more relevant features, which were then added to a new RSF. standard deviation) imaging were significantly associated
In order to optimize the RSF process, we used an approach, based with biochemical recurrence. By using the Cattell’s Scree
on the Cattell’s Scree test, that involves the integration of the first approach and the RSF method, we were able to select five
n-th higher predictive features according to the C-index value features that led to a model with a better association with
obtained in univariate Cox analysis.28 Thus, these n features were
biochemical recurrence with a C-index of 0.90 6 0.09 (100
integrated into the RSF model and the evolution of the C-index
iterations and 2000 trees per iteration). The five features
according to the number n of explored features was analyzed.
Afterwards, this C-index evolution was examined to select the fea-
were tumor volume, T2-w difference variance mean, T2-w
tures associated with the most significant break. The logic behind contrast mean, maximal tumor area, and ADC median.
this method is that this point divides the important or major fea- Their OOB data importance values according to RSF are
tures from the minor or trivial factors. shown in Fig. 4. Among the Haralick textural features, in F4
MRI features and clinical information (PSA rate, clinical T- the Cox regression, only the T2-w features were associated
stage, MRI T-stage, MRI seminal vesicles invasion, MRI extracap- with biochemical recurrence in the entire cohort.

TABLE 5. Significant Parameters Associated With Biochemical Recurrence Using Cox Proportional Hazards Model (Univariate Analysis)
Features Feature values Harrell’s Univariate Cox regression
January 2017
(mean 6 SD) C-index
P value HR 95% CI
Haralick features T2-w Contrast 458 2.47 6 0.36 0.82 <0.001 3.03 (1.66-5.52)
T2-w Difference Variance 458 0.027 6 0.002 0.82 0.001 0.30 (0.15-0.60)
T2-w Difference Entropy 458 1.29 6 0.05 0.80 0.001 3.01 (1.54-5.90)
T2-w Inverse Difference Moment 458 0.53 6 0.03 0.80 0.001 0.32 (0.16-0.62)
T2-w Inverse Difference Moment mean 0.57 6 0.03 0.79 0.003 0.33 (0.16-0.68)
a
T2-w Difference Variance mean 0.031 6 0.003 0.79 0.002 0.32 (0.16-0.66)
a
T2-w Contrast mean 2.07 6 0.35 0.78 0.001 2.80 (1.51-5.17)
T2-w Contrast 908 1.72 6 0.44 0.78 0.001 2.88 (1.51-5.50)
T2-w Contrast 1358 2.43 6 0.45 0.77 0.003 2.80 (1.41-5.58)
T2-w Information Measure Of Correlation 1 908 20.28 6 0.05 0.75 0.013 0.37 (0.17-0.81)
T2-w Information Measure Of Correlation 1 mean 20.26 6 0.04 0.75 0.025 2.58 (1.13-5.93)
T2-w Information Measure Of Correlation 2 mean 0.75 6 0.03 0.72 0.027 0.43 (0.21-0.91)
Stage:

T2-w Contrast 08 1.64 6 0.31 0.70 0.024 1.96 (1.09-3.52)
111
Page: 111
112
TABLE 5: Continued
Features Feature values Harrell’s Univariate Cox regression

(mean 6 SD) C-index
P value HR 95% CI

T2-w Sum Average 1358 7.97 6 0.50 0.66 0.049 1.94 (1.00-3.75)
Gradient-based features T2-w Kirsch 100.7 6 28.2 0.69 0.034 0.57 (0.33-0.96)
First-order features T2-w Standard Deviation 1523 6 900 0.76 0.011 0.30 (0.12-0.76)
T2-w Average 37.1 6 11.7 0.76 0.003 0.42 (0.24-0.75)
a
ADC Median 74.2 6 16.4 0.76 0.028 0.39 (0.17-0.91)
ADC Standard Deviation 6039 6 2565 0.76 0.036 0.29 (0.09-0.92)
ADC Average 74.7 6 16.2 0.75 0.028 0.39 (0.17-0.90)
T2-w Median 35.9 6 11.5 0.75 0.005 0.43 (0.24-0.78)
3 a
Geometrical features Tumor Volume (mm ) 3626 6 4694 0.79 <0.001 2.13 (1.44-3.17)
2 a
Maximal Tumor Area (mm ) 924 6 934 0.77 0.004 1.84 (1.21-2.78)
Maximal Tumor Diameter (mm) 42 6 21 0.75 0.006 2.22 (1.26-3.94)
Maximal Tumor Perimeter (mm) 132 6 89 0.75 0.002 1.74 (1.22-2.47)
Clinical parameters Gleason Score 0.69 0.036 2.68 (1.07-6.73)
Seminal Vesical Invasion 0.65 0.037 3.55 (1.08-11.65)
Nodal Status 0.59 0.047 3.85 (1.02-14.56)
HR, hazard ratio; CI, confidence interval.
Data are in arbitrary units unless otherwise specified.
Considering the high risk group only (n 5 42), the parameters that remained significant were nodal status, the four geometrical measurements and T2-w Haralick’s textural features (con-
trast, difference entropy, difference variance, inverse difference moment, information measure of correlation 1 and 2). angles are not specified here. The features are classified by order of
C-index within each type of features.
a
Features that were implemented in the RSF model (T2-w contrast mean, T2-w difference variance mean, tumor volume, maximal tumor area and ADC median).
Stage:

Volume 45, No. 1
Page: 112
FIGURE 4: Out-of-bag data covariate importance values according to random survival forest (RSF). The five most important fea-
tures (according to the Cattell Scree test) that led to a model with a better association with biochemical recurrence with a C-index
of 0.90 6 0.09 (1000 iterations and 2000 trees per iteration). The five features were tumor volume, T2-w difference variance
mean, T2-w contrast mean, maximal tumor area, and ADC median.
In the high-risk population, first-order descriptors of sion) to be significantly associated with biochemical
T2-w and ADC images were not significantly associated recurrence.
with biochemical recurrence. According to Harrell’s C- In both populations (entire and high-risk subgroups),
index, most of the T2-w Haralick textural features were the the type of MRI scan used was not associated with bio-
strongest parameters (identified in Table 5 for Cox regres- chemical recurrence.
FIGURE 5: Biochemical recurrence rate according to tumor volume (A), T2-w difference variance mean (B) and T2-w contrast mean
(C). For each parameter, the median value was chosen to divide the entire cohort into two parts. Except for tumor volume, values
are given in arbitrary units. Statistical significance between curves was determined using the log-rank test.
January 2017 113

Kaplan–Meier curves of three features associated in the study revealed that an appropriate determination of the fea-
F5 RSF model are presented in Fig. 5 (tumor volume, T2-w ture subset is thus a crucial step in having higher perform-
difference variance mean and T2-w contrast mean). ance while speeding up and simplifying the process. The
log-rank approximate approach offered the fastest computa-
Discussion tion, followed by the conservation of events rule. It is
Our study demonstrated that Haralick textural features, important to note that the RSF model, once it has been
derived primarily from T2-w MRI scanning, correlate with trained, can be stored in software (such as R) and shared
Gleason score and are associated with biochemical recur- among institutions to be used for any given patient.
rence following prostate RT. Conversely, the two D’Amico Gleason score is probably the most powerful prognos-
risk classification criteria, clinical T-stage and PSA, were not tic factor for prostate cancer.33 We thus chose to explore the
significantly associated with recurrence in our cohort. Fur- correlation between the imaging parameters and Gleason
thermore, several first-order, gradient-based, and geometrical score. Previous studies have demonstrated the capability of
features were also associated with biochemical recurrence, MRI quantification to be associated with Gleason score
although to a lesser extent. These results were also observed with first-order descriptors, such as mean or median values,
within the high-risk tumor subgroup. both on T2-w34,35 and ADC images.36–40 Furthermore,
Haralick textural features exploit a wide range of fac- some of the Haralick textural features were also correlated
tors that statistically describe the image content, such as ori- with Gleason score in two recently published studies on T2-
ented information, energy, and disorder, among others. w and ADC images.20,21
These features are well suited to describe fine textures that Even though D’Amico risk groups of patients are
contain no obvious regularities regarding spatial arrange- clearly established, they are heterogeneous in terms of recur-
ments and patterns compared to expected MRI content.30 rence risk, highlighting our need to identify new independ-
Texture representation depends on the relationship between ent predictors within each risk group. Association between
the signal and noise within the image. Depending on the MRI and recurrence following prostate cancer radiotherapy
prostate region and imaging modality, different textural fea- have been identified in four studies6–9 corresponding to
tures seem to be relevant for the clinical endpoint of bio- qualitative parameters (seminal vesicle invasion, extracapsu-
chemical recurrence. Texture analysis algorithms were lar extension) or simple quantitative parameters (tumor size
initially designed to compute textural properties in a 2D or malignant metabolic tumor volume on spectroscopic
image space. However, given the evolution of imaging tech- MRI). The first study, conducted by Nguyen et al in a
nologies, new types of image data with volumetric character- cohort of 250 patients with clinical T1- or T2-stage prostate
istics have emerged. Directly applying traditional 2D texture cancer, found that endorectal MRI evidence of seminal vesi-
analysis algorithms to these new imaging data types is insuf- cle invasion was significantly associated with biochemical
ficient to fully explore 3D textural features in the volumetric recurrence in multivariate Cox analysis.6 The second study,
datasets. To address this issue, several studies have extended carried out by McKenna et al, analyzed a cohort of 80
the 2D GLCM texture computation into a 3D form.16,31,32 patients with localized or locally advanced prostate cancer,
Concerning the actuarial analysis, on the one hand, reporting that the mean diameter of extracapsular extension
the Cox model clearly represented the impact of each cova- at MRI (range: 1–22 mm) prior to RT was the sole inde-
riate on overall survival and its relationship with other cova- pendent variable, among clinical, biological, and pathologi-
riates (in a multivariate analysis); on the other hand, the cal parameters, that was associated with metastatic
RSF approach was not able to provide hazard ratios and recurrence (HR: 2.06; P 5 0.007).7 The third study was car-
associated p values, and may be considered a “black box” ried out by Joseph et al and involved a cohort of 67 patients
when interpreting the model, due to the large number of with localized or locally advanced prostate cancer, revealing
generated trees (2000 trees) and the number of iterations that the volume of malignant metabolism on spectroscopic
(100 iterations). However, the advantage of RSF is that it is MRI (range: 0.12-8.46 mm3) was the only variable showing
much better suited to automation of survival analysis than a an association with biochemical recurrence in multivariate
Cox model, as it requires less input from the user in data analysis (HR: 1.63; P 5 0.028).8 The two independent vari-
settings where covariates are highly intercorrelated. We con- ables associated with metastatic failure were MRI tumor vol-
sider RSF method to be a well-adapted explorative “data- ume (range: 0–13.48 cm3) (HR: 1.34; P 5 0.028) and
mining” tool for large datasets with covariates that have an seminal vesicle invasion on MRI (HR: 28.05; P 5 0.0008).
as-yet undetermined effect on the model output variable. The fourth study, by Riaz et al, analyzed a cohort of 279
The RSF approach using five selected features (according to patients with intermediate- or high-risk prostate cancer, all
the Cattell’s Scree test) did, in fact, demonstrate a higher having undergone endorectal MRI prior to brachytherapy
performance than the approach using all available informa- and RT.9 In their multivariate analysis, only extracapsular
tion and a Cox model with a C-index of 0.90 6 0.09. Our extension (HR: 3.79; P 5 0.003) and Gleason score (HR:

3.57; P 5 0.002) were independently associated with recur- parameters, we were able to identify subgroups of tumors
rence. Our study produced similar results for geometrical with unfavorable prognosis inside the D’Amico high-risk
characteristics, as we identified four parameters related to group population who were likely to be good candidates for
tumor size (diameter, perimeter, area, and volume) that were treatment intensification, such as RT dose escalation or
significantly associated with biochemical recurrence. We chemotherapy.
found the seminal vesicle invasion to be indicative of an The vast majority (100%) of Haralick parameters,
increased risk of biochemical recurrence within our cohort including the most associated with biochemical recurrence,
of intermediate- and high-risk tumor patients, but not the were found in T2-w images on the Cox regression. We
extracapsular extension. In conclusion, these four studies, found that both normalized (on the urine signal) and non-
along with our own,6–9 demonstrate that geometrical meas- normalized ADC Haralick parameters were poorly associ-
urements and qualitative parameters are indeed strongly ated with biochemical recurrence. Moreover, the low
associated with recurrence following prostate RT. resolution of the ADC images compared to the T2-w images
Haralick textural features from MRI in the field of may have impacted the textural information contained in
cancer have so far only been investigated for assessing tumor the Haralick parameters. The co-occurrence distance infor-
response in a few studies involving lymphoma, breast, gli- mation required in order to compute GLCM should be
oma, sarcoma and nasopharyngeal cancers.12–17 Harrison adjusted for ADC images. These results suggest that this
et al studied 19 patients with non-Hodgkin’s lymphoma modality could be analyzed using more specific texture
who had undergone MRI before and during chemother- descriptors that take into account other issues, such as orien-
apy.12 Texture parameters (eight sorts of Haralick features tation and multiple resolution. The T2-w modality offers
with histogram, run-length matrix, absolute gradient, and the highest resolution, and is therefore richest in texture
autoregressive model parameters) were shown to also be use- information.
ful for chemotherapy response monitoring. Ahmed et al Our study had several limitations. First, it was retro-
explored all Haralick parameters on 100 breast cancer spective in design, incurring a possible recruitment bias. All
patients who underwent preneoadjuvant chemotherapy the patients, however, did undergo PSA level analysis every
MRI.13 They found significant differences between 6 months during follow-up. Furthermore, we did not
responders and nonresponders to chemotherapy for contrast, include transitional zone tumors, even if MRI computer-
sum of squares, difference variance, sum entropy, cluster aided-diagnosis systems could be an essential tool to diag-
shade, and cluster prominence. Michoux et al studied nine nose these kind of tumors, frequently missed by ultrasound-
Haralick parameters (among other textural features derived guided biopsies. Second, the MRI database contains images
from the run length matrix) on 69 breast cancer patients from two different MRI scanners, consequently providing
who underwent preneoadjuvant chemotherapy MRI.41 different image properties. However, the choice of MRI
Energy, entropy, contrast, inverse difference moment, and scanner was not found to impact the risk of recurrence,
difference variance significantly differed between responders either for the entire cohort or the high-risk subgroup. More-
and nonresponders to chemotherapy. Yang et al and Upad- over, a normalization step was applied to both T2-w and
haya et al explored a cohort of 83 and 40 patients, respec- ADC voxels in order to decrease signal inhomogeneity, lim-
tively, who underwent prechemoradiotherapy multimodal iting the potential impact of scanner choice on the feature
MRI for glioblastoma.15,17 Both studies showed that textural computation. In any case, Haralick textural features reflect
features (Haralick’s among others) were potent prognostic the relative spatial distribution of neighboring pixels. In
factors of survival. Liu et al prospectively studied 20 Hara- addition, two studies by Mayerhoefer et al and one by
lick parameters among 126 textural features found on pre- Herlidou-Même et al demonstrated MRI protocol analysis
chemoradiotherapy MRI scans of 53 patients with to only minimally impact Haralick textural analysis.42–44
nasopharyngeal carcinoma.14 The authors built three predic- The third limitation lies in our capacity to only perform
tive models with selected parameters on T1-w, T2-w, and univariate Cox regression due to the limited number of
diffusion-w images. Our study is the first to demonstrate a recurrence events not allowing for a multivariate analysis
significant relationship between Haralick textural features (classically, one tested parameter per 10 events).45 However,
from MRI and biochemical recurrence of prostate cancer. we provided both the P value and Harrell’s C-index, the lat-
Most studies, including ours, reported to be informative in ter notably enabling comparison of the association capabil-
associating treatment response (for lymphoma, breast, and ities of the tested parameters in univariate analysis.
nasopharyngeal cancers) or biochemical failure (for prostate Moreover, we were able to identify new MRI-based parame-
cancer) difference variance and contrast. To a lesser extent, ters within a given risk group (high risk). To tackle the issue
inverse difference moment, sum of squares, difference of multivariate analysis, we proposed using random survival
entropy, and sum average also proved potential associations forests, which resulted in an improvement of the C-index
with recurrence in three studies, including ours. Using these values. Nevertheless, a larger cohort is clearly required to
January 2017 115

confirm the prognostic value of these parameters, particu- 4. Barentsz JO, Weinreb JC, Verma S, et al. Synopsis of the PI-RADS v2
guidelines for multiparametric prostate magnetic resonance imaging
larly the Gleason score. A fourth limitation was our consid- and recommendations for use. Eur Urol 2016;69:41–49.
eration of only one value (eight gray levels) in the 5. Hamoen EHJ, de Rooij M, Witjes JA, Barentsz JO, Rovers MM. Use of
quantization involved in the GLCM generation. In addition, the Prostate Imaging Reporting and Data System (PI-RADS) for pros-
tate cancer detection with multiparametric magnetic resonance imag-
the impact of varying the number of gray levels employed ing: a diagnostic meta-analysis. Eur Urol 2015;67:1112–1121.
needs to be investigated. It should, however, be noted that
6. Nguyen PL, Whittington R, Koo S, et al. Quantifying the impact of
this issue was explored in another study assessing the effi- seminal vesicle invasion identified using endorectal magnetic reso-
cacy of an MRI textural analysis in the associated response nance imaging on PSA outcome after radiation therapy for patients
with clinically localized prostate cancer. Int J Radiat Oncol 2004;59:
to chemotherapy in breast cancer. Those authors suggested 400–405.
that reducing the number of gray levels from 64 to 8 had a 7. McKenna DA, Coakley FV, Westphalen AC, et al. Prostate cancer: role
minimal effect on the computed feature values and their of pre-treatment MR in predicting outcome after external-beam radia-
tion therapy-initial experience. Radiology 2008;247:141–146.
ability to be associated with response to treatment.13 Finally,
8. Joseph T, McKenna DA, Westphalen AC, et al. Pretreatment endorec-
we did not study any other texture-based features except tal magnetic resonance imaging and magnetic resonance spectro-
Haralick features. Fractal analysis,46 discrete cosine trans- scopic imaging features of prostate cancer as predictors of response
to external beam radiotherapy. Int J Radiat Oncol 2009;73:665–671.
form (DCT),47 Gabor and Haar wavelet-based features18
have already been studied in previous trials in order to 9. Riaz N, Afaq A, Akin O, et al. Pretreatment endorectal coil magnetic
resonance imaging findings predict biochemical tumor control in pros-
improve prostate cancer characterization. Moreover, multi- tate cancer patients treated with combination brachytherapy and
parametric prostate MRI includes other possible sequences, external-beam radiotherapy. Int J Radiat Oncol 2012;84:707–711.
such as diffusion-weighted, DCE perfusion, or spectroscopic 10. Ganeshan B, Goh V, Mandeville HC, Ng QS, Hoskin PJ, Miles KA.
Non-small cell lung cancer: histopathologic correlates for texture
images, which have not been explored in our study. parameters at CT. Radiology 2013;266:326–336.
In conclusion, MRI T2-w Haralick textural features, 11. Haralick RM, Shanmugam K, Dinstein I. Textural features for image
along with geometrical parameters, were shown to be classification. IEEE Trans Syst Man Cybern 1973;SMC–3:610-621.
strongly associated with biochemical recurrence following 12. Harrison LC, Luukkaala T, Pertovaara H, et al. Non-Hodgkin lym-
prostate cancer radiotherapy, particularly in the high-risk phoma response evaluation with MRI texture classification. J Exp Clin
Cancer Res 2009;28:1.
prostate cancer group. All these parameters could aid the cli-
13. Ahmed A, Gibbs P, Pickles M, Turnbull L. Texture analysis in assess-
nician in making decisions in terms of intensifying or ment and prediction of chemotherapy response in breast cancer.
diminishing treatments (such as androgen deprivation, J Magn Reson Imaging 2013;8:89–101.
chemotherapy, and radiotherapy dose modifications). Future 14. Liu J, Mao Y, Li Z, et al. Use of texture analysis based on contrast-
enhanced MRI to predict treatment response to chemoradiotherapy
studies must now be conducted involving a larger number in nasopharyngeal carcinoma. J Magn Reson Imaging 2016.
of patients with prebiopsy MRI. 15. Yang D, Rao G, Martinez J, Veeraraghavan A, Rao A. Evaluation of
tumor-derived MRI-texture features for discrimination of molecular
subtypes and prediction of 12-month survival status in glioblastoma.
Med Phys 2015;42:6725–6735.
Acknowledgments 16. Vallieres M, Freeman CR, Skamene SR, Naqa IE. A radiomics model
Contract grant sponsor: “Region Bretagne,” the regional from joint FDG-PET and MRI texture features for the prediction of
lung metastases in soft-tissue sarcomas of the extremities. Phys Med
government, and a French government grant by way of the Biol 2015;60:5471.
CominLabs excellence laboratory, managed by the National 17. Upadhaya T, Morvan Y, Stindel E, Le Reste P-J, Hatt M. A framework
Research Agency in the “Investing for the Future” program; for multimodal imaging-based prognostic model building: preliminary
contract grant number: ANR-10-LABX-07-01.11 study on multimodal MRI in glioblastoma multiforme. IRBM 2015;36:
345–350.
We thank Geoffrey Roman Jimenez, William Komegni, and
18. Viswanath SE, Bloch NB, Chappelow JC, et al. Central gland and
Thibault Madouasse for contributions to this study. peripheral zone prostate tumors have significantly different quantitative
imaging signatures on 3 Tesla endorectal, in vivo T2-weighted mag-
netic resonance imagery. J Magn Reson Imaging 2012;36:213–224.
19. Duda D, Kretowski M, Mathieu R, Crevoisier R de, Bezy-Wendling J.

Multi-image texture analysis in classification of prostatic tissues from
References MRI. Preliminary results. In: Pi^
etka E, Kawa J, Wieclawek W, editors.
1. D’Amico AV, Whittington R, Malkowicz S, et al. Biochemical outcome Information technology and biomedicine, vol. 3. Springer International
after radical prostatectomy, external beam radiation therapy, or inter- Publishing; 2014:139–150 [Advances in intelligent systems and com-
stitial radiation therapy for clinically localized prostate cancer. JAMA puting, vol. 283].
1998;280:969–974.
20. Vignati A, Mazzetti S, Giannini V, et al. Texture features on T2-
2. Cahlon O, Hunt M, Zelefsky MJ. Intensity-modulated radiation ther- weighted magnetic resonance imaging: new potential biomarkers for
apy: supportive data for prostate cancer. Semin Radiat Oncol 2008; prostate cancer aggressiveness. Phys Med Biol 2015;0:2685.
18:48–57.
21. Wibmer A, Hricak H, Gondo T, et al. Haralick texture analysis of pros-
3. Lambin P, Rios-Velazquez E, Leijenaar R, et al. Radiomics: extracting tate MRI: utility for differentiating non–cancerous prostate from pros-
more information from medical images using advanced feature analy- tate cancer and differentiating prostate cancers with different Gleason
sis. Eur J Cancer 2012;48:441–446. scores. Eur Radiol 2015:1-11.

22. Abramowitz MC, Li T, Buyyounouski MK, et al. The Phoenix definition 36. Vargas HA, Akin O, Franiel T, et al. Diffusion-weighted endorectal MR
of biochemical failure predicts for overall survival in patients with imaging at 3 T for prostate cancer: tumor detection and assessment
prostate cancer. Cancer 2008;112:55–60. of aggressiveness. Radiology 2011;259:775–784.
23. Itatani R, Namimoto T, Yoshimura A, et al. Clinical utility of the nor- 37. Turkbey B, Shah VP, Pang Y, et al. Is apparent diffusion coefficient
malized apparent diffusion coefficient for preoperative evaluation of associated with clinical risk scores for prostate cancers that are visible
the aggressiveness of prostate cancer. Jpn J Radiol 2014;32:685–691. on 3-T MR images? Radiology 2011;258:488–495.
24. Delgado-Gonzalo R, Thevenaz P, Seelamantula CS, Unser M. Snakes 38. Oto A, Yang C, Kayhan A, et al. Diffusion-weighted and dynamic
with an ellipse-reproducing property. IEEE Trans Image Process 2012; contrast-enhanced MRI of prostate cancer: correlation of quantitative
21:1258–1271. MR parameters with gleason score and tumor angiogenesis. Am J
Roentgenol 2011;197:1382–1390.
25. Breiman L. Random forests. Mach Learn 2001;45:5–32.
39. Hambrock T, Somford DM, Huisman HJ, et al. Relationship between
26. Ishwaran H, Kogalur UB, Blackstone EH, Lauer MS. Random survival
apparent diffusion coefficients at 3.0-T MR imaging and Gleason
forests. Ann Appl Stat 2008;2:841–860.
grade in peripheral zone prostate cancer. Radiology 2011;259:453–
27. Datema FR, Moya A, Krause P, et al. Novel head and neck cancer sur- 461.
vival analysis approach: random survival forests versus cox propor-
40. Peng Y, Jiang Y, Yang C, et al. Quantitative analysis of multipara-
tional hazards regression. Head Neck 2012;34:50–58.
metric prostate MR images: differentiation between prostate cancer
28. Jolliffe I. Principal component analysis. In: Everitt BS, Howell DC, edi- and normal tissue and correlation with Gleason score — a
tors. Encyclopedia of statistical behavior science. Chichester, UK: computer-aided diagnosis development study. Radiology 2013;267:
John Wiley & Sons; 2005. 787–796.
29. Harrell FE, Lee KL, Mark DB. Multivariable prognostic models: issues 41. Michoux N, Van den Broeck S, Lacoste L, et al. Texture analysis on
in developing models, evaluating assumptions and adequacy, and MR images helps predicting non-response to NAC in breast cancer.
measuring and reducing errors. Stat Med 1996;15:361–387. BMC Cancer 2015;15:574.
30. Lerski RA, Straughan K, Schad LR, Boyce D, Bl€ uml S, Zuna I. MR 42. Herlidou-Même S, Constans JM, Carsin B, et al. MRI texture analysis
image texture analysis—an approach to tissue characterization. Magn on texture test objects, normal brain and intracranial tumors. Magn
Reson Imaging 1993;11:873–887. Reson Imaging 2003;21:989–993.
31. Tsai F, Chang C-K, Rau J-Y, Lin T-H, Liu G-R. 3D computation of 43. Mayerhoefer ME, Breitenseher MJ, Kramer J, Aigner N, Hofmann S,
gray level co-occurrence in hyperspectral image cubes. In: Yuille AL, Materka A. Texture analysis for tissue discrimination on T1-weighted
Zhu S-C, Cremers D, Wang Y, editors. Energy minimization methods in MR images of the knee joint in a multicenter study: transferability of
computer visualizaion pattern recognition. Berlin, Heidelberg: Springer; texture features and comparison of feature selection methods and
2007:429–440 [Lecture Notes in Computer Science, vol. 4679]. classifiers. J Magn Reson Imaging 2005;22:674–680.
32. Hatt M, Majdoub M, Vallières M, et al. 18F-FDG PET uptake charac- 44. Mayerhoefer ME, Szomolanyi P, Jirak D, Materka A, Trattnig S.
terization through texture analysis: investigating the complementary Effects of MRI acquisition parameter variations and protocol het-
nature of heterogeneity and functional tumor volume in a multi- erogeneity on the results of texture analysis and pattern discrimi-
cancer site patient cohort. J Nucl Med 2015;56:38–44. nation: an application-oriented study. Med Phys 2009;36:1236–
1243.
33. D’Amico AV, Chen M-H, Renshaw AA, Loffredo B, Kantoff PW. Risk of
prostate cancer recurrence in men treated with radiation alone or in 45. Peduzzi P, Concato J, Kemper E, Holford TR, Feinstein AR. A simula-
conjunction with combined or less than combined androgen suppres- tion study of the number of events per variable in logistic regression
sion therapy. J Clin Oncol 2008;26:2979–2983. analysis. J Clin Epidemiol 1996;49:1373–1379.
34. Rosenkrantz AB, Kopec M, Kong X, et al. Prostate cancer vs. post- 46. Lv D, Guo X, Wang X, Zhang J, Fang J. Computerized characteriza-
biopsy hemorrhage: Diagnosis with T2- and diffusion-weighted imag- tion of prostate cancer by fractal analysis in MR images. J Magn
ing. J Magn Reson Imaging 2010;31:1387–1394. Reson Imaging JMRI 2009;30:161–168.
35. Giannini V, Vignati A, Mirasole S, et al. MR-T2-weighted signal inten- 47. Niaf E, Rouvière O, Mège-Lechevallier F, Bratan F, Lartizien C. Com-
sity: a new imaging biomarker of prostate cancer aggressiveness. puter-aided diagnosis of prostate cancer in the peripheral zone using
Comput Methods Biomech Biomed Eng Imaging Vis 2014;0:1–5. multiparametric MRI. Phys Med Biol 2012;57:3833–3851.
January 2017 117

ORIGINAL RESEARCH
Incremental Value of MRI for

Preoperative Penile Cancer Staging
ao Lucchesi MD, PhD,1* Rodoldo Borges Reis MD, PhD,2
Fabiano Rubi~
Eliney Ferreira Faria MD, PhD,3 Roberto Dias Machado MD,3
Rodrigo Ribeiro Rossini MD,1 Leonardo D. Borregales MD,4
Gyl Eanes Barros Silva MD, PhD,5 and Valdair Francisco Muglia MD, PhD6
Purpose: To evaluate the incremental value of magnetic resonance imaging (MRI), compared to clinical examination, for
penile cancer (PC) local staging.
Materials and Methods: Twenty-five consecutive patients with histologically proven PC were evaluated prospectively.
MRI staging was performed on 1.5 and 3.0T scanners using high-resolution T2-weighted and postcontrast T1-weighted
images. Two blinded observers interpreted MR images. Clinical local staging was performed by experienced urologists.
The pathology report was used as the standard of reference.
Results: The interobserver agreement for MRI staging, using a kappa test for T-staging was moderate, 0.52 (95% confi-
dence interval [CI] 5 0.24–0.78), P 5 0.001, although a high correlation for N-staging, 0.72 (95% CI 5 0.42–1.00),
P 5 0.001, was detected. Clinical staging was correct in 52.0% (13/25) of patients. After pathological staging, five
(20.0%) lesions were upstaged and seven (28.0%) lesions were downstaged compared to clinical examination. MRI accu-
rately defined T-staging in 18/25 lesions (72.0%). After pathologic staging, five (20.0%) were upstaged and two down-
staged (8.0%), compared to MRI. Fifteen patients were submitted to inguinal and pelvic lymphadenectomy and
considered for comparison of accuracy of nodal staging by physical examination and MRI. Clinical staging accurately
staged 7/15 patients (46.7%). After histopathologic analysis, six cases had nodal staging upgraded and two cases were
downgraded. MRI correctly staged 13/15 (86.7%). Using a chi-square for comparison, differences in proportion of cor-
rected staging between clinical examination and MRI were not significant for T-staging (P 5 0.14), but were significant
for nodal staging (P 5 0.02).
Conclusion: According to our results, MRI improves local staging of PC patients, particularly for those with limited phys-
ical examination.
P enile cancer is a rare disease in developed countries,

with an incidence in the United States for 2015 of
1820 cases with 310 related deaths.1 However, incidence in
also detected. The cellular type is fundamental, since impor-
tant differences in outcomes are observed according to histo-
logic subtypes.2
emergent countries is higher,2,3 arguably related to differen- Accurate staging and grading of tumors is a key point,
ces in socioeconomic aspects, religious conditions, and the since it is essential to define clinical management and
prevalence of the human papilloma virus.4,5 follow-up, and also to predict outcome. Today, TNM stag-
The majority of penile cancers are squamous cell carci- ing system is an important tool for risk stratification. Inva-
nomas (>95%), affecting mainly the glans, but other histo- sion of subepithelial connective tissue, corpus spongiosum
logic types as melanocytic lesions, mesenchymal and and/or corpora cavernosa, urethra and adjacent structures
neuroendrocrine tumors, lymphomas, and metastases are are hallmarks for "T" classification. Histologic features such
*Address reprint requests to: F.R.L., Department of Radiology, Barretos Cancer Hospital, Pio XII Foundation, Rua Antenor Duarte Villela, 1331 Barretos, SP,
Brazil 14784-400. E-mail: fabianolucchesi@hotmail.com
From the 1Department of Radiology, Barretos Cancer Hospital, Pio XII Foundation, Barretos, Brazil; 2Department of Surgery, Urology Division, Ribeirao
Preto Medical School, University of Sao Paulo, Brazil; 3Department of Urology, Barretos Cancer Hospital, Pio XII Foundation, Barretos, Brazil; 4Department
of Urology, University of Texas MD Anderson Cancer Center, Houston, Texas, USA; 5Department of Pathology, Ribeirao Preto Medical School, University of
Sao Paulo, Brazil; and 6Internal Medicine Department, Imaging Division, Ribeirao Preto Medical School, University of Sao Paulo, Brazil

V

Lucchesi et al.: MRI in Penile Cancer Staging
TABLE 1. MRI Protocol Parameters
TR (msec) TE (msec) ST (mm) FOV (mm) MATRIX NEX
Axial T2 5645/5400 90/100 3.0 168x168 170x170 576x576 540x540 2/3

Axial T2 pelvis 2500/2200 80 5.0 320x380 320x360 476x421 420x380 1
Sag T2 3679/2950 100 4.0 240x240 290x240 416x327 398x318 2
Cor T2 5375/4050 100 3.0/4.0 170x180 200x180 272x189 300x192 2
Axial T1 756/666 24/12 3.0/4.0 1803180 2003180 2523208 2
First parameter (before slash) is from 3T scanner. Second parameter is from 1.5 scanners. TR, repetition time; TE, echo time; SL,
slice thickness; FOV, field of view; NEX, number of excitations.
as lymphovascular invasion and histologic grade are important The inclusion criteria were: 1) clinical suspicion of penile
prognostic factors. Nonetheless, the presence of nodal involve- cancer, confirmed by posterior histopathological study; and 2)
ment is the single most important prognostic factor for out- those who underwent an MRI study (with dedicated protocol)
come.6–10 Particularly, extracapsular extension is considered a prior to surgery. Exclusion criteria included: 1) any procedure
factor of poor prognosis and is staged as N3, while retroperi- involving index lesion or lymph nodes prior to MRI; 2) interval
greater than 4 months between the MRI and surgery; and 3) sub-
toneal nodal involvement indicates distant metastases.10
optimal or incomplete MRI. In this period, 41 patients were diag-
Current clinical staging is based on physical examina-
nosed with penile cancer in both institutions. Nine were excluded
tion and provides inaccurate information in up to 25% of
due to suboptimal or no MRI, two patients had sentinel lymph
patients,9 with a higher proportion in patients with second- node biopsy prior to MRI, two patients had surgery 4 months after
ary infection in the tumor, when edema and inflammatory MRI, and three did not undergo a surgical procedure. After exclu-
endurance may mimic neoplastic infiltration.11 Clinical stag- sions, 25 patients were enrolled with an average age of 58.9 years
ing in obese patients can be a challenge due to an excess of (ranging from 43 to 89).
adipose tissue, which makes inguinal nodes evaluation
extremely difficult. Enlarged inguinal lymph nodes can be Imaging
identified in about 30–60% of the patients with penile can- All MRI exams were analyzed in PACS (picture archiving and
communication system) of the two participating institutions. Penile
cer at the time of diagnosis, but only about 50% will be
MRI studies were performed with two 1.5T scanners (Signa, GE
related to metastasis and the other half will be reactive ones
Medical Systems, Milwaukee, WI, or Achieva, Philips Healthcare,
due to infection.10,12–14 However, false-positive is not the Best, The Netherlands) and one 3T (Intera, Philips Healthcare)
unique limitation of clinical staging because 10–25% of whole-body MRI unit and using pelvic multichannel 16-channel,
patients with no palpable lymph node can harbor occult phased-array coil.
metastases.6,13,15 A similar examination protocol was used at both institutions
Magnetic resonance imaging (MRI) has not been fre- and included axial T1-weighted sequences and T2 of whole pelvis
quently used to evaluate penile cancer. The first reports and T2-weighted sequences in the sagittal, axial, and coronal using
dated from the early 1990s,16 but recently interest was penile position as a reference, with high-resolution with small
renewed, with more studies published.17–22 field-of-view. All examinations were performed with the penis in
Since there is a paucity of data in this issue, we con- flaccid status, and no pharmacologically induced erection was per-
ducted this study to evaluate the incremental value of MRI formed. After intravenous injection of gadolinium-based contrast
media, T1-weighted, volumetric acquisitions were obtained in the
for staging penile cancer, focusing in the local extent of the
short and long axis of penis and for the whole pelvis. Details of
tumor (T) and in locoregional lymph node (N) involvement
the MRI acquisition parameters are presented in Table 1. T1
and comparison to preoperative clinical staging.
MRI Analysis
Materials and Methods All studies were separately analyzed by two radiologists with experi-
Patients ence in oncologic abdominal radiology (VFM, 16 years and FRL,
This prospective study was conducted from April 2011 to July 12 years), blinded to clinical and pathologic results. Discrepant
2014, at two participating institutions, after approval by the Insti- results were reviewed to reach a consensus. Another radiologist
tutional Review Board, with the need for a signed informed con- (RRR, 8 years of experience) analyzed the surgery and pathology
sent. During this period, any patient with penile lesions suspected report, independently of the other two radiologists.
for cancer was assigned as a potential participant in the study and The primary penile lesion, pelvic, and inguinal lymph nodes
an MRI exam of the penile region, including when the whole pel- status were evaluated using the American Joint Committee on
vis was indicated. Cancer Tumor, Nodes, and Metastasis (TNM) system. On MRI,
January 2017 119

TABLE 2. Demographic Data

Results
For the primary lesion, the average size was 3.0 cm (mini-
Age (years) 58.9 (43–89) mum 1.0 and maximum 7.4 cm), measured in the most
favorable plane in MR images, according to lesion format
Lesion size (cm) 3.0 (1.0–7.4) and orientation.
Histology Squamous cell The mean interval between MRI and surgery was 30
carcinoma (all) days (median of 17), ranging from 2 to 123 days. All the
Mean/median of interval Mean 5 30 (2–123) penile cancers were squamous cell carcinomas. The demo-
MRI-surgery (days) Median 5 17 graphic characteristics of the studied population are
described in Table 2. T2
The interobserver agreement for MRI staging, for T-
penile cancer usually presents as solitary, irregular, and infiltrating staging was moderate, 0.52 (95% confidence interval
mass lesion, with low signal compared to corpus spongiosum or [CI] 5 0.24–0.78), P 5 0.001, and there was good correla-
cavernosum on T2-weighted images.17–19 The lesions were meas- tion for N-staging, 0.72 (95% CI 5 0.42–1.00), P 5 0.001.
ured in the most suitable plane. The MRI criteria for involvement
of these structures were disruption of low signal of albuginea and
clear extension to the urethra.
Pelvic lymph nodes were considered suspect when they met T-Staging
the following criteria: short axis greater than 7 mm, loss of elliptical During the office evaluation, the diagnostic accuracy of clin-
form changing to round shape, and/or loss of internal architecture. ical T-staging using physical examination alone was 52.0%,
as only 13 out of 25 were correctly evaluated. After patho-
Clinical Staging logical staging, five (20.0%) lesions were upstaged, includ-
Clinical staging was obtained by palpation of the primary lesion ing four patients with clinical staging T2 to T3 and two
and bilateral inguinal regions. It was performed at the time of the patients with T1 to T2; and seven (28.0%) lesions were
first clinical consultation by the urologic staff of two participating downstaged, including three patients with clinical staging
institutions, always with participation of a senior urologist of each T2 to T1; two T3 to T2 and two T3 to T1.
institution (RBR, 22 years of experience and EFF, with 12 years of MRI accurately defined T-staging (Fig. 1) in 18 out of F1
experience). 25 lesions (72.0%). After surgery, five (20.0%) had a higher
pathologic staging and two were downstaged (8.0%), com-
Surgical and Pathologic Analysis pared to MRI. Among patients whose lesions were upstaged,
The standard of reference for this study was a combination of sur- MRI classified all five patients as T1, but four were T2
gical and pathological features. Patients with nonpalpable lymph (Fig. 2) and one was T3 at pathologic staging. Of the two F2
nodes were categorized into low (Tis, Ta, T1a), intermediate patients who were downstaged, there was one lesion initially
(T1b), and high risk (any T2 or G3 or G4) to present nodal dis- staged as T2 and the other staged as T3 (Fig. 3) by MRI F3
ease. Modified bilateral inguinal lymph node dissection23 was indi- that were subsequently moved to T1, after final pathologic
cated for all patients in the intermediate and high-risk groups. All staging. Results for every patient are presented in Table 3. T3
patients with palpable inguinal lymph nodes underwent bilateral
inguinal lymphadenectomy. Pelvic lymph node dissection was rec-
ommended when at least two histologically proven inguinal node
metastases were detected.24 A suspicious node detected by MRI
was also an indication for inguinal lymphadenectomy. Surgical
specimens were examined by one dedicated pathologist (GEBS)
with 11 years of experience in urological oncology. The pathologist
provided a detailed description for T- and N-staging according to
the AJCC format.
Statistical Analysis FIGURE 1: A 57-year-old male presenting with penile and ingui-
Demographic data were provided as descriptive statistics. Accuracy nal masses. Clinical staging was T3, N2. A: Coronal T2-weighted
of clinical and MRI staging were compared using the chi-square imaging showing distal mass (*) in the penis, abutting, but not
test for comparison of proportions. violating the tunica albuginea (arrow), indicating a T1 stage,
confirmed after surgery. A large necrotic and ulcerated lymph
The rate of agreement among observers was evaluated using node (N) is seen in left inguinal region. B: Axial T2-weighted
a kappa test. Data were processed with SSPS software, v. 21 image confirms the same findings. After excision, a spoor-
(Chicago, IL) and the level of significance was set at 0.05. differentiated, squamous cell carcinoma was diagnosed.

FIGURE 2: A 63-year-old man presenting with a small mass in glans. A,B: Axial and coronal T2-weighted images showing an isoin-
tense mass to glans, just distal to balanopreputial sulcus (*), causing loss of the low signal of albuginea, but without signs of
extension to the glans (arrow). C: A postcontrast, coronal T1-weighted image confirming the enhancing, partially exophytic mass,
suggesting a stage T1 lesion. At surgery and histopathologic analysis, a T2 lesion was described as squamous cell carcinoma.
N-Staging gested N0 and N1, but MRI staged as N3 (Fig. 4) and F4

Of the 25 patients, only 15 were submitted to inguinal and both were confirmed as true-positive after the final pathol-
pelvic lymph node dissection and were considered for analy- ogy report.
sis of the MRI accuracy of lymph node staging. Using histo-
pathologic analysis as the standard of reference, clinical
Discussion
examination accurately staged 7 out of 15 patients (46.7%).
Penile cancer is a rare, but devastating neoplasia, associated
After histopathologic analysis, six cases (40%) had nodal
with high mortality rates. Surgery remains the best thera-
staging upgraded, including four N1 moved to N2, one N0
to N1, and one N1 to N3. Two cases were downgraded peutic option and is targeted to removal of primary lesions
(13.3%). as well as lymph node dissection. The goals of surgical ther-
MRI correctly staged 13 out of 15 (86.7%). After apy are lesion removal and organ sparing, to the extent
comparison of MRI results with final pathological report, possible.
two cases were downstaged (false-positive) and none As clinical assessment of the patient may be impaired
upstaged (false-negative). The two false-positive cases were by associated conditions such as superimposed infection or
due to reactive inguinal lymph nodes (N1 on MRI and N0 body habitus, imaging methods have gained acceptance for
at pathology). MRI missed one case of inguinal lymph node penile cancer staging. Our data showed a significant
involvement. No cases of pelvic lymph nodes involvement improvement for local staging of penile cancer when com-
(three cases) were missed by MRI and, in two cases, the sur- pared to clinical staging. And our results are consistent with
gical approach was changed, since clinical examination sug- previous reports in the literature, as in the studies of Scar-
dino et al,20 and Petraglia et al,21 and later, in 2008, also
FIGURE 3: An 82-year-old man presenting with a glans mass. A,B: Sagittal and coronal T2-weighted images showing a mass with
intermediate signal (*), apparently disrupting the low signal of albuginea (arrow). C: An adjacent sagittal image indicates involve-
ment of corpus spongiosum and urethra (U), suggesting a stage T3 lesion. After surgery and histopathologic analysis, a T1 lesion
was assigned for this squamous cell carcinoma.
January 2017 121

TABLE 3. Clinical, MRI, and Pathological T- and N-Staging For All Patients
Patient Age T-staging N-staging Size (cm)
CL MRI PAT CL MRI PAT
1 55 T3 T3 T3 N1 N3 N3 5.8
2 64 T2 T1 T3 N3 N3 N3 5.3
3 50 T1 T1 T1 Nx Nx Nx 1.5
4 74 T2 T1 T2 N0 N0 N0 1.9
5 53 T2 T1 T1 Nx Nx Nx 1.5
6 61 T2 T3 T3 N0 N1 N1 3.3
7 59 T3 T2 T2 Nx Nx Nx 3.3
8 57 T3 T3 T3 N1 N1 N0 2.9
9 57 T3 T1 T1 N3 N3 N3 4.7
10 47 T2 T3 T3 Nx Nx Nx 4.3
11 49 T2 T2 T2 Nx Nx Nx 6.0
12 51 T2 T2 T2 N1 N2 N2 3.7
13 89 T2 T2 T2 Nx Nx Nx 3.6
14 60 T2 T2 T2 N1 N1 N0 2.8
15 54 T3 T2 T2 N1 N1 N1 5.4
16 82 T3 T3 T1 Nx Nx Nx 3.9
17 55 T1 T1 T2 Nx Nx Nx 1.9
18 57 T2 T2 T1 N0 N0 N0 3.8
19 68 T2 T1 T1 N0 N0 N0 1,0
20 46 T2 T1 T2 Nx Nx Nx 2.0
21 81 T2 T2 T2 N0 N0 N0 3.2
22 43 T1 T1 T2 Nx Nx Nx 1.7
23 54 T3 T3 T3 N1 N2 N2 7.4
24 55 T2 T2 T2 N1 N2 N2 6.0
25 53 T3 T3 T3 N1 N2 N2 3.8
CL, clinical staging; MRI, MRI staging; Pat, pathological staging.
assessed the value of MRI for cancer local staging (T-stag- The intraoperative decision for primary tumor assess-
ing), aiming at the tendency of organ preservation. ment is usually based on physical examination findings,
Although both previous studies reported good accuracy which might be a useful method for estimating penile
(88.8 and 92.3%, respectively), in both the MRI was per- tumor size and also for predicting urethral and corpus caver-
formed after a pharmacologically induced penile erection nosum infiltration. However, clinical staging of the primary
and evaluated a limited group of patients (9 and 13 tumor can be incorrect in a significant proportion of
patients, respectively). Recently, another study assessed the patients, as in the study of Lont et al25 and ours. Clinical
accuracy of MRI for locoregional invasion of penile cancer, understaging may result from histologic infiltration not clin-
showing sensitivity and specificity for predicting invasion of ically evident and overstaging from edema and infection.26
urethra of 82.1%, 79.1%, and 62.5, 86.9%, respectively,22 These aspects affecting the therapeutic approach, conserva-
in a series of 100 patients. But none of these studies tive or more aggressive, may be improved by imaging.
assessed, either local and lymph nodes staging, in the same Aligned with our study, a recent series showed good accu-
group of patients. racy of MRI for detecting urethra and corpus cavernosum

FIGURE 4: A: Sagittal T2-weighted image of a 57-year-old male showing a mass in glans (*), disrupting the low signal of tunica
albuginea. B: Axial T2-weighted image demonstrated a large lymph node in right inguinal region, with areas of high signal inten-
sity suggestive of necrosis. By clinical staging it was a T3-N2 lesion. C: Axial T2-weighted image of the whole pelvis, showing an
enlarged and irregular lymph node in the external iliac site, with loss of internal architecture. After partial penectomy and lymph-
adenectomy, a T3-N3 lesion was confirmed.
involvement,22 using the same criteria as ours. Of the cases difficult, even in surgical specimens.30 Also, lymph nodes
that were upstaged by MRI in the present study, most with signs of extracapsular extension were considered N3,
lesions measured between 3.1 and 3.9 cm. The two lesions regardless of its location. Inguinal lymphadenopathy is rela-
downstaged after surgery and histopathological analysis, tively common in patients with penile cancer, either due to
compared to MRI, were located in glans and balanoprepu- true metastatic dissemination or due to reactive enlargement
tial sulcus, and presented with an infiltrative growth pattern. secondary to superimposed infection of the penis. Therefore,
Arguably, the ill-defined borders of these infiltrative tumors the false-positive rate for palpable lymph nodes is reported
may have induced the interpretation of more advanced to be as high as 50%.14,30 The false-negative rate is the
lesions. One case of supposed urethral invasion assigned by other great limitation of clinical staging, since palpation of
MRI was not confirmed at pathology, resulting in a false- the inguinal region may be difficult due to obesity and
positive finding (Fig. 3). Discrepant results are especially rel- edema.
evant in penile cancer cases, because in this situation MRI Although not previously considered for detection of
may induce more aggressive resection, with the inherent nodal involvement in penile cancer,10,31,32 our study sug-
implications. In five cases, MRI missed a more advanced gests a potential role of MRI for detection of inguinal and
stage. In four of them, lesions had large burden, measuring iliac lymph nodes extension in penile cancer.
up to 3.8 cm, and all were located in the distal penis and In our study, MRI performed significantly better than
glans, the most common site for penile cancer. palpation for nodal staging. Our results represent a signifi-
Although MRI for penile cancer has been described cant improvement over those number of Zhu et al,33 who,
since the early 1990s, only recently has the technique using computed tomography (CT) exams, found limited
evolved to a more standardized approach, with pelvic sensitivity, only 37.5%, although with 100% specificity. In
phased-array, with multiple channels coils, yielding high- addition to the prognostic factor of nodal involvement, the
resolution, high-quality images, essential for correct interpre- definition of involvement of iliac and other retroperitoneal
tation. Early studies used induced erection for MRI of pen- sites allow surgeons to plan a single surgical procedure that
ile cancer,20,21 while ours and the most recent one22 did will include primary lesion removal, inguinal and pelvic ret-
not, but a direct comparison between the accuracy of the roperitoneal dissection. We did not assess for the presence
two techniques is beyond the scope of this study. The only of extracapsular extension, since this finding has no imaging
possible conclusion is that MRI accuracy for T-staging of criterion defined, so far, for this condition.34
both studies22 without tumescence, so far, is comparable. Our study has several limitations. First, the number of
Nodal involvement is one of best predictors of penile patients was limited, but we have to consider that penile
cancer outcomes.27–29 Several studies have demonstrated cancer is a relatively rare disease in developed countries,
that overall survival correlates well with the number of although slightly more prevalent in emergent countries.
lymph nodes involved. The N-staging for penile cancer has Also, we limited our analysis of nodal staging to those who
been recently modified, with removal of the anatomical dis- underwent lymph node dissection, in order to have surgical
tinction for superficial and deep inguinal nodes, which were and pathological reports as the standard of reference. As
January 2017 123

only the more advanced cases are generally referred for 15. Barski D, Georgas E, Gerullis H, Ecke T. Metastatic penile carci-
noma—an update on the current diagnosis and treatment options.
lymph node dissection, this introduces a potential selection Cent Eur J Urol 2014;67:126–132.
bias. However, it would be difficult to gain approval from 16. Catalona W. Role of lymphadenectomy in carcinoma of the penis.
any IRB for a study in which inguinal and pelvic lymph Urol Clin N Am 1980;3:785–792.
node dissection were performed for all patients, even for 17. de Kerviler E, Ollier P, Desgrandchamps F, et al. Magnetic resonance
those with reduced risk for lymphadenopathy in penile can- imaging in patients with penile carcinoma. Br J Radiol 1995;68:704–711.
cer, as both are invasive procedures. 18. Vossough A, Pretorius ES, Siegelman ES, Ramchandani P, Banner MP.
Magnetic resonance imaging of the penis. Abdom Imaging 2002;27:
In conclusion, our study provides some additional evi- 640–659
dence that MRI may add significant information for penile 19. Kirkham A. MRI of the penis. Br J Radiol 2012;85:S86–S93
cancer staging, particularly for N-staging, especially when
20. Scardino E, Villa G, Bonomo G, et al. Magnetic resonance imaging
clinical examination is considered suboptimal. combined with artificial erection for local staging of penile cancer.
Urology 2004;63:1158–1162.
References 21. Petralia G, Villa G, Scardino E, et al. Local staging of penile cancer
using magnetic resonance imaging with pharmacologically induced
1. Siegel RL, Miller KD, Jemal A. Cancer statistics. CA: Cancer J Clin
penile erection. Radiol Med 2008;113:517–528.
2015;65:5–29.
22. Hanchanale V, Yeo L, Subedi N, et al. The accuracy of magnetic reso-
2. Pow-Sang MR, Ferreira U, Pow-Sang JM, Nardi AC, Destefano V. Epi-
demiology and natural history of penile cancer. Urology 2010;76:S2– nance imaging (MRI) in predicting the invasion of the tunica albuginea
S6. and the urethra during the primary staging of penile cancer. BJU Int
2016;117:439–443.
3. Favorito LA, Nardi AC, Ronalsa M, Zequi SC, Sampaio FJB, Glina S.
Epidemiologic study on penile cancer in Brazil. Int Braz J Urol 2008; 23. Catalona W. Modified inguinal lymphadenectomy for carcinoma of
34:587–593. the penis with preservation of saphenous veins: technique and prelim-
inary results. J Urol 1988;140:306–310.
4. Bleeker MCG, Heideman DAM, Snijders PJF, Horenblas S, Dillner J,
Meijer CJLM. Penile cancer: epidemiology, pathogenesis and preven- 24. Protela C, Alcaraz A, Horenblas S, Pizzocaro G, Zlotta A, Hakenberg
tion. World J Urol 2009;27:141–150. OW. Lymphadenectomy in the surgical management of penile cancer.
Eur Urol 2009;55:1075–1088.
5. Chaux A, Lezcano C, Cubilla AL, Tamboli P, Ro Jae, Ayala A. Compar-
ison of subtypes of penile squamous cell carcinoma from high and 25. Lont AP, Besnard AP, Gallee MP, van Tinteren H, Horenblas S. A
low incidence geographical regions. Int J Surg Pathol 2010;18:268– comparison of physical examination and imaging in determining the
277. extent of primary penile carcinoma. BJU Int 2003;91:493–495.
6. Paiva GR, de Oliveira Ara

ujo IB, Athanazio DA, de Freitas LA. Penile 26. Heyns CF, Mendoza-Vald es A, Pompeo AC. Diagnosis and staging of
cancer: impact of age at diagnosis on morphology and prognosis. Int penile cancer. Urology 2010;76(2 Suppl 1):S15–23.
Urol Nephrol 2015;47:295–299. 27. Gupta S, Rajesh A. Magnetic resonance imaging of penile cancer.
7. Horenblas S. Lymphadenectomy in penile cancer. Urol Clin N Am Magn Reson Imaging Clin N Am 2014;22:191–199.
2011;38:459–469. 28. Ravi R. Correlation between the extent of nodal involvement and sur-
8. Ravi R. Correlation between the extent of nodal involvement and survival following groin dissection for carcinoma of the penis. Br J Urol
vival following groin dissection for carcinoma of the penis. Br J Urol 1993;72:817–819.
1993;72:817–819. 29. Svatek RS, Munsell M, Kincaid JM, et al. Association between lymph
9. Heyns CF, Fleshner N, Sangar V, Schlenker B, Yuvaraja TB, van node density and disease specific survival in patients with penile can-
Poppel H. Management of the lymph nodes in penile cancer. Urology cer. J Urol 2009;182:2721–2727.
2010;76(Suppl 2A):S43–S57.
30. Edge SB, Byrd DR, Carducci MA, et al. AJCC Cancer Staging Manual,
10. Hakenberg OW, Comp erat EM, Minhas S, Protzel C, Watkin N. EAU 7th ed. New York: Springer; 2009.
guidelines on penile cancer: 2014 update. Eur Urol 2015;67:142–150.
31. Horenblas S. Lymphadenectomy for squamous cell carcinoma of the
11. Lont AP, Kroon BK, Gallee MPW, van Tinteren H, Mooneb LMF, penis. Part 2: role and technique of lymph node dissection. BJU Int
Horenblas S. Pelvic lymph node dissection for penile carcinoma: 2001;88:473–483.
extent of inguinal lymph node involvement and survival. J Urol 2007;
32. Heyns CF, Fleshner N, Sangar V, Schlenker B, Yuvaraja TB, van
177:947–952.
Poppel H. Management of the lymph nodes in penile cancer. Urology
12. Heyns CF, Mendoza-Vald es A, Pompeo ACL. Diagnosis and staging 2010;76(2 Suppl 1):S43–57.
of penile cancer. Urology 2010;76(Suppl 2A):S15–S23.
33. Zhu Y, Zhang SL, Ye DW, et al. Predicting pelvic lymph node metasta-
13. Suh CH, Baheti AD, Tirumani SH, et al. Multimodality imaging of pen- ses in penile cancer patients: a comparison of computed tomography,
ile cancer: what radiologists need to know. Abdom Imaging 2015;40: Cloquet’s node, and disease burden of inguinal lymph nodes. Onko-
424–435. logie 2008;31:37–41.
14. Horenblas S. Lymphadenectomy for squamous cell carcinoma of the 34. Graafland NM, Moonen LM, van Boven HH, et al. Prognostic signifi-
penis. Part 1: Diagnosis of lymph node metastasis. BJU Int 2001;88: cance of extranodal extension in patients with pathological node posi-
467–472. tive penile carcinoma. J Urol 2010;184:1347–1353.

ORIGINAL RESEARCH
Optimal High b-Value for Diffusion

Weighted MRI in Diagnosing High Risk
Prostate Cancers in the Peripheral Zone
Harsh K. Agarwal, PhD,1,2 Francesca V. Mertan, BSME,1 Sandeep Sankineni, MD,1
Marcelino Bernardo, BS,1,3 Julien Senegas, PhD,4 Jochen Keupp, PhD,4
Dagane Daar, RT,1,3 Maria Merino, MD,5 Bradford J. Wood, MD,6
Peter A. Pinto, MD,7 Peter L. Choyke, MD,1 and Baris Turkbey, MD1*
Purpose: To retrospectively determine the optimal b-value(s) of diffusion-weighted imaging (DWI) associated with inter-
mediate–high risk cancer in the peripheral zone (PZ) of the prostate.
Materials and Methods: Forty-two consecutive patients underwent multi b-value (16 evenly spaced b-values between 0
and 2000 s/mm2) DWI along with multi-parametric MRI (MP-MRI) of the prostate at 3 Tesla followed by trans-rectal
ultrasound/MRI fusion guided targeted biopsy of suspicious lesions detected at MP-MRI. Computed DWI images up to
a simulated b-value of 4000 s/mm2 were also obtained using a pair of b-values (b 5 133 and 400 or 667 or 933 s/mm2)
from the multi b-value DWI. The contrast ratio of average intensity of the targeted lesions and the background PZ was
determined. Receiver operator characteristic curves and the area under the curve (AUCs) were obtained for separating
patients eligible for active surveillance with low risk prostate cancers from intermediate–high risk prostate cancers as
per the cancer of the prostate risk assessment (CAPRA) scoring system.
Results: The AUC first increased then decreased with the increase in b-values reaching maximum at b 5 1600 s/mm2
(0.74) with no statistically significant different AUC of DWI with b-values 1067–2000 s/mm2. The AUC of computed DWI
increased then decreased with the increase in b-values reaching a maximum of 0.75 around b 5 2000 s/mm2. There was
no statistically significant difference between the AUC of optimal acquired DWI and either of optimal computed DWI.
Conclusion: The optimal b-value for acquired DWI in differentiating intermediate–high from low risk prostate cancers in
the PZ is b 5 1600 s/mm2. The computed DWI has similar performance as that of acquired DWI with the optimal perfor-
mance around b 5 2000 s/mm2.
M ulti-parametric MRI (MP-MRI) is widely used to

detect, localize, and stage prostate cancer (PCa).1–5
Diffusion weighted imaging (DWI) is an integral part of
been added to the recommended repertoire of sequences
obtained during MP-MRI of PCa.6,7
HBV-DWI detects the increased restriction of water in
the prostate MP-MRI.6–10 ADC maps derived from DWI the PCa relative to normal tissue without any post process-
have not only been shown to detect PCa but also ADC val- ing.8 In normal tissue, water diffuses freely with little or
ues correlate with the histologic grade of PCa.10–14 Howev- almost no MR signal in HBV-DWI. Conversely, PCa tissue
er, lesions on ADC maps can be subtle and it has been has increased diffusion restriction leading to a hyperintense
observed that inclusion of high b value (HBV) DWI signal pattern on HBV-DWI. Therefore, HBV-DWI, in
improves the conspicuity of PCa.15–19 Thus, HBV-DWI has effect, suppresses background normal tissue while highlighting
Received Dec 28, 2015, Accepted for publication Jun 7, 2016.
*Address reprint requests to: B.T., Molecular Imaging Program, National Cancer Institute, 10 Center Dr, MSC 1182, Bldg 10, Room B3B85, Bethesda, MD
20892-1088. E-mail: turkbeyi@mail.nih.gov
From the 1Molecular Imaging Program, NCI, NIH, Bethesda, Maryland, USA; 2Philips Research North America, Cambridge, Massachusetts, USA; 3Leidos
Biomedical Research, Inc., Frederick National Laboratory for Cancer Research, Maryland, USA; 4Philips Research Laboratories, Hamburg, Germany;
5
Laboratory of Pathology, NCI, NIH, Bethesda, Maryland, USA; 6Center for Interventional Oncology, NCI and Radiology and Imaging Sciences, Clinical
Center, NIH, Bethesda, Maryland, USA; and 7Urologic Oncology Branch, NCI, NIH, Bethesda, Maryland, USA
Dr Agarwal and Ms Mertan contributed equally to the manuscript and jointly share first authorship.

V 125

PCa, making it easier to detect.20,21 The recently published s/mm2) was obtained. Gradient overplus technique for diffusion
Prostate Imaging and Reporting and Data System (PI- encoding was used to reduce the echo time (TE) to 58 ms for b 5
RADSv2) criteria, recommends use of acquired or computed 2000 s/mm2. mbDWI was obtained using SENSE accelerated
HBV-DWI with b-values above 1400 s/mm2 to improve the (SENSE factor 2) partial-Fourier (halfscan factor 5 0.732), single-
shot SE-EPI with TR/TE 5 3990 ms/58 ms, FOV 5 140 3 140
conspicuity of clinically significant cancers.6,7 Multiple studies
mm, scan resolution 5 2.19 3 2.19 mm, recon resolution 5
have tested the diagnostic accuracy of HBV-DWI for the
1.09 3 1.09 mm. A limited coverage of prostate with slice thick-
detection of PCa,15,16,18–26 however, there has been little
ness/gap 5 2.73 mm/0.27 mm, number of slices 5 16, number of
agreement on the optimal b-value for HBV-DWI.16,20–26 averages 5 2 was used to limit the scan time to 6 min 11 s. The
Typically, the highest b-value used for HBV-DWI is 2000 acquisition time was 8 seconds for b 5 0 s/mm2 DWI and 24 s to
s/mm2 and the most recent recommendation for HBV-DWI acquire 3 directions of the each of the non-zero DWI.
is 1400–2000 s/mm2.6,7 Therefore, in this manuscript, we Calculated-DWI images were also obtained using two DWI
explore a range of b-values from 0 to 2000 s/mm2 for images from mbDWI.15 Calculated-DWI at a given b-value
acquired and up to 4000 s/mm2 for computed HBV-DWI to (cDWIb) is computed as follows,
determine the optimal b-value for detecting clinically signifi-
cant PCa in the peripheral zone based on quantitative con- b2b1 DWIb2
cDWIb 5 DWIb1 exp ln ; b2 > b1 (1)
b22b1 DWIb1
trast ratios of lesion to background.
where DWIb1 and DWI b2 are acquired DWI from mbDWI at b-
Materials and Methods values of b1 and b2, respectively. Calculated DWI was obtained
Study Population for 33 b-values from 0 to 4000 s/mm2 by fixing the smaller b-
This retrospective, single-institution study was approved by the value at 133 s/mm2 (the lowest non-zero acquired b-value) while
local institutional review board and was compliant with the Health the higher b-value was set at 400, 667, or 933 s/mm2 to mimic
Insurance Portability and Accountability Act. Informed consent different b-value combinations used in clinical practice for calculat-
was obtained from each patient. This study included 42 consecu- ed HBV-DWI.7,15,18,28
tive patients between March and October 2014 who underwent
MP-MRI of the prostate with an additional multi b-value DWI TRUS/MRI Fusion Biopsies and ROI Drawings
(mbDWI) and subsequently underwent trans-rectal ultrasound Lesions were identified prospectively on MP-MRI (triplane T2W
(TRUS)/MRI fusion guided targeted biopsy of peripheral zone MRI, ADC maps derived from axial DWI, b 5 2000 s/mm2
DWI from axial HBV-DWI and raw DCE MRI) by the same two
(PZ) lesions identified on MP-MRI.27 Patients were excluded for
radiologists (B.T. with 9 years and P.L.C. with 15 years of experi-
prior focal therapy of the prostate (1 patient) or excessive motion
ence in reading prostate MRI) in consensus using an in-house scor-
on the multi b-value DWI (2 patients). The final study population
ing system for all patients as part of standard clinical
consisted of 39 patients (mean age: 64.8 years; median age 5 64.1
interpretation.4 PI-RADSv2 was not available at the data acquisi-
years; range 5 44.3–79.5 years) with a mean serum prostate spe-
tion phase of this study, therefore, it was not used. Fusion-guided
cific antigen (PSA) of 10.20 ng/mL (median 5 11.56 ng/mL;
biopsies were performed for all identified lesions along with stan-
range 5 6.17–39.00 ng/mL) and an average of 40.4 days between
dard 12 core systemic biopsy. Biopsies were performed on a com-
scan time and biopsy date (median 5 21 days; range 5 1–128
mercial TRUS/MRI fusion guided biopsy system (UroNav, Invivo,
days).
Gainesville, FL), where these identified lesions marked on high res-
olution T2W axial MRI is registered in real time with TRUS imag-
MR Imaging
ing plane to guide and obtained the biopsies from the lesions
Prostate MP-MRI was obtained on a 3 Tesla (T) clinical MR scan-
identified on MP-MRI. All lesions that were identified histopatho-
ner (Achieva-TX, Philips Healthcare, Best, The Netherlands) using
logically as PCa from targeted biopsies were contoured on a single
the anterior half of a 32-channel SENSE cardiac coil (Invivo;
slice of b 5 0 s/mm2 DWI from mbDWI and on a single slice of
Gainesville, FL), and an endorectal coil (ERC) (BPX-30, Medrad,
the ADC map using an open source software (Medical Image Proc-
Pittsburgh, PA). The ERC was placed using a semi-anesthetic gel
essing, Analysis, and Visualization [MIPAV]; National Institutes of
(Lidocaine, AstraZeneca, London) while the patient was in the left
Health, Bethesda MD) In addition, the entire PZ region was also
lateral decubitus position. The balloon surrounding the coil was
contoured on the same slice where the targeted lesion(s) was/were
distended with perfluorocarbon (3 mol/L-Fluorinert, 3M, St. Paul,
identified (Figure 1) for background signal measurement. F1
MN) to a volume of 45 mL to reduce susceptibility artifacts. The
MP-MRI protocol included tri-planar T2W-MRI, axial DWI with PCa Classification
five evenly spaced b values between 0 and 750 s/mm2 referred to Biopsy specimens were evaluated by an experienced genitourinary
as standard DWI, axial HBV-DWI with b-values of 0, 1000, and pathologist (M.J.M., with over 20 years of experience). All PCa
2000 s/mm2, axial DCE MRI and whole abdomen axial T1W lesions were classified either as low or intermediate–high risk using
MRI. the CAPRA scoring system.29,30 This classification was used to
Additionally, axial mbDWI with 16 evenly spaced b-values identify low risk patients eligible for active surveillance with PSA
between 0 and 2000 s/mm2 (b 5 0, 133, 266, 400, 533, 667, less than or equal to 6.0 ng/mL, less than 34% of a biopsy core
800, 933, 1067, 1200, 1333, 1467, 1600, 1733, 1867, and 2000 involved with cancer and no primary Gleason pattern of 4 and 5.

Agarwal et al.: Optimal High b-Value DWI for PCa
C
O
L
O
R
FIGURE 1: ROIs drawn for a right peripheral zone (PZ) lesion of a 61-year-old PCa patient with PSA of 5.29 ng/mL on Axial T2W
MRI (A), ADC map (B), and high b value DWI acquired at b 5 0 s/mm2 (C). TRUS/MRI fusion guided biopsy showed Gleason 7
(413) in the right PZ lesion.
Image Analysis comparisons between the AUC were done using Delong meth-
MATLAB (The MathWorks, Inc., Natick, MA) was used to per- od.32,33 p-values below 0.05 were considered statistically significant.
form image and some statistical analysis. The contrast between
PCa and normal PZ on the same section was computed for DWI
Results
images on multi b-value DWI, 3-b-value DWI, and ADC. The
contrast was defined as,
The final study population of 39 patients had 53 cancers (n 5
17 Gleason 6(313), n 5 16 Gleason 7(314), n 5 5 Gleason
Ic 2Ib 7(413), n 5 1 Gleason 8(315), n 5 8 Gleason 8(414), n 5 6
C5 (2)
Ib Gleason 9(415)). A total of 29 patients had one cancer lesion, 7
patients had two cancer lesions, 2 patients had 3 cancer lesions,
where Ic is the mean value over the PCa ROI and Ib is the mean
and 1 patient had 4 cancer lesions. As per CAPRA criteria, of 53
value over the background PZ tissue.
cancer lesions, 29 cancer lesions were identified as intermediate–
high risk and 24 lesions were identified as low risk cancers. Biop-
Statistical analysis was performed using R (The R Project for Statisti- sy cores rather than PSA dictated the low and intermediate–high
cal Computing). Performance of contrast in ADC maps obtained CAPRA scoring in 10 patients with multiple cancer lesions.
using standard DWI and in acquired and computed DWI images The mean ROI area drawn on mbDWI was 99.4
from mbDWI was evaluated using the area the under the curve mm2 6 89.8 mm2 (median: 75.4 mm2; 1st and 3rd quar-
(AUC) of receiver operating characteristics (ROC) analysis.31 Paired tile: 41.9 mm2 and 104.7 mm2, respectively; range: 14.3–
C
O
L
O
R
FIGURE 2: The AUC obtained from the ROC of diffusion quantitates to separate active surveillance patients with low risk PCa
from active treatment patients with intermediate–high risk PCa as per CAPRA criteria. AUC was obtained for acquired mbDWI
and computed DWI obtained from pair of b-values from mbDWI.
January 2017 127

C
O
L
O
R
FIGURE 3: A 69-year-old PCa patient with serum PSA of 3.05 ng/mL and CAPRA scored low risk PCa in midline anterior peripheral
zone is seen on axial T2W (A), ADC map (B), high b value DWI acquired at b 5 1600 s/mm2 (C), and computed high b value DWI
at b 5 1600 s/mm2computed using acquired DWI images at b 5 133 and 667 s/mm2 (D). Targeted TRUS/MRI fusion guided biop-
sy revealed Gleason 314 with 20% core involvement. The lesion is clearly visible on T2W MRI (A) and ADC map (B); however, it
shows isointense signal pattern compared with the rest of the prostate on both acquired b 5 1600 DWI (C) and computed b1600
DWI (D) (arrows).
394.8 mm2). The ADC values obtained from standard axial s/mm2 reaching maximum of 0.75 at b 5 1500 s/mm2, 0.76 at
DWI for low and intermediate–high risk PCa were 1.1942 b 5 2125 s/mm2 and 0.75 at b 5 1375 s/mm2 for b 5 133 and
6 0.2091 mm2/s and 1.0795 6 0.2120 mm2/s, respectively. 400 s/mm2, b 5 133 and 667 s/mm2 and b 5 133 and 933 s/
The ADC values for low risk PCa patients obtained from mm2, respectively. Beyond the b-value with the highest AUC,
133 s/mm2 and 400, 667, and 933 s/mm2 were 1.4105 6 there is a minor decrease in AUC until b 5 4000 s/mm2. There
0.1624, 1.3269 6 0.1433, and 1.2459 6 0.1232 mm2/s, is no statistically significant difference between the AUC of opti-
respectively. Similarly, the ADC values for intermediate– mal acquired DWI at b 5 1600 s/mm2 and optimal calculated
high risk PCa patients obtained from 133 s/mm2 and 400, DWI. Figures 3 and 4 show sample patients with low risk and F3
667, and 933 s/mm2 were 1.1112 6 0.2309, 1.0282 6 intermediate–high risk PCa, respectively. F4
0.2144, and 0.9800 6 0.1975 mm2/s, respectively.
F2 Figure 2 shows the AUC of the ROC curve for differen- Discussion
tiating intermediate–high risk PCa from low risk PCa. The DWI plays a critical role in MRI of the prostate. Hypoin-
AUC of contrast ratio of ADC maps obtained from standard tense signal pattern on ADC maps and hyperintense signal
axial DWI was 0.55. The AUC for the acquired DWI images on HBV-DWI are used as an important imaging marker for
from mbDWI increased rapidly with higher b values between the presence of PCa.6,7,34 Although HBV-DWI has been
b 5 266 s/mm2 (AUC 5 0.52) and b 5 1067 s/mm2 (0.70) hypothesized to detect high risk PCa, the current literature
reaching maximum at b 5 1600 s/mm2 (0.74). Beyond b 5 is focused on differentiating PCa from normal prostate tis-
1600 s/mm2, there is a minor decrease in AUC to 0.70 at b 5 sue15,16,18–26 with no clear choice of optimal b-value. In
2000 s/mm2. There was no statistically significant difference this manuscript, we evaluated the performance of acquired
between the AUC of DWI at b 5 1600 s/mm2 and AUC of and calculated HBV-DWI to distinguish low risk PCa from
the acquired DWI in b-values ranging from 1067 s/mm2 to intermediate–high risk PCa using 16 equally spaced b-value
2000 s/mm2. images ranging from 0 to 2000 s/mm2. Our results show
Similarly, the AUC of calculated DWI from multi b-value that DWI obtained with b 5 1600 s/mm2 has the optimal
DWI increased rapidly with increase in b-values up to b 5 1200 AUC (0.74) for discriminating low risk from intermediate–

C
O
L
O
R
FIGURE 4: A 77-year-old PCa patient with serum PSA of 15.08 ng/mL and CAPRA scored high risk PCa in right apical-mid periph-
eral zone identified on axial T2W (A), ADC map (B), high b value DWI acquired at b 5 1600 s/mm2 (C), and computed high b value
DWI at b 5 1600 s/mm2computed using acquired DWI images at b 5 133 and 667 s/mm2 (D). Targeted TRUS/MRI fusion guided
biopsy revealed Gleason 315 with 80% core involvement. The lesion is clearly visible on T2W MRI (A) and ADC map (B); moreover
it shows hyperintense signal pattern compared to the rest of the prostate on both acquired b 5 1600 DWI (C) and computed
b1600 DWI (D) (arrows).
high risk cancers. However, the differences in AUC among lesions on HBV-DWI.6,7 This work concurs with several
the different b-values >1000 s/mm2 were small suggesting prior studies that state that b 5 2000 s/mm2 HBV-DWI
that even a b-value around 1067 s/mm2 can be sufficient has better diagnostic performance in comparison to b 5
for the depiction of high risk cancers. The calculated DWI 1000 s/mm2 (see references 16,21–26) and also in comparison
images derived from relatively low b-values (<1000 s/mm2) to b 5 500, 1000, and 2000 s/mm2.20 Both b 5 1500 s/
can also be used to replace direct HBV-DWI. mm2 and b 5 2000 s/mm2 HBV-DWI have similar and
PI-RADSv2 recommends DWI to be used as the prima- improved diagnostic performance compared with b 5 1000
ry pulse sequence for lesions in the peripheral zone. This for- s/mm2 and b 5 2500 s/mm2.19 However, the distinction
mulation includes qualitative evaluation of ADC maps and between clinically insignificant (low risk) and clinically sig-
high b-value DWI (with b-values between 1400 and 2000 s/ nificant (intermediate–high risk) cancers was not made in
mm2, if adequate signal-to-noise ratio [SNR] permits).7 HBV- these studies. This study differs from previous studies
DWI is advantageous because it suppresses background normal because it compares the ability of HBV-DWI to distinguish
tissue and most of the low risk PCa. The higher diffusion of low from intermediate–high risk tumors. Patient selection
background tissue and low risk PCa quickly decays to noise for biopsy is an important consideration in reducing health
levels with increases in b-value due to its higher ADC value, care costs related to the use of MRI for PCa. By identifying
then plateaus at higher b-values. However, the DWI intensity patients with abnormal but low risk disease, some biopsies
of intermediate–high risk PCa decreases more slowly with the could be avoided or at least deferred.
increase in b-value due to the impeded diffusion in these This study has several limitations. First, we used
tumors. In this manuscript, we have shown that the perfor- TRUS/MRI fusion guided biopsy as a validation method
mance of acquired high b-value DWI peaked at b 5 1600 s/ instead of radical prostatectomy. The advantage of doing
mm2, then decreased with the further increase in b-values. Fur- this is that it allows a broader class of patients to be includ-
thermore, there was no statistically significant difference in the ed in the study (i.e., active surveillance and radiation thera-
performance of acquired high b-value DWI between b-values py candidates as well as surgical candidates). There is a
of 1000 s/mm2 and 2000 s/mm2. Therefore, this manuscript chance that a tumor will be missed on MP-MRI or missed/
supports the consensus recommendation in PI-RADSv2. undersampled at biopsy because this study did not use
In most MP-MRI studies, DWI is evaluated qualita- whole-mount histopathology as the gold standard. However,
tively based on ADC maps and the appearance of bright TRUS/MRI fusion guidance has been proven highly
January 2017 129

effective in diagnosing patients with intermediate/high grade 4. Turkbey B, Mani H, Aras O, et al. Prostate cancer: can multiparametric
MR imaging help identify patients who are candidates for active sur-
tumors compared with traditional systematic biopsy.35 A veillance? Radiology 2013;268:144–152.
second limitation of this study is that only lesions positive 5. Pinto PA, Chung PH, Rastinehad AR, et al. Magnetic resonance imag-
on MP-MRI were included. Across multiple studies, the ing/ultrasound fusion guided prostate biopsy improves cancer detec-
tion following transrectal ultrasound biopsy and correlates with
false negative rate for the detection of intermediate–high
multiparametric magnetic resonance imaging. J Urol 2011;186:1281–
risk cancer is <5% using MP-MRI so this should have a 1285.
small impact on the results. AUC is not perfect, indicating 6. Barentsz JO, Richenberg J, Clements R, et al. ESUR prostate MR
that this method still misses some tumors and false calls guidelines 2012. Eur Radiol 2012;22:746–757.
others positives. HBV-MRI, while a step forward, is not the 7. Weinreb JC, Barentsz JO, Choyke PL, et al. Prostate imaging and
reporting and data system: version 2. Eur Radiol 2016;69:16–40.
definitive method for diagnosing such cancers. Other meth-
ods, alone or in combination, may be needed to further 8. Patterson DM, Padhani AR, Collins DJ. Technology insight: water dif-
fusion MRI--a potential new biomarker of response to cancer therapy.
improve the diagnosis of intermediate high risk PCa. A Nat Clin Pract Oncol 2008;5:220–233.
third limitation of this study is that the use of an endorectal 9. Kajihara H, Hayashida Y, Murakami R, et al. Usefulness of diffusion-
coil and images acquired at a field strength of 3T may not weighted imaging in the localization of prostate cancer. Int J Radiat
Oncol Biol Phys 2009;74:399–403.
be performed in all centers. Scans obtained with lower SNR
10. Yoshimitsu K, Kiyoshima K, Irie H, et al. Usefulness of apparent diffu-
may have different optimal high b-values compared with
sion coefficient map in diagnosing prostate carcinoma: correlation
what we found in this study. A fourth limitation of this with stepwise histopathology. J Magn Reson Imaging 2008;27:132–
study included using the CAPRA method on a per lesion 139.
basis instead of on a per patient basis. The CAPRA scoring 11. Turkbey B, Shah VP, Pang Y, et al. Is apparent diffusion coefficient
associated with clinical risk scores for prostate cancers that are visible
system is one of the very few available validated nomograms on 3-T MR images? Radiology 2011;258:488–495.
that can allow patients with secondary Gleason pattern 4 to
12. Itou Y, Nakanishi K, Narumi Y, et al. Clinical utility of apparent diffu-
be an active surveillance candidate. Using this system on a sion coefficient (ADC) values in patients with prostate cancer: can
per patient basis would not contribute to our primary objec- ADC values contribute to assess the aggressiveness of prostate can-
cer? J Magn Reson Imaging 2011;33:167–172.
tive to find the best high b-value in the detection of high
13. Vargas HA, Akin O, Franiel T, et al. Diffusion-weighted endorectal MR
risk prostate cancer in the peripheral zone as the sample size imaging at 3 T for prostate cancer: tumor detection and assessment
would artificially dilute our lesion categorizations. Thus, of aggressiveness. Radiology 2011;259:775–784.
using the CAPRA scoring system on a per lesion basis was 14. Hambrock T, Somford DM, Huisman HJ, et al. Relationship between
useful as an extra step to biopsy confirmation. Finally, this apparent diffusion coefficients at 3.0-T MR imaging and gleason grade
in peripheral zone prostate cancer. Radiology 2011;259:453–461.
is a retrospective study with a relatively limited patient/
15. Grant KB, Agarwal HK, Shih JH, et al. Comparison of calculated and
lesion population size which could introduce a potential acquired high b value diffusion-weighted imaging in prostate cancer.
patient selection bias that would be less likely to occur pro- Abdom Imaging 2015;40:578–586.
spectively. However, the patients in this study were consecu- 16. Tamada T, Kanomata N, Sone T, et al. High b value (2,000 s/mm2)
tive reducing the likelihood of a specific selection bias. diffusion-weighted magnetic resonance imaging in prostate cancer at
3 Tesla: comparison with 1,000 s/mm2 for tumor conspicuity and dis-
In conclusion, our study shows that b 5 1600 s/mm2 crimination of aggressiveness. Plos One 2014;9:e96619.
is the optimal b-value for HBV-DWI in separating high risk 17. Rosenkrantz AB, Mannelli L, Kong X, et al. Prostate cancer: utility of
PCa from low risk PCa. HBV-DWI at b 5 1600 s/mm2 fusion of T2-weighted and high b-value diffusion-weighted images for
peripheral zone tumor detection and localization. J Magn Reson
and HBV-DWI between b 5 1067 and b 5 2000 s/mm2 Imaging 2011;34:95–100.
has similar performance indicating that a HBV-DWI with
18. Maas MC, Futterer JJ, Scheenen TW. Quantitative evaluation of com-
b-value between 1067 and 2000 s/mm2 can be comfortably puted high B value diffusion-weighted magnetic resonance imaging
used to detect high risk PCa lesions. The computed HBV- of the prostate. Invest Radiol 2013;48:779–786.
DWI computed without perfusion effect performed similar 19. Metens T, Miranda D, Absil J, et al. What is the optimal b value in
diffusion-weighted MR imaging to depict prostate cancer at 3T? Eur
to that of acquired HBV-DWI with best performance Radiol 2012;22:703–709.
around b 5 2000 s/mm2. 20. Ohgiya Y, Suyama J, Seino N, et al. Diagnostic accuracy of ultra-high-
b-value 3.0-T diffusion-weighted MR imaging for detection of prostate
References cancer. Clin Imaging 2012;36:526–531.
1. Hoeks CM, Barentsz JO, Hambrock T, et al. Prostate cancer: multi- 21. Katahira K, Takahara T, Kwee TC, et al. Ultra-high-b-value diffusion-
parametric MR imaging for detection, localization, and staging. Radi- weighted MR imaging for the detection of prostate cancer: evaluation
ology 2011;261:46–66. in 201 cases with histopathological correlation. Eur Radiol 2011;21:
188–196.
2. Delongchamps NB, Rouanne M, Flam T, et al. Multiparametric magnet-
ic resonance imaging for the detection and localization of prostate can- 22. Kim CK, Park BK, Kim B. High-b-value diffusion-weighted imaging at
cer: combination of T2-weighted, dynamic contrast-enhanced and 3 T to detect prostate cancer: comparisons between b values of
diffusion-weighted imaging. BJU Int 2011;107:1411–1418. 1,000 and 2,000 s/mm2. AJR Am J Roentgenol 2010;194:W33–W37.
3. Kurhanewicz J, Vigneron D, Carroll P, et al. Multiparametric magnetic 23. Ueno Y, Kitajima K, Sugimura K, et al. Ultra-high b-value diffusion-
resonance imaging in prostate cancer: present and future. Curr Opin weighted MRI for the detection of prostate cancer with 3-T MRI.
Urol 2008;18:71–77. J Magn Reson Imaging 2013;38:154–160.

24. Rosenkrantz AB, Hindman N, Lim RP, et al. Diffusion-weighted imag- prediction of recurrence after radical prostatectomy. Cancer 2006;
ing of the prostate: comparison of b1000 and b2000 image sets for 107:2384–2391.
index lesion detection. J Magn Reson Imaging 2013;38:694–700.
30. Cooperberg MR, Pasta DJ, Elkin EP, et al. The University of California,
25. Manenti G, Nezzo M, Chegai F, et al. DWI of prostate cancer: optimal San Francisco Cancer of the Prostate Risk Assessment score: a
b-value in clinical practice. Prostate Cancer 2014;2014:868269. straightforward and reliable preoperative predictor of disease recur-
rence after radical prostatectomy. J Urol 2005;173:1938–1942.
26. Kitajima K, Takahashi S, Ueno Y, et al. Clinical utility of apparent diffu-
sion coefficient values obtained using high b-value when diagnosing 31. Hanley JA, McNeil BJ. The meaning and use of the area under a receiv-
prostate cancer using 3 tesla MRI: comparison between ultra-high b- er operating characteristic (ROC) curve. Radiology 1982;143:29–36.
value (2000 s/mm(2)) and standard high b-value (1000 s/mm(2)).
J Magn Reson Imaging 2012;36:198–205. 32. Hanley JA, McNeil BJ. A method of comparing the areas under
receiver operating characteristic curves derived from the same cases.
27. Xu S, Kruecker J, Turkbey B, et al. Real-time MRI-TRUS fusion for Radiology 1983;148:839–843.
guidance of targeted prostate biopsies. Comput Aided Surg 2008;13:
255–264. 33. Robin X, Turck N, Hainard A, et al. pROC: an open-source package for R
and S1 to analyze and compare ROC curves. BMC Bioinform 2011;12:77.
28. Feuerlein S, Davenport MS, Krishnaraj A, et al. Computed high b-
value diffusion-weighted imaging improves lesion contrast and con- 34. Tamada T, Sone T, Jo Y, et al. Diffusion-weighted MRI and its role in
spicuity in prostate cancer. Prostate Cancer Prostatic Dis 2015;18: prostate cancer. NMR Biomed 2014;27:25–38.
155–160.
35. Siddiqui MM, Rais-Bahrami S, Turkbey B, et al. Comparison of MR/
29. Cooperberg MR, Freedland SJ, Pasta DJ, et al. Multiinstitutional vali- ultrasound fusion-guided biopsy with ultrasound-guided biopsy for
dation of the UCSF cancer of the prostate risk assessment for the diagnosis of prostate cancer. JAMA 2015;313:390–397.
January 2017 131

ORIGINAL RESEARCH
Endogenous Assessment of Diffuse

Myocardial Fibrosis in Patients
With T1q-Mapping
Joep W.M. van Oorschot, PhD,1,2* Fatih G€ u, MD,2 Sanne de Jong, PhD,3
uçl€
Steven A.J. Chamuleau, MD, PhD,2 Peter R. Luijten, Prof,4 Tim Leiner, Prof,4 and
Jaco J.M. Zwanenburg, PhD4
Purpose: Recently, it was shown that a significantly higher T1q is found in compact myocardial fibrosis after chronic myo-
cardial infarction. In this study, we investigated the feasibility of native T1q-mapping for the detection of diffuse myocar-
dial fibrosis in patients with dilated cardiomyopathy (DCM).
Materials and Methods: T1q-mapping was performed on three explanted hearts from DCM patients at 3 Tesla (T). His-
tological fibrosis quantification was performed, and compared with the T1q-relaxation times in the heart. Furthermore,
twenty DCM patients underwent an MRI at 1.5T. Native T1q-maps, native T1-maps, and extracellular volume (ECV)-maps
were acquired. Additionally, eight healthy volunteers were scanned for reference values.
Results: A significant correlation (Pearson r 5 0.49; P 5 0.005) was found between ex vivo T1q-values and fibrosis frac-
tion from histology. Additionally, a significantly higher T1q-relaxation time (55.2 6 2.7 ms) was found in DCM patients
compared with healthy control subjects (51.5 6 1.2 ms) (P 5 0.0024). The relation between in vivo T1q-values and
ECV-values was significant (Pearson r 5 0.66). No significant relation was found between native T1- and ECV-values in
this study (P 5 0.89).
Conclusion: This study showed proof of principle for the endogenous detection of diffuse myocardial fibrosis with T1q-
MRI. Ex vivo and in vivo experiments showed promising results that T1q-MRI can be used to measure the extent of dif-
fuse myocardial fibrosis in the myocardium.
A dverse structural remodeling of the myocardium occurs

in almost every cardiac disease. A major result of the
adverse remodeling process is the deposition of collagen.1
However, the visual detection of diffuse interstitial myocar-
dial fibrosis with LGE is suboptimal because there is an
absence of an area with clearly unaffected myocardium to
Reactive fibrosis corresponds to collagen that is deposited use for comparison.3
under pathological circumstances without cell loss, and To detect the presence of diffuse fibrosis, methods for
increases stiffness of the heart, which contributes to diastolic quantitative mapping of extracellular volume (ECV) have
failure and reduced contractility. In addition, diffuse reactive been developed.4,5 ECV-mapping requires quantitative
fibrosis can alter the electrical activity pattern in the heart, measurements of the longitudinal relaxation time, T1, before
which may result in arrhythmias.1 and after the administration of a contrast agent.6
The usual imaging method to assess cardiac fibrosis Although the application of contrast enhanced myocar-
non-invasively is late gadolinium enhanced (LGE) MR dial T1-mapping and quantification of ECV provides a valu-
method.2 This technique relies on the difference in contrast able tool to study reactive fibrosis, there are still limitations
agent washout between normal and diseased myocardium. that stimulate the search for other imaging techniques. Both

*Address reprint requests to: J.v.O., Department of Radiology, University Medical Center Utrecht, Heidelberglaan 100 3508 GA Utrecht.
E-mail: j.w.m.oorschot@outlook.com
From the 1Philips Healthcare, Best, The Netherlands; 2Department of Cardiology, University Medical Center Utrecht, Utrecht, The Netherlands;
3
Department of Medical Physiology, University Medical Center Utrecht, Utrecht, The Netherlands; and 4Department of Radiology, University Medical Center
Utrecht, Utrecht, The Netherlands

V

van Oorschot et al.: Detecting Diffuse Cardiac Fibrosis With T1q-MRI
acute and delayed reactions due to administration of the gad-

TABLE 1. Imaging Parameters for the Balanced SSFP
olinium based contrast agent have been described, and this T1q-Mapping Method Used in the Experiments
has led to the recommendation not to use these agents in
patients with severe renal failure.7,8 Furthermore, the calcula- Ex vivo In vivo
tion of an ECV-map requires the acquisition of two different imaging imaging
T1-maps, one before and one 15–20 min after contrast injec-
tion, which requires a long scan time. These drawbacks can Field strength 3T 1.5T
potentially be overcome when endogenous MRI contrast BW/pixel 985 Hz 723 Hz
mechanisms are used to detect myocardial fibrosis.9 TE/TR 1.7/3.4 ms 1.74/3.5 ms
A promising endogenous MRI contrast mechanism to 2
Resolution 0.75 x 0.75 mm 2 x 2 mm2
measure myocardial fibrosis is T1q-MRI. The T1q-relaxation
time describes relaxation while the magnetization is in the FOV 120 x 120 mm2 288 x 288 mm2
transverse plane, in the presence of a so-called SL radiofre- Slice thickness 0.75 mm 8 mm
quency (RF) pulse. This parameter is known to be sensitive Flip angle 10 degrees 35 degrees
to macromolecular interactions, and is successfully used in
SENSE factor 2 2
the field of orthopedics to assess cartilage degeneration.10,11
It is important to note the difference between T1- No. of startup 0 8
pulses
relaxation and T1q-relaxation. The T1-relaxation time
describes the relaxation mechanisms of spin interactions at SL pulse 0, 10, 20, 30, 40 0, 10, 20, 30, 40
duration
the Larmor frequency created by the main magnetic field
B0, which is typically between 50 and 300 MHz in MR sys- SL amplitude 500 Hz 500 Hz
tems. On the other hand, T1q-relaxation describes the relax-
ation of magnetization during a spin-lock (SL) pulse. In
Before cardiac transplantation, all patients gave written informed
presence of a SL pulse, spins can interact with the surround-
consent for using the explanted hearts for research purposes. The
ing at frequencies close to the SL amplitude, which is typi- study was approved by the local Medical Ethics Committee of the
cally between 300 and 1000 Hz, far below the Larmor University Medical Centre Utrecht.
frequency. The T1q-relaxation time is, therefore, sensitive to After explanation, hearts were cut in transverse slices (6 1 cm
low frequency interactions between water and macromole- thickness). Per heart one fresh slice was kept at 48C and imaged by
cules, which have slower tumbling rates. MR within 24 h.
It has been shown that higher T1q-values are found in com-
pact myocardial fibrosis after chronic myocardial infarction in MRI Methods
experimental animal models.12–14 Recently, a study in patients MR imaging was performed on a 3 Tesla (T) clinical MR Scanner
has shown that native T1q-mapping can be used to detect myo- (Achieva TX, Software release 3.2.1, Philips Healthcare, Best, The
Netherlands) using an eight-channel head coil. The explanted heart
cardial scar in patients with a chronic myocardial infarction.15
slice was placed in a perfluoropolyether solution (Fomblin, Solvay,
Another promising endogenous MR contrast for the
Brussels, Belgium) in a larger container, to reduce the magnetic
detection of diffuse fibrosis is native T1-mapping. Signifi- field inhomogeneities around the tissue boundaries. Image acquisition
cant changes in native T1-relaxation time have been found was performed at room temperature.
in patients with dilated cardiomyopathy,16 and a correlation T1q-mapping was performed using a three-dimensional
was shown with collagen volume fraction in other cardiomy- (3D), T1q-prepared, multi-shot gradient echo sequence. T1q-prepa-
opathies such as aortic stenosis.17,18 ration was performed using a SL pulse consisting of 2 RF pulses,
Here, we demonstrate the feasibility of native T1q- with opposite phase to compensate for B1 variations, and a refo-
MRI to detect interstitial myocardial fibrosis. T1q-mapping cusing pulse in between to compensate for B0 errors.19 The ampli-
was performed on human hearts of patients suffering from tude of the SL pulse was set to 500 Hz (11.7 lT), and five images
end-stage dilated cardiomyopathy (DCM), and subsequently with different SL preparation times were acquired (SL 5 1, 10,
compared with histology to evaluate the relation between 20, 30, 40 ms). Other parameters were: bandwidth (BW)/pixel 5
T1q-relaxation time and myocardial collagen levels. Subse- 985 Hz, echo time/repetition time (TE/TR) 5 1.7/3.4 ms, spatial
resolution 5 0.75 3 0.75 mm2, slice thickness 5 0.75 mm, field
quently, we studied the feasibility of T1q-mapping to detect
of view (FOV) 5 120 3 120 mm2, flip angle 5 10 degrees, turbo
diffuse myocardial fibrosis in vivo in DCM patients.
field echo (TFE) factor 5 64, number of averages (NSA) 5 1,
and shot interval 5 3000 ms (Table 1). T1
Explanted Hearts Study Histology
Explanted hearts (n 5 3) from DCM patients admitted to the Uni- Immediately after the image acquisition was completed, cardiac
versity Medical Centre Utrecht for heart transplantation, were used. slices were fixated in 4% formaldehyde. Fixated slices were divided
January 2017 133

C
O
L
O
R
FIGURE 1: T1q-map of an ex vivo cardiac slice and a corresponding histological section. Representative example of a T1q-map of a trans-
verse cardiac slice from an explanted human heart. In the T1q-map the ROI is depicted corresponding with the cardiac piece analyzed
with Masson’s trichrome shown on the right. The inset shows a detailed view of the stained myocardium; scale bar 5 200 lm. Blue 5
fibrosis; red 5 cardiomyocytes; white (not stained) 5 adipose tissue.
into 9–12 transmural pieces based on a dissection scheme and sub- T1-maps were acquired using the MOLLI 4(1)3(1)2 scheme.21
F1 sequently embedded in paraffin (Fig. 1). Embedded pieces were Other parameters were: bandwidth/pixel 5 723 Hz, TE/TR 5
sectioned (4 lm thickness) and stained with Masson’s trichrome 1.74/3.5 ms, resolution 5 2 3 2 mm2, slice thickness 5 8 mm,
for detection and quantification of myocardial fibrosis, cardiomyo- slice gap 5 15 mm, FOV 5 288 3 288 mm2, flip angle 5 35
cytes and adipose tissue. Stained sections were scanned at 203 degrees, SENSE acceleration 5 2. Blood samples were acquired
magnification and subsequently extracted for analysis as described directly before the MR exam to determine hematocrit. In the
previously.20 healthy control subjects, only the T1q- and native T1-maps were
acquired. Postcontrast T1-maps and hematocrit was not acquired
Patient Study in the control subjects, because we did not have Medical Ethical
permission for gadolinium administration and blood withdrawal in
SUBJECTS. Twenty patients with idiopathic DCM underwent healthy subjects.
cardiac MRI. Eight healthy control subjects (six male, two female; Left ventricular ejection fraction in patients was determined
age 51 6 6 years) were imaged to obtain reference T1q-values. using a CINE acquisition. Fifteen short axis slices with 30 heart
Written informed consent was obtained from all participants, and phases were acquired using a bSSFP sequence. Other parameters
the local Ethical Review Board of the University Medical Center were: bandwidth/pixel 5 1036 Hz, TE/TR 5 2.4/4.7 ms, resolu-
Utrecht approved the study. tion 5 1.67 3 1.75 mm2, slice thickness 5 8 mm, FOV 5 320
MRI IMAGING. All subjects were imaged on a clinical 1.5T MR 3 320 mm2, flip angle 5 60 degrees, SENSE acceleration 5 3.
scanner (Philips Ingenia, Software release 5.1.7, Philips Healthcare, POSTPROCESSING AND ANALYSIS. T1q-maps were calculated
Best, The Netherlands). T1q-mapping was performed using a 2D, by pixelwise fitting of a mono-exponential decay function using
single shot T1q-prepared steady-state free precession (SSFP) Matlab (Version R2014b, Mathworks, Natick, MA). A mono-
sequence. Five images were obtained with different SL preparation exponential two-parameter decay model was applied using a
times (SL 5 0, 10, 20, 30, 35 ms) in diastole. The amplitude of Levenberg-Marquardt algorithm of nonlinear estimation to fit the
the SL pulse was set to 500 Hz (11.7 lT). Other parameters were: T1q-relaxation time.22
bandwidth/pixel 5 723 Hz, TE/TR 5 1.74/3.5 ms, resolution 5 From the histological slices of the explanted hearts, the per-
2 3 2 mm2, slice thickness 5 8 mm, FOV 5 288 3 288 mm2, centage of fibrosis, cardiomyocytes, and adipose tissue in each sec-
flip angle 5 35 degrees, SENSE acceleration 5 2, shot interval 5 tion was determined using MATLAB (release 2014b, Mathworks,
3 heart beats. In all patients, the T1q-mapping was performed in Natick, MA). Technical details about the quantification method
three short axis slices. The middle slice was located at the mid ven- performed by MATLAB were described previously.20 Subsequently,
tricular level, and the slice gap was 15 mm. Data for each T1q- regions of interest (ROIs) were outlined on the T1q-maps, match-
map were acquired in a single breath-hold in end-expiration of 13 ing with the dissection scheme and guided by the histology images.
heart beats duration. Mean T1q-relaxation time per ROI was calculated and compared
To assess the ability of T1q-mapping to detect diffuse myo- with the percentage fibrosis per piece.
cardial fibrosis in vivo, corresponding T1-maps were acquired in all Using the pre- and postcontrast T1-maps and the hematocrit
patients, before and 15 min after administration of 0.2 mmol/kg value, ECV-fraction maps were calculated for each slice in patients,
contrast agent (Gadovist, Bayer Healthcare, Berlin, Germany). The using the same approach as described elsewhere.6
T1- and T1q-maps were acquired on the same position and orienta- The myocardium of the left ventricle was delineated by man-
tion, and with the same number of slices. Native T1-maps were ually contouring the endocardial and epicardial border on the T1q-
acquired using a MOLLI 5(3)3 scheme, and the contrast enhanced maps of both patients and healthy volunteers. These ROIs were

control subjects (51.5 6 1.2 ms), P 5 0.0024, see Figure 4. F4

A significant correlation was found between the T1q-relaxa-
tion time and the ECV in patients, with a Pearson correla-
tion coefficient r 5 0.66 (P 5 0.0015) (Fig. 5a). Mean F5
ECV-value in the DCM patients was 0.32 6 0.04.
The native T1-relaxation time was significantly higher
in the DCM patients (1166 6 62 ms), compared with the
healthy control subjects (1026 6 21 ms, P < 0.0001).
However, no significant correlation was found between the
T1q-relaxation time and native T1-values in patients (Pear-
son r 5 0.26; P 5 0.28), and between the native T1- and
ECV-fraction in patients (Pearson r 5 0.03; P 5 0.89), as
shown in Figure 5b.
FIGURE 2: Correlation between amount of fibrosis and T1q-
relaxation time. A significant but moderate correlation with a
Discussion
Pearson r of 0.49 was found between amount of fibrosis and T1q-
relaxation time in explanted hearts of end stage DCM patients. In this study, we demonstrated the proof of principle for in
vivo detection of diffuse myocardial fibrosis in patients with
used to calculate the mean T1q-relaxation time and the mean end-stage DCM using native T1q-mapping. We found a sig-
native T1-relaxation time for each slice, and for the patients, also nificant correlation between T1q-values and fibrosis as deter-
the ECV fraction. mined by histology in ex vivo human end-stage heart failure
myocardium due to DCM. In vivo, we found a significantly
higher T1q-relaxation time in DCM patients compared with
Statistical analysis was performed with GraphPad Prism (GraphPad
healthy control subjects, and a significant correlation
Software, La Jolla, CA). The Pearson correlation coefficient was
calculated to determine the correlation between the percent fibrosis
between T1q-values and ECV-fractionin DCM patients.
at histology and T1q-values. Group comparisons were performed Myocardial native T1 was also significantly higher in the
using an unpaired Student’s t-test, and considered significant at DCM patients, but no correlation was found between native
P < 0.05. The correlation between native T1q-relaxation time and T1 and T1q or between native T1 and ECV.
ECV fraction, as well as the correlation between native T1q and In chronic myocardial infarct scar tissue, the T1q-relax-
native T1-values was also calculated using the Pearson correlation ation time has already been found to be significantly higher
coefficient. compared with remote myocardium. The ex vivo results
presented here show a modest but significant correlation
Results between T1q-values and fibrosis fraction, suggesting that
Explanted Hearts Study T1q-MRI may also provide information on diffuse fibrosis
In total, 32 pieces of the three explanted heart slices were formation. The modest correlation coefficient might be
assessed for histology. The mean T1q-value of all these was explained by challenges in the experiment setup. When vali-
73.5 6 7.5 ms (range, 60.0–88.0 ms). Figure 1 shows a dating MR images with histology as a reference, there are
representative T1q-map with a corresponding picture of a some pitfalls that may limit the correlation. First, in this
histological image stained with Masson’s trichrome. Percent- study the MR images gather information from tissue slices
age fibrosis in all pieces, as determined by quantification of with a thickness of 0.75 mm thickness.
Masson’s trichrome staining, ranged from 8.3 to 58.0% In contrast, histology is performed on slices of only 4
(mean value 24.5 6 9.5%). When comparing the histologi- lm thickness. This suggests that MR images average much
cally determined mean percentage of fibrosis with the mean more tissue information compared with histological slices,
T1q-value of each piece, a Pearson correlation coefficient which might hamper a proper correlation. Second, due to
F2 r 5 0.49 (P 5 0.005) was observed (Fig. 2). technical aspects, there is always some discrepancy between
the MRI planimetry and the cutting plane of the tissue for
Patient Study histology, which makes accurate matching of both histology
The patient characteristics of the 20 patients are listed in and MR images challenging. Similarly, the in-plane ROIs
T2 Table 2. Mean ejection fraction in the DCM patients was may not perfectly match the sections made for histology. A
25 6 13 %. An example of a T1q -map, pre- and postcon- direct comparison of the T1q-relaxation times found in the
trast T1 -maps, and corresponding ECV-map is given in explanted hearts and the in vivo patients is difficult, because
F3 Figure 3. the measurements are performed at a different field strength,
The T1q-relaxation time was significantly higher in the and at a different temperature. The T1q-values we found in
DCM patients (55.2 6 2.7 ms), compared with the healthy the ex vivo hearts were higher than in in vivo. We believe
January 2017 135

TABLE 2. Patient Characteristics
Controls (n 5 8) DCM patients (n 5 20)
Sex Male n 5 6; 75% Male n 5 7; 35%

Female n 5 2; 25% Female n 5 13; 65%
Age, years 51 6 6 60 6 13
BMI 22 6 2 25 6 5
Smoking n 5 1; 5% n 5 2; 10%
Hypertension n 5 0; 0% n 5 8; 40%
Diabetes n 5 0; 0% n 5 2; 10%
Hyperlipidemia n 5 0; 0% n 5 0; 0%
Family history of vascular disease n 5 0; 0% n 5 10; 50%
Antihypertensive drugs n 5 0; 0% b-Blockers n 5 10; 50%
Diuretics n 5 16; 80%
Angiotensin-converting enzyme
inhibitors n 5 8; 40%
Angiotensin antagonists n 5 5; 25%
Statin use n 5 0; 0% n 5 8; 40%

Antithrombotic therapy n 5 0; 0% n 5 10; 50%
this is mainly caused by a more severe state of disease and with this study, one should realize that, in HCM, larger
more fibrosis in these explanted hearts, and not by the dif- patches of fibrosis are formed, where in DCM patients dif-
ference in main magnetic field strength. A future study fuse interstitial fibrosis is formed.
should be performed at the same field strength, with the Results of biopsy studies have shown that the ECV
same T1q-mapping readout, to obtain a more direct correla- fraction, calculated with native T1, contrast enhanced T1,
tion between the ex vivo and in vivo results. and hematocrit value combined, is an excellent predictor of
In vivo, both significantly higher T1q-relaxation times the amount of myocardial fibrosis.25,26 The significant
and significantly higher native T1-relaxation times were correlation between T1q-relaxation time and ECV suggests
found in DCM patients. The higher native T1-values are in that a single T1q experiment could potentially provide
accordance with other studies in literature where it was similar information.
shown that higher native T1-values can be found in DCM In other studies, the use of native T1-mapping for
patients.23 However, this is the first study that reports a sig- detection of diffuse myocardial fibrosis has been investi-
nificantly higher T1q-relaxation time in DCM patients. gated. It has been shown that T1-mapping without contrast
Another recent study has shown that in patients with hyper- agent can provide information on diffuse fibrosis.17,27
trophic cardiomyopathy (HCM) an increase in T1q-relaxa- Another study found that native T1-values can provide diag-
tion time is found in the heart, and that the fibrosis in the nostic accuracy to discriminate between normal and diffuse
heart can be assessed using this method.24 In comparing fibrosis in patients with nonischemic DCM, and is superior
C
O
L
O
R
FIGURE 3: In vivo short axis T1q-, T1-, and ECV-maps. In vivo short axis T1q-map, with corresponding pre- and postcontrast T1-map
and resulting ECV-map, in a DCM patient.

Main limitation for the use of T1q-mapping for fibrosis

detection is that the T1q-relaxation time found in DCM
patients is only approximately 2 SDs higher compared with
healthy control subjects. Although this appears to be sufficient
to distinguish healthy and diseased myocardium, more contrast
could improve the sensitivity of the method. It was shown in
infarct tissue that a higher B1 amplitude for the SL pulse
would generate more contrast between healthy and diseased
myocardium.15 Because the B1 of the SL pulse is limited by
the specific absorption rate (SAR) and RF performance of the
MR system, there is a limit on the maximum SL amplitude.
New MRI contrast developments such as relaxation along a fic-
titious field (RAFF) may have the potential to generate more
FIGURE 4: In vivo T1q-relaxation times. Whisker box plot show- contrast between infarct and remote tissue for a given B1
ing the T1q-values for DCM patients and healthy controls The amplitude and SAR level.29
mean T1q-relaxation time in DCM patients is significantly higher
(55.2 6 2.7 ms) compared with healthy myocardium (51.5 6 In conclusion, we demonstrate the feasibility of endog-
1.2 ms) (P 5 0.0024). enous detection of diffuse myocardial fibrosis with T1q-
MRI. Ex vivo and in vivo experiments showed promising
to ECV at distinguishing between controls and cardiomyop- results, and a significant correlation between the T1q-relaxa-
athy patients.23 In this study, no significant relation between tion time and the ECV-fraction was found in DCM
native T1- and T1q-relaxation time was found. This is unex- patients. Native T1q-mapping requires no separate pre- and
pected, because both the native T1 and the T1q-relaxation
time are increased in the DCM patients, and both relaxa-
tion parameters have shown significant correlation with col-
lagen volume fraction on histology, as was shown for T1q in
this study, and for native T1 in a study by Bull et al.17
Main limitation of the study design is the relatively
small number of patients, and the fact that only DCM
patients were included in the study. Although the results in
this study in a small group of patients are promising, much
larger studies in different cardiac diseases, and different dis-
ease stages, are necessary to investigate the potential of T1q-
mapping for the detection of diffuse myocardial fibrosis.
In addition, there was no significant relation between
native T1- and ECV-fraction. Because we did find a signifi-
cant correlation between T1q- and ECV in this study, and
although the patients sample size is small in our study, these
results suggest that the T1q-relaxation time is potentially a
better predictor of the ECV-value than the native T1-value.
It is important to keep in mind that T1- and T1q-relaxation
are different tissue parameters, affected by other relaxation
mechanisms.
Patients with heart failure commonly suffer from con-
comitant renal impairment and use of a gadolinium based con-
trast agent might, therefore, be contra-indicated in these
patients.28 Because the T1q-mapping method does not require
the use of a contrast agent, this method could function as an
alternative in these patients. Another advantage of T1q-mapping
over ECV-mapping, is that T1q-mapping only requires a single FIGURE 5: Correlation between ECV and T1q and T1. a: Correla-
acquisition, which requires one breathhold per slice. This results tion between the ECV-fraction and T1q-relaxation time in vivo
in DCM patients. A significant correlation with a Pearson r of
in a significant decrease in total scan time, because ECV-
0.66 was found (P 5 0.0015). b: No significant correlation was
mapping requires a pre- and postcontrast T1-map, and a 15–20 found between the ECV-fraction and native T1-relaxation time
min waiting time after contrast agent injection. in DCM patients (Pearson r 5 0.03; P 5 0.89).
January 2017 137

postcontrast acquisitions with mandatory delays, and 14. Musthafa HS, Dragneva G, Lottonen L, et al. Longitudinal rotating
frame relaxation time measurements in infarcted mouse myocardium
removes the need for hematocrit measurement. Thus, native in vivo. Magn Reson Med 2013;69:1389–1395.
T1q-mapping is easier to incorporate in a clinical protocol 15. Van Oorschot JW, El Aidi H, Jansen Of Lorkeers SJ, et al. Endoge-
as opposed to ECV-mapping. In further studies, it is impor- nous assessment of chronic myocardial infarction with T(1q)-mapping
in patients. J Cardiovasc Magn Reson 2014;16:104.
tant that a larger patient group is studied, and patients with
other cardiomyopathies should be included, to study the 16. Puntmann VO, Arroyo Ucar E, Hinojar Baydes R, et al. Aortic stiffness
and interstitial myocardial fibrosis by native T1 are independently
changes in T1q. associated with left ventricular remodeling in patients with dilated car-
diomyopathy. Hypertension 2014;64:762–768.
References 17. Bull S, White SK, Piechnik SK, et al. Human non-contrast T1 values
1. Kawara T, Derksen R, de Groot JR, et al. Activation delay after prema- and correlation with histology in diffuse fibrosis. Heart 2013;99:
ture stimulation in chronically diseased human myocardium relates to 932–937.
the architecture of interstitial fibrosis. Circulation 2001;104: 18. Lee S-P, Lee W, Lee JM, et al. Assessment of diffuse myocardial fibro-
3069–3075. sis by using MR imaging in asymptomatic patients with aortic steno-
2. Bondarenko O, Beek A, Hofman M, et al. Standardizing the definition sis. Radiology 2015;274:359–369.
of hyperenhancement in the quantitative assessment of infarct size 19. Witschey WR Jr, Borthakur A, Elliott MA, et al. Artifacts in T1q-
and myocardial viability using delayed contrast-enhanced CMR. weighted imaging: compensation for B1 and B0 field imperfections.
J Cardiovasc Magn Reson 2005;7:481–485. J Magn Reson 2007;186:75–85.
3. Mewton N, Liu CY, Croisille P, Bluemke D, Lima JA. Assessment of 20. Gho JM, van Es R, Stathonikos N, et al. High resolution systematic
myocardial fibrosis with cardiovascular magnetic resonance. J Am Coll digital histological quantification of cardiac fibrosis and adipose tissue
Cardiol 2011;57:891–903. in phospholamban p.Arg14del mutation associated cardiomyopathy.
4. Sibley CT, Noureldin RA, Mudd JO. T1 mapping in cardiomyopathy PLoS One 2014;9:e94820.
at cardiac MR: comparison with endomyocardial biopsy. Radiology 21. Kellman P, Arai AE, Xue H. T1 and extracellular volume mapping in
2012;265724–732. the heart: estimation of error maps and the influence of noise on pre-
5. Flett AS, Sado DM, Quarta G, et al. Diffuse myocardial fibrosis in cision. J Cardiovasc Magn Reson 2013;15:56.
severe aortic stenosis: an equilibrium contrast cardiovascular magnetic 22. Akçakaya M, Basha TA, Weing€artner S, Roujol S, Berg S, Nezafat R.
resonance study. Eur Heart J Cardiovasc Imaging 2012;13:819–826. Improved quantitative myocardial T2 mapping: impact of the fitting
6. Kellman P, Wilson JR, Xue H, Ugander M, Arai AE. Extracellular vol- model. Magn Reson Med 2014. doi: 10.1002/mrm.25377.
ume fraction mapping in the myocardium, part 1: evaluation of an 23. Puntmann VO, Voigt T, Chen Z, et al. Native T1 mapping in differen-
automated method. J Cardiovasc Magn Reson 2012;14:63. tiation of normal myocardium from diffuse disease in hypertrophic
7. Bellin M-F, Van Der Molen AJ. Extracellular gadolinium-based contrast and dilated cardiomyopathy. JACC Cardiovasc Imaging 2013;6:475–
media: an overview. Eur J Radiol 2008;66:160–167. 484.
8. Murphy KP, Szopinski KT, Cohan RH, Mermillod B, Ellis JH. Occur- 24. Wang C, Zheng J, Sun J, et al. Endogenous contrast T1rho cardiac
rence of adverse reactions to gadolinium-based contrast material and magnetic resonance for myocardial fibrosis in hypertrophic cardiomy-
management of patients at increased risk: a survey of the American opathy patients. J Cardiol 2015;66:520–526.
Society of Neuroradiology Fellowship Directors. Acad Radiol 1999;6:
25. Sibley CT, Noureldin RA, Gai N, et al. T1 Mapping in cardiomyopathy
656–664.
at cardiac MR: comparison with endomyocardial biopsy. Radiology
9. Van Oorschot JWM, Gho JMIH, van Hout GPJ, et al. Endogenous 2012;265:724–732.
contrast MRI of cardiac fibrosis: beyond late gadolinium enhance-
26. aus dem Siepen F, Buss SJ, Messroghli D, et al. T1 mapping in
ment. J Magn Reson Imaging 2015;41:1181–1189.
dilated cardiomyopathy with cardiac magnetic resonance: quantifica-
10. Akella S V, Regatte RR, Gougoutas AJ, et al. Proteoglycan-induced tion of diffuse myocardial fibrosis and comparison with endomyocar-
changes in T1rho-relaxation of articular cartilage at 4T. Magn Reson dial biopsy. Eur Hear J Cardiovasc Imaging 2014;16:210–216.
Med 2001;46:419–423.
27. Dass S, Suttie JJ, Piechnik SK, et al. Myocardial tissue characterization
11. Borthakur A, Mellon E, Niyogi S, Witschey W, Kneeland JB, Reddy R. using magnetic resonance noncontrast T1 mapping in hypertrophic
Sodium and T1rho MRI for molecular and diagnostic imaging of artic- and dilated cardiomyopathy. Circ Cardiovasc Imaging 2012;5:
ular cartilage. NMR Biomed 2006;19:781–821. 726–733.
12. Muthupillai R, Flamm SD, Wilson JM, Pettigrew RI. Radiology acute 28. Silverberg D, Wexler D, Blum M, Schwartz D, Iaina A. The association
myocardial infarction: tissue characterization with T1rho -weighted MR between congestive heart failure and chronic renal disease. Curr Opin
imaging — initial experience. Radiology 2004;232:606–610. Nephrol Hypertens 2004;13:163–170.
13. Witschey WR, Zsido GA, Koomalsingh K, et al. In vivo chronic myocar- 29. Liimatainen T, Sorce DJ, O’Connell R, Garwood M, Michaeli S. MRI
dial infarction characterization by spin locked cardiovascular magnetic contrast from relaxation along a fictitious field (RAFF). Magn Reson
resonance. J Cardiovasc Magn Reson 2012;14:37. Med 2010;64:983–994.

ORIGINAL RESEARCH
MRI-Based Computational Hemodynamics

in Patients With Aortic Coarctation Using
the Lattice Boltzmann Methods: Clinical
Validation Study
Hanieh Mirzaee PhD,1* Thomas Henn MS,2 Mathias J. Krause PhD,2
Leonid Goubergrits PhD,3,4 Christian Schumann MS,1 Mathias Neugebauer PhD,1
Titus Kuehne MD,4 Tobias Preusser PhD,1 and Anja Hennemuth PhD1
Purpose: To introduce a scheme based on a recent technique in computational hemodynamics, known as the lattice
Boltzmann methods (LBM), to noninvasively measure pressure gradients in patients with a coarctation of the aorta
(CoA). To provide evidence on the accuracy of the proposed scheme, the computed pressure drop values are com-
pared against those obtained using the reference standard method of catheterization.
Materials and Methods: Pre- and posttreatment LBM-based pressure gradients for 12 patients with CoA were simulat-
ed for the time point of peak systole using the open source library OpenLB. Four-dimensional (4D) flow-sensitive phase-
contrast MRI at 1.5 Tesla was used to acquire flow and to setup the simulation. The vascular geometry was recon-
structed using 3D whole-heart MRI. Patients underwent pre- and postinterventional pressure catheterization as a refer-
ence standard.
Results: There is a significant linear correlation between the pretreatment catheter pressure drops and those computed
based on the LBM simulation, r5:85, P < :001. The bias was -0.58 6 4.1 mmHg and was not significant (P50:64Þ with a
95% confidence interval (CI) of -3.22 to 2.06. For the posttreatment results, the bias was larger and at -2.54 6 3.53
mmHg with a 95% CI of -0.17 to -4.91 mmHg.
Conclusion: The results indicate a reasonable agreement between the simulation results and the catheter measure-
ments. LBM-based computational hemodynamics can be considered as an alternative to more traditional computational
fluid dynamics schemes for noninvasive pressure calculations and can assist in diagnosis and therapy planning.
C oarctation (or narrowing) of the aorta (CoA) is a con-

genital heart disease and accounts for 3–11% of all
birth heart defects.1 The narrowing usually occurs distal to
To evaluate the severity of the coarctation either pre-
operatively or postoperatively, the hemodynamic analysis is
usually performed through measuring the pressure gradient
the carotids and its severity degree is variable. CoA causes across the coarctation site by means of an invasive and
upper-body hypertension and lower-body hypotension.2 expensive cardiac catheterization.4 However, with develop-
Clinical guidelines recommend treatment by surgery, stent ments in combining noninvasive MRI with modeling
placement, or by balloon angioplasty.3 schemes, numerical simulations of hemodynamics could
*Address reprint requests to: H.M., Universit€

atsallee 29, Bremen, 28359, Germany. E-mail: hanieh.mirzaee@mevis.fraunhofer.de
From the 1Fraunhofer MEVIS, Institute for Medical Image Computing, Bremen, Germany; 2Institute for Mechanical Process Engineering and Mechanics,
Karlsruher Institute of Technology, Karlsruhe, Germany; 3Biofluid Mechanics Laboratory, Charit atsmedizin, Berlin, Germany; and 4Non-Invasive
e-Universit€
Cardiac Imaging in Congenital Heart Disease Unit, Charit e-Universit€
atsmedizin, Berlin, and German Heart Institute, Berlin, Germany
This work is part of the EU project CARDIOPROOF (partially funded by the European Commission under ICT-2013.5.2, Grant Agreement: 611232).
This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution
in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made.
C 2016 The Authors Journal of Magnetic Resonance Imaging published by Wiley Periodicals, Inc.
V
on behalf of International Society for Magnetic Resonance in Medicine 139

assist in diagnosis.5,6 Recently, three-dimensional spatial

TABLE 1. Heart Rates and Pressure Measurements
encoding combined with three-directional phase-contrast for CoA Casesa
MRI (4D flow [PC] MRI) has drawn increased attention.7
Four-dimensional (4D) flow MRI offers the ability to mea- MRI Cath
sure and to visualize the temporal evolution of complex
Case HR Psys Pdia HR Psys Pdia
blood flow patterns within an acquired 3D volume. These
4D flow MRI along with anatomical images can be used to 1 65 140 60 66 113 65
extract hemodynamic and wall boundary conditions respec-
2 88 136 58 77 129 74
tively, when performing patient-specific computational fluid
dynamics (CFD) simulations. In this regard, several authors 3 65 146 66 66 95 54
used similar techniques to simulate patient-specific hemody- 4 96 128 62 84 120 82
namics.8 We further add that based on imaging techniques 5 75 128 62 76 122 74
such as Doppler ultra sound and flow MRI, pressure com-
6 62 140 75 65 126 65
putation schemes have been suggested.9–11 However, pres-
sure gradients obtained from Doppler ultra sound suffer 7 65 158 83 61 149 73
from overestimation and those based on flow MRI tend to 8 68 140 55 54 111 53
underestimate pressure drop values. 9 80 128 62 74 111 69
In conventional computational studies, CFD schemes 10 85 153 72 70 114 70
are based on Navier-Stokes equations and are performed on
11 75 118 63 81 110 66
body-fitted volume meshes. In this work, however, our
hemodynamics computation is based on an approach known 12 74 196 88 78 144 73
a
as the lattice Boltzmann methods (LBM).12 The LBM is a Heart rates (HR) and pressure measurements (in mmHg) dur-
ing MRI and catheterization (Cath). Psys is the systolic pres-
relatively new development in CFD and even newer in the
sure, and Pdia is the pressure at diastole. For the MRI
field of blood flow simulation.13–16 They are easy and acquisitions, cuff pressures are reported.
straightforward to implement, benefit from an automated
and efficient mesh preprocessing, and due to their local
angioplasty and case 4, which was not treated as the
operations, are an ideal candidate for a highly scalable paral-
catheterization-based pressure drop (<20 mmHg) together with an
lelization.17 These promising features motivated us to use
analysis of the anatomy (stenosis degree) and other parameters
LBM as the method of choice for the purposes of this work.
(e.g., hypertension) did not assign this patient to the treatment
Previously, Henn et al investigated a patient-specific
group according to the clinical guidelines. The study was approved
blood flow simulation using the lattice Boltzmann methods by the institutional Research Ethics Committee, following the ethi-
for one coarctation case and showed how realistic pressure cal guidelines of the 1975 declaration of Helsinki. Written
values can be obtained when sufficient spatial and temporal informed consent was obtained from the participants and their
resolutions are used.18 In this work, the aim was to demon- guardians, where applicable.
strate the fidelity of LBM for treatment planning through
considering a versatile coarctation patient cohort and pre- MRI
liminary validation against catheter measurements. MRI was done with a whole-body 1.5 Tesla MR scanner Achieva
R 3.2.2.0 using a five-element cardiac phased-array coil (Philips
Materials and Methods Medical System, Best, Netherlands). Anatomy of the aorta was
acquired by a navigator-triggered 3D WH MRI sequence in end
Patient Data
The data include 12 patients (5 males, 7 females), all with clinical diastole. The sequence parameters were: field of view (FOV)
indication for catheterization, with the age ranging from 14 to 62 212 3 212 3 121 mm, matrix size 320 3 320, 76 slices, acquired
years, and mean and standard deviation of 26 6 15 years. Patients voxel 0.66 3 0.66 3 3.2 mm, reconstructed voxel 0.66 3 0.66 3
were first scanned for 3D whole-heart (WH) and 4D flow MRI. 1.6 mm, repetition time (TR) 4.0 ms, echo time (TE) 2.0 ms, flip
Then catheterization and pressure measurements were performed. angle (FA) 90 8, and number of signal averages 3. The scan time
The averaged time lapse between MRI acquisition and the cathe- was 8 min.
terization procedure was 4 days (range between 1 day and 4 The 4D flow MRI of thorax was performed using an aniso-
T1 weeks). Table 1 provides the heart rates and pressure measurements tropic 4D segmented k-space phase contrast gradient echo sequence
of these 12 cases during the MRI acquisition and catheterization. with retrospective electrocardiographic gating but without navigator
F1 Except for cases 6, 7, and 11 (Fig. 1) with native coarctation, the gating of respiratory motion to minimize acquisition time. The
remaining cases were due to recoarctation. All CoAs were treated sequence parameters were: FOV 180 3 216 3 75 mm, matrix size
by an implantation of a bare Cheatham platinum (CP) stent pre- 100 3 128, 30 slices, acquired voxel 2.5 3 2.5 3 2.5 mm, recon-
mounted on a BIB (balloon in balloon) catheter (NuMED, Nich- structed voxel 1.7 3 1.7 3 2.5 mm, TR 3.5 ms, TE 2.2 ms, FA 5 8,
olville, NY), except case 9, which was treated by balloon 25 reconstructed cardiac phases, velocity encoding 4.0 m/s, and

Mirzaee et al.: MRI-Based Lattice Boltzmann Hemodynamics
C
O
L
O
R
FIGURE 1: a: Pretreatment coarctation geometries. b: Posttreatment geometries. Case 4 was not treated. Treatment was per-
formed through a stenting procedure to expand the narrowed region in the aortic arch.
number of signal averages 1. Scan time varied between 9 and Hemodynamic Computation with the Lattice
14 min, depending on the size of the patient’s chest. The high Boltzmann Methods
velocity encoding in all three directions was used to avoid phase The proposed hemodynamic modeling scheme in this work is
wraps in the presence of stenosis forming complex 3D flow. based on the lattice Boltzmann methods. Rather than relying on
the classical Navier-Stokes equations as in conventional CFD solv-
Segmentation, Flow Analysis, and Posttreatment ers, LBM is a special finite-difference discretization of the simpli-
Geometry Reconstruction fied Boltzmann equation from kinetic theory, which describes
Segmentation of the aorta was performed on the 3D WH images transport phenomena in a mesoscopic scale. Quantities of interest,
using level set methods similar to Goubergrits et al.19 Figure 1a such as velocity uðx; tÞ and pressure pðx; tÞ, where x is a point in
depicts the 12 preoperative geometries used in this work. space and t represents time, are computed indirectly through simu-
The 4D flow MRI was analyzed using the MEVISFlow soft- lating the dynamics of particle distributions. Fundamental to an
ware (Fraunhofer MEVIS, Bremen, Germany). After the routine LBM is a uniform lattice which defines the possible paths that fic-
preprocessing of the PC MRI data including phase-offset error cor- titious particles are allowed to take. Central to the scheme are the
rection and antialiasing, the extracted anatomy was fused with the streaming and collision operations. For instance, in a three-
flow data. For this purpose, first a PC MR angiography (PC dimensional implementation, a typical scenario is shown in Figure
MRA) image was computed and then coregistered with the anato- 3 where particles can stream in 19 different paths (the so-called F3
my image. As both anatomy and flow measurements were acquired D3Q19 LBM scheme). Collision can be viewed as relaxation
in the same session using gated imaging sequences, only small per- toward local equilibrium which should satisfy conservation of
turbations occurred between the two which could be accounted for mass, momentum, and energy.
using rigid registration techniques.20 Based on this fusion, the nec- In one time step, particles can move along the lattice only
essary flow information required for CFD modeling could be in the given lattice directions. The distribution functions fi ðx; tÞ
extracted. More specifically, flow rate curves in the ascending aorta are defined at each grid point x and time t and represent the
just above the sinotubular junction and in the descending aorta are likelihood of particles moving along the certain directions i: The
F2 computed. Figure 2 demonstrates this processing pipeline. LBM iteration with Bhatnagar-Gross-Krook (BGK) collision
The posttreatment corresponding geometries can be seen in operator is
Figure 1b. Posttreatment MRI anatomy data were not available for
1 eq

this patient cohort; therefore, the treated geometry was recon- fi ðx1ei dx; t1dtÞ2fi ðx; tÞ52 fi ðx; tÞ2fi ðu; qÞ (1)
s
structed using projection image data acquired during catheteriza-
tion similar to Goubergrits et al.8 These images provide an where fei g are the set of discrete lattice direction vectors defining a
accurate (0.1 mm resolution) measurement of the posttreatment local neighborhood of a grid point.21 dx and dt are the lattice
diameters. spacing and time step. s is a relaxation parameter and is s5m=cs2 1
January 2017 141


P
qu5 i fi ei , and the pressure is given by p5qcs2. The LBM
reproduce the results of the Navier-Stokes equations in the nearly
incompressible limit and is second-order accurate for the velocity
and first-order accurate for the pressure distributions in the body
of the fluid.23
Simulation Setup
The computational domain, in our case the aorta, was divided into
a uniform lattice. Additionally, it was further divided into blocks
that were processed by separate processors for parallelization pur-
poses. Figure 4 demonstrates a LB-based discretization. A velocity F4
boundary condition as given by Bouzidi et al with a Poiseuille flow
profile based on the measured flow volume at the ascending aorta
opening as shown in Figure 2 was used.24 A smooth startup phase
was added to suppress undesired pressure fluctuations. Due to the
limited resolution of the 4D flow MRI, the flow rates at the side
branches are prone to noise and could not be extracted with
enough accuracy. Therefore, the difference between flow rates in
the ascending and descending aorta was distributed between side
branches of the aortic arch according to their cross sectional area.
The boundary condition at the descending aorta was set as a stress-
free condition, i.e., the pressure is fixed to zero. Moreover, a no-
slip (velocity zero) Bouzidi boundary condition was considered at
the walls. The same boundary conditions are applied to the post-
treatment geometries.
As for the blood, it is assumed to be Newtonian with a
density of 1000 kg=m3 and a kinematic viscosity of 0:00431023
m2 =s. In addition, a D3Q19 BGK-LBM, supported by a Smagor-
insky turbulence model with constant 0.12 was applied to account
for the relatively coarse spatial resolution.25 The simulations are
done using the open source lattice Boltzmann library OpenLB
C
O (www.openlb.net). A fixed spatial resolution of 0:00130:0013
L 0:001m was used in all the cases and the computation times were
O on average around 30 min for each case on Intel Core i7-2600k
R
FIGURE 2: Extraction of flow rate curves at the inlet (red con-
CPU @ 3.4 GHz. Computations were done using 4 cores and no
tour) and outlet (green contour) of the aorta geometry for sim- hyperthreading.
ulation setup. a: The 3D WH MRI (anatomy). b: PC MR
angiography (PC MRA). c: Anatomy after alignment with PC
MRA through registration. d: The segmented geometry. e,f:
Represent the magnitude of the velocity. g: Red: inlet flowrate
curve. Green: outlet flowrate curve. The flow is distributed into
the branches by subtracting the inlet and outlet flow rate at
the time point of peak systole (red vertical line).
1
2 with m the kinematic viscosity and cs being the lattice speed of
eq
sound. fi is the equilibrium distribution function and is a low
Mach number approximation to the Maxwellian distribution
defined as,

eq 9 3
fi ðu; qÞ5 xi q 113ei :u1 ðei :uÞ2 2 u2 (2)
2 2
where xi are weight factors that can be found in literature.22

The hydrodynamic density q, and the macroscopic velocity u are
determined from the particle distribution functions using the laws
P FIGURE 3: Lattice configuration D3Q19 in three dimensions
of conservation of mass and momentum, q5 i fi and with 19 discrete velocity directions.

C
O
L
O
R
FIGURE 4: Lattice Boltzmann voxelization for a patient-specific aorta geometry. a: Blocks are for parallelization and are distributed
between processors. Each block normally contains between 1000 to 10,000 grid cells. b: The boundaries are fitted by taking the
exact distances along the 19 paths from each voxel which is inside the computational domain and close to the wall. c: Streamline
visualization of the simulated velocity.
Simulation Outcome and Pressure Computation mean flow rate at the outlet (descending aorta) was 192 6
At the end of the simulation, a velocity vector field was generated 58 mL/s. A statistical analysis of pressure and heart rate data
corresponding to the time point of peak systole. This velocity field shown in Table 1 found no significant differences for heart
was then used to compute the pressure field using the pressure rates during MRI acquisition and catheterization
Poisson equation (PPE) as described in Krittian et al and Meier (71.9 6 12.8 versus 68.5 6 10.5 bpm; P 5 0.112), diastolic
et al.10,11 As can be noticed, we did not use the pressure values
pressure (66.7 6 12.4 versus 62.1 6 12.3 mmHg;
computed directly from the LB simulation. Henn et al demonstrat-
P 5 0.366), and mean aortic pressure (95.5 6 12.9 versus
ed that a very high spatial resolution is required to get pressure val-
89.5 6 10.9 mmHg; P 5 0.129). The peak systolic pressure
ues compared with the reference measured values.18 In our work,
however, as we are adhering to a fixed spatial resolution for effi-
acquired during MRI was, however, significantly higher
ciency purposes, we rely on the higher order of convergence of the than the one measured by the catheterization (143.9 6 19.8
velocity field in LBM and use this for pressure field computations versus 112.8 6 14.6 mmHg; P 5 0.001). A comparison of
using the PPE.
TABLE 2. Pre- and Posttreatment Pressure Dropsa
The statistical analysis in this work was carried out using Excel Case Cath Pre Cath Post LBM Pre LBM Post
data analysis toolbox and are expressed as mean 6 standard devia-
tion (SD). Effects were considered significant at P < 0.05. Normal-
01 25 9 21 7
ity tests using normal probability plots was performed to ensure
the normal distribution of the data. The Pearson’s correlation coef- 02 30 1 35 1
ficient was used to compute the linear relation between the mea- 03 25 5 30 3
surement and computed results. The agreement between the values
04 11 – 9 –
was investigated using Bland-Altman plots.26 The difference
between the methods was analyzed using a paired t-test for com- 05 15 2 14 5
paring the means of the samples and the suitable sample size was 06 18 3 11 4
determined using a power analysis with the following parameters: 07 15 0 23 7
two-tails a 5 .05, power 5 1-b 5 .8, effect size 5 means SD difference
5
1.08. In computing the effect size, a 5 mmHg difference in means 08 16 0 16 0
between measurement techniques was assumed with a SD of pres- 09 11 0 12 4
sure drops measured with catheter at 4.6 mmHg. Based on these 10 15 1 15 5
values, the required sample size for this study was 9 patients.
11 17 0 16 9
Results 12 12 0 15 4
a
The mean flow rate at the inlet (ascending aorta) was 389 6 Ascending-descending aorta pressure drops (mmHg) computed
111 mL/s which corresponds to a mean Reynolds number of for the cases in Figure 1 using catheter (Cath) and lattice Boltz-
mann methods (LBM), respectively.
5299 6 1675 for a mean diameter of 26.4 6 5.18 mm. The
January 2017 143

FIGURE 5: Bland-Altman plots demonstrating the agreement between pressure gradients measured by catheter and LBM. a: pre-
treatment. b: posttreatment. Reference lines are mean and 6 1.96 3 SD.
the pre- and posttreatment pressure drops between the cath- coarctation of the aorta. LBM has been used in a wide range
eter measurements and LBM-based computations are proof geometries in medical applications and a few studies have
T2 vided in Table 2. also reported on the agreement between LBM and Finite-
Considering the pretreatment results, the mean pres- element-based CFD.27,28
sure drop measured by catheter was 17.5 6 6 mmHg and Given the results in the previous section, we observed
computed based on the LBM was 18 6 7.8 mmHg. Figure that the bias between the pretreatment catheter and LBM
F5 5a demonstrates the agreement between pressure gradients pressure gradient values was small. On the other hand, we
using a Bland-Altman plot. There is a significant relation- noticed that the bias was much larger in the posttreatment
ship between the pretreatment catheter pressure drops and results, but with a comparable standard deviation to pre-
those computed based on the LBM simulation, rð10Þ5:85, treatment findings. For the posttreatment LBM values, case
P < 0:001. The bias in Figure 5a is -0.58 6 4.1 mmHg 7 and 11 produced the highest differences comparing to the
with a 95% confidence interval (CI) of -3.22 to 2.06 catheter measurements. It is not immediately clear to the
mmHg based on a t-distribution with 11 degrees of free- authors why these cases have resulted in higher pressure
dom. The lower and upper limits of agreement are -8.7 drops compared with their catheter counterparts and more
(95% CI: -13.45–-3.94) and 7.5 (95% CI: 2.75–12.25) investigation is required. If we, however, remove these two
mmHg correspondingly. The paired t-test indicated no sig- outliers, the bias will reduce to -1.33 6 2.5 mmHg.
nificant difference between the pretreatment catheter and When comparing the modelling results with the clini-
LBM pressure drop values ðP50:64). cal reference values, we should also take into account the
Regarding the posttreatment results, the mean pressure sources of uncertainty and discrepancies in measurement
drop measured by catheter was 1.9 6 2.84 mmHg and com- and simulation during the whole processing pipeline. In our
puted based on the LBM was 4.45 6 2.62 mmHg. As can study, these can, among others, be generally related to (1)
be noticed, treatment resulted in a reduced pressure drop catheter measurements, (2) MRI postprocessing, and (3)
computed by means of LBM for all the cases. The agree- simulation setup.
ment between pressure gradients is demonstrated in Figure For this work, catheter measurements and the MRI
5b. The bias is -2.54 6 3.53 mmHg with a 95% CI of - acquisitions were performed sequentially and some days
0.17 to -4.91 mmHg based on a t-distribution with 10 apart in sedated and awake patients, respectively. Comparing
degrees of freedom. Based on the paired t-test, P50:04, the heart rate and the mean pressure, there was no statisti-
indicating a difference between the sample means and, cally significant difference between MRI and catheterization,
therefore, rejecting the null hypothesis. The lower and however, the peak systolic pressure in MRI was significantly
upper limits of agreement are -9.46 (95% CI: -13.34–- higher. We note that the difference in systemic pressure
5.58) and 4.37 (95% CI: 0.49–8.25) mmHg, respectively. does not automatically mean a difference in the flow rate
and subsequently in the pressure drop. Nonetheless, further
Discussion investigation needs to be performed to analyze the impact
In this study, MRI flow and anatomy data was used in a of MRI-based boundary setup on the discrepancies between
CFD scheme based on the lattice Boltzmann methods to simulation results and catheter measurements. Moreover, in
estimate pressure gradient values in patients with a clinical routines, systolic pressure drops are reported as

peak-to-peak gradients, even though this nonphysiological In conclusion, despite the aforementioned uncertain-
difference might result in inaccuracies.29 Here, we simulated ties, we demonstrated the accuracy of the lattice Boltzmann
the pressure gradient along the vessel centerline and only at methods for pressure drop calculations in coarctation
the time point of peak systole. Riesenkampff et al used a patients. Our findings are comparable to those obtained by
dynamic pressure calibration technique based on the cathe- Goubergrits et al using a traditional CFD approach.8 Both
ter data to compare peak-to-peak pressure gradients based pre- and posttreatment results indicated a reasonable agree-
on catheterization and 4D flow MRI.30 A similar approach ment with catheter measurements. In case of CoA, stenting
can be performed using a time-harmonic simulation to gain is a recommended therapy to reduce the pressure gradients
more insight on the discrepancies between the methods. across the narrowed vessel segment.35 However, as we men-
This is the subject of future research. We, moreover, note tioned earlier, this procedure is associated with several com-
that the locations in the ascending and descending aorta plications. Recently Neugebauer et al, developed a virtual
chosen for the pressure drop calculation, corresponded only stenting tool that enables an interactive placement of stents
roughly to those measured by catheter. into the narrowed aortic region.36 Combining this with an
In case of MRI data postprocessing, there is a level of LBM-based flow simulation, provides a means for a
uncertainty when constructing the 3D posttreatment geome- computer-aided treatment planning and predicting possible
try based on 2D angiography data, which ultimately con- hemodynamic alterations. Our first validation study in this
tributes to the differences between the postinterventional work, provides evidence on the applicability of such tools in
simulation and catheterization pressure gradients. Moreover, clinical settings.
uncertainties manifest themselves when extracting flow rates
for a vessel cross-section based on the 4D flow MRI data,
which in turn affects the accuracy of the simulation
outcome.31 Acknowledgments
Considering the LBM simulation, its relatively easy Contract grant sponsor: This work is part of the EU project
setup makes it an attractive alternative to traditional solvers. CARDIOPROOF (partially funded by the European Com-
However, modeling blood flow in larger vessels with LBM mission under ICT-2013.5.2, Grant Agreement: 611232).
involves a challenging set of constraints. Among these, accu-
rate open and wall boundary conditions suited to the appli-
cation area and the geometry under investigation is still an
References
active topic of research.32–34 For this work, we adhered to a
1. Lindinger A, Schwedler G, Henze HW. Prevalence of congenital heart
somewhat classic way of imposing boundary conditions, that defects in newborns in Germany: results of the first registration year
is, velocity Poiseuille at the inlet, pressure zero at the outlet, of the PAN study. Klin Padiatr 2010;222:321–326.
and a second order no-slip at the wall boundaries. The aim 2. Cohen M, Fuster V, Steele PM, Driscoll D, McGoon DC. Coarctation
of the aorta. Long-term follow-up and prediction of outcome after sur-
was to keep the computational effort as low as possible to
gical correction. Circulation 1989;80:840–845.
examine the suitability of LBM for clinical applications. In
3. Krueger JJ, Ewert P, Yilmaz S, et al. Magnetic resonance imaging-
this regard, more investigations should be done to study the guided balloon angioplasty of coarctation of the aorta. Circulation
effect of MRI-based inflow on an LBM simulation, similar 2006;113:1093–1100.
to Goubergrits et al.19 Additionally, for the posttreatment 4. Ralovich K, Itu L, Mihalef V, et al. Hemodynamic assessment of pre-
and post-operative aortic coarctation from MRI. Med Image Comput
cases there was no flow information available and, therefore, Assist Interv 2012;15:486–493.
the same boundary conditions used in pretreatment geome- 5. LaDisa JF, Figueroa CA, Vignon-Clementel IE, et al. Computational
tries, were applied. In future studies, MRI measurements simulations for aortic coarctation: representative results from a sam-
after intervention should also be included. pling of patients. J Biomech Eng 2011;133:091008.
Considering the limitations in this study, we note that 6. Menon PG, Pekkan K, Madan S. Quantitative hemodynamic evaluation
in children with coarctation of aorta: phase contrast cardiovascular
the patient data for this work was obtained from one clini- MRI versus computational fluid dynamics. STACOM 2012. LNCS 2013;
cal site. A multicenter study should be carried on to exam- 7746:9–16.
ine the impact of different MR scanners and protocols on 7. Markl M, Kilner PJ, Ebbers T. Comprehensive 4D velocity mapping of
the heart and great vessel by cardiovascular magnetic resonance.
the robustness and accuracy of the results. Moreover, we
J Cardiovasc Magn Reson 2011;7:13.
only had access to 12 CoA patients with the required imag-
8. Goubergrits L, Riesenkampff E, Tevtushenko P, et al. MRI-based com-
ing data. The statistical analysis can certainly benefit from a putational fluid dynamics for diagnosis and treatment prediction: clini-
larger patient cohort. Furthermore, one disease model was cal validation study in patients with coarctation of Aorta. J Magn
considered here. The merits of the LBM for pressure com-
9. Baumgartner H, Bonhoeffer P, De Groot NM, et al. ESC Guidelines
putation should be further investigated through its applica- for the management of grown-ups congenital heart disease. Eur Heart
tion to a diverse range of medical problems. J 2010;31:2915–2957.
January 2017 145

10. Krittian SBS, Lamata P, Michler C. A finite-element approach to the 23. Chen S, Doolen GD. Lattice Boltzmann method for fluid flows. Annu
direct computation of relative cardiovascular pressure from time- Rev Fluid Mech 1998;30:329–364.
resolved MR velocity data. Med Image Anal 2012;16:1029–1037.
24. Bouzidi MH, D’humieres D, Lallemand P, Luo LS. Lattice Boltzmann
11. Meier S, Hennemuth A, Drexl J, Bock J, Jung B, Preusser T. A fast equation on a 2D rectangular grid. J Comput Phys 2001;172:704–717.
and noise-robust method for computation of intravascular pressure
25. Smagorinsky J. General circulation experiments with the primitive
difference maps from 4D PC-MRI data. STACOM 2012. LNCS 2013;
equations. Monthly Weather Rev 1963:91:99–164.
7746:215–224.
26. Bland JM, Altman DG. Statistical methods for assessing agreement
12. Succi S. The lattice Boltzmann equation for fluid dynamics and
between two methods of clinical measurement. Lancet 1986;8476:
beyond. Oxford, UK: Oxford University Press; 2001. 287 p.
307–310.
13. Boyd J, Buick JM. Three-dimensional modelling of the human carotid
27. Geller S. Krafczyk M, Toelke I, Turek S, Hron I. Benchmark computa-
artery using the lattice Boltzmann method. I: model and velocity anal-
tions based on lattice-Boltzmann, finite element, and finite volume
ysis. Phys Med Biol 2008;53:5767–5779.
methods for laminar flows. Comput Fluids 2006;35:888–897.
14. Axner L, Hoekstra AG, Jeays A, Lawford P, Hose R, Sloot PM. Simula-
28. Chopard B, Quared R, Ruefenacht D, Yilmaz H. Lattice Boltzmann
tions of time harmonic blood flow in the Mesenteric artery: comparing modeling of thrombosis in giant aneurysms. Int J Mod Phys C 2007;
finite element and lattice Boltzmann methods. Biomed Eng Online 18:712–721.
2009;8:23.
29. Nishimura RA, Carabello BA. Hemodynamics in the cardiac catheteriza-
15. Krafczyk M, Cerrolaza M, Schulz M, Rank E. Analysis of 3D transient tion laboratory of the 21st Century. Circulation 2012;125:2138–2150.
blood flow passing through an artificial aortic valve by lattice-
Boltzmann methods. J Biomech 1998;453:462. 30. Riesenkampff E, Fernandes JF, Meier S, et al. Pressure fields by flow-
sensitive 4D, velocity-encoded CMR in patients with aortic coarcta-
16. Cosmin N, Itu LM, Suciu C. GPU accelerated blood flow computation tion. JACC Cardiovasc Imaging 2014;7:920–926.
using the lattice Boltzmann method. In: High Performance Extreme
Computing Conference (HPEC), Waltham, Massachusetts, 2013. p 31. Nordmeyer S, Riesenkampff E, Crelier G, et al. Flow-sensistive four-
1–6. dimensional cine magnetic resonance imaging for offline blood flow
quantification in multiple vessels: a validation study. J Magn Reson
17. Fietz J, Krause MJ, Schulz C, et al. Optimized hybrid parallel lattice Imaging 2010;32:677–683.
Boltzmann fluid flow simulations on complex geometries. LNCS 2012;
7484:818–829. 32. Carver HB, Nash RW, Bernabeu MO, et al. Choice of boundary condi-
tion and collision operator for lattice-Boltzmann simulation of
18. Henn T, Heuveline V, Krause M, Ritterbusch S. Aortic coarctation sim- moderate-Reynolds-number flow in complex domains. Phys Rev E
ulation based on the lattice Boltzmann method: benchmark results. Stat Nonlin Soft Matter Phys 2014;89:023303.
STACOM 2012. LNCS 2013;7746:34–43.
33. Latt J, Chopard B, Malaspinas O, Deville M, Michler A. Straight veloc-
19. Goubergrits L, Mevert R, Yevtushenko P, et al. The impact of MRI- ity boundaries in the lattice Boltzmann method. Phys Rev E Stat Non-
based inflow for the hemodynamic evaluation of aortic coarctation. lin Soft Matter Phys 2008;77:056703.
Ann Biomed Eng 2013;41:2575–2587.
34. Boyd J, Buick JM, Cosgrove JA, Stansell P. Application of the lattice
20. Mirzaee H, Hennemuth A. Noninvasive measurement of intravascular Boltzmann method to arterial flow simulation: investigation of bound-
pressure gradients based on 3D anatomy and 4D flow image fusion. ary conditions for complex arterial geometries. Australas Phys Eng Sci
In: Proceedings of the 23rd Annual Meeting of ISMRM, Toronto, Can- Med 2004;27:207–212.
ada, 2015.
35. Feltes TF, Bacha E, Beekman RH. Indication for cardiac catheterization
21. Bhatnagar PL, Gross EP, Krook M. A model for collision processes in and intervention in pediatric cardiac disease. Circulation 2011;123:
Gases. I. Small amplitude processes in charged and neutral one- 2607–2652.
component systems. Phys Rev Lett 1954;94:511–525.
36. Neugebauer M, Gl€ ockler M, Goubergrits L, Kelm M, Kuehne T,
22. Chopard B, Droz M. Cellular automata modeling of physical systems. Hennemuth A. Interactive virtual stent planning for treatment of coarc-
Cambridge: Cambridge University Press; 1998. 341 p. tation of the aorta. Int J Comput Assist Radiol Surg 2016;11:133–144.

ORIGINAL RESEARCH
Effects of Cortisol on the Heart:

Characterization of Myocardial
Involvement in Cushing’s Disease by
Longitudinal Cardiac MRI T1 Mapping
Charles Roux, MD, MSc,1,2* Nadjia Kachenoura, PhD,1,2,3 Zainab Raissuni, MD,4
Elie Mousseaux, MD, PhD,5,6 Jacques Young, MD, PhD,7,8,9
Martin J. Graves, PhD,10 Christel Jublanc, MD,2,11 Philippe Cluzel, MD, PhD,1,2,3,12
Philippe Chanson, MD, PhD,8,9,10 Peter Kamenicky, MD, PhD,8,9,10 and
Alban Redheuil, MD, PhD1,2,3,12
Purpose: Cushing’s disease (CD) is associated with alterations in cardiac geometry and function, shown to be reversible
after treatment. Our aim was to study cortisol-related changes in myocardial content in CD at baseline and after treat-
ment using MR myocardial T1 times.
Materials and Methods: This is a longitudinal study performed in 10 patients with active CD matched with 10 hyperten-
sive and 10 healthy controls. All subjects had MR after CD diagnosis and 6 months after cortisol normalization. The 1.5
Tesla MR protocol included left ventricular geometry and function assessment and MOLLI sequences before and after
contrast injection as well as late gadolinium enhancement.
Results: At baseline, native myocardial T1 was significantly higher in CD patients compared with controls and the hypertensive
group (1056 6 139 ms versus 929 6 80 ms, P 5 0.023; 1056 6 139 ms versus 952 6 51, P 5 0.049). After treatment, native and
postcontrast myocardial T1 decreased in CD patients versus controls (1056 6 139 ms versus 832 6 78, P 5 0.006 and 483 6 69
ms versus 395 6 39 ms, P 5 0.010) reaching values even lower than found in controls (P 5 0.038 and P 5 0.001, respectively).
Conclusion: Native myocardial T1 is increased in Cushing’s disease independently from hypertension and notably decreases
after effective treatment, highlighting its potential to detect subclinical diffuse myocardial involvement in this condition.
C ushing’s syndrome results from prolonged exposure to

excessive glucocorticoids. Endogenous Cushing’s syn-
drome is rare and includes adrenocorticotropin (ACTH) -
Cushing’s syndrome and is due to ACTH oversecretion from a
pituitary corticotroph adenoma.1
Cardiovascular complications are the main cause of
dependent and ACTH-independent causes. Cushing’s disease death in CD, even after effective treatment.2,3 Until recent-
(CD) accounts for approximately 70% of endogenous ly, cardiac changes secondary to cortisol excess had been
*Address reprint requests to: C.R., D

epartement d’Imagerie Cardiovasculaire, La Piti
e Salp^
etrière, ICAN Imaging Core Lab, 75013, Paris, France.
E-mail charles.roux@icloud.com
From the 1Institut National de la Sant e et de la Recherche Medicale (INSERM) UMR 7371, UMR_S 1146, Laboratoire d’Imagerie Biom edicale, ICAN Imaging
Core Lab, Paris, France; 2Sorbonne Universit es, UPMC Univ Paris 06, Faculte de Medecine, Paris, France; 3Institute of Cardiometabolism and Nutrition,
ICAN, Paris, France; 4Facult e de m edecine et de pharmacie de Tanger, Maroc; 5Universit e Paris Descartes, INSERM UMR 970, Paris, France; 6Assistance
Publique-H^ opitaux de Paris, H^ opital Europeen George Pompidou, Service de Radiologie Cardiovasculaire, Paris, France; 7Institut National de la Sant e et de
la Recherche M edicale (INSERM) U693, F-94276, Le Kremlin Bic^ etre, France; 8Assistance Publique-H^ opitaux de Paris, Hôpital de Bicêtre, Service
d’Endocrinologie et des Maladies de la Reproduction, F-94276, Le Kremlin Bic^ etre, France; 9Universit
e Paris-Sud, Faculte de M edecine Paris-Sud, UMR-S
693, F-94276, Le Kremlin Bic^ etre, France; 10Radiology, Cambridge University Hospitals NHS Foundation Trust, UK; 11Assistance Publique-H^ opitaux de Paris,
12
Hôpital Piti etrière, service d’endocrinologie, IE3M, Paris, France; and Assistance Publique-H^
e-Salp^ opitaux de Paris, Hôpital Piti
e-Salp^
etrière, D
epartement
d’Imagerie Cardiovasculaire, Paris, France
Drs. Kamenicky and Redheuil contributed equally to this work.

V 147

mostly studied using 2D ultrasound (US)4 and only focused by complete trans-sphenoidal adenoma resection in all patients
on left ventricular (LV) morphology and function.5 Ultra- associated with adjuvant anti-cortisol treatment in 6 patients.
sound studies revealed LV hypertrophy as well as systolic All subjects gave informed written consent. The research pro-
and diastolic dysfunction in CD patients.5–7 Early work tocol was approved by our institutional ethics committee and com-
using echocardiographic-derived integrated backscatter sug- plies with the Declaration of Helsinki.
gests a possible role for fibrosis in the myocardial patho-
Clinical Investigations
physiology occurring in CD; however no histological studies
Before cardiac imaging, patients were admitted to a specialized aca-
to confirm this have yet been performed.8 Fibrosis, either
demic endocrinology center for clinical and laboratory examina-
dense scar fibrosis from myocardial infarction or diffuse tions as previously described.9
interstitial fibrosis, could indeed play a role in cardiac mor-
bidity and mortality in this syndrome. Cardiac MR
In a pilot cardiac MR study of cardiac structure and In patients, a baseline MR exam was performed soon after CD
function, we have recently demonstrated that Cushing’s syn- diagnosis, before any specific treatment and the second exam was
drome is associated with biventricular and left atrial systolic performed after obtaining complete biological remission of CD,
dysfunction and increased LV mass (LVM), reversible after based on the normalization of circulating and excreted cortisol
treatment.9 LV dysfunction and hypertrophy seen in hyper- rates.
cortisolism may be associated with complex myocardial IMAGE ACQUISITIONS. MR examinations were all performed on
composition changes including global water and fat content a 1.5 Tesla (T) system (GEMS, Milwaukee, USA), with an eight-
and also microvascular density and myocyte mass changes channel cardiac-phased array surface coil and ECG gating. All acquis-
beyond the effect of associated hypertension. However, we itions were made during breath-holding and included: (1) 8 to 12
did not detect late gadolinium enhancement (LGE) in any steady-state free precession (SSFP) cine short axis slices, which were
of these patients.9 MR is the reference method to detect acquired using the following parameters: echo time (TE) 5 1.5 ms;
and quantify interstitial myocardial fibrosis using T1 map- repetition time (TR) 5 3.5 ms; slice thickness 5 7 mm; inter-slice
ping techniques.10 T1 mapping techniques have been shown gap 5 1 mm; acquisition matrix 5 288 3 288; spatial resolu-
tion 5 1.31 mm; flip angle 5 75 8. (2) LGE imaging was performed
to be useful for the diagnosis and the assessment of various
using a T1 inversion recovery sequence in both short and long axis
diseases such as cardiac amyloidosis,11 myocardial infarc-
slices during diastole10 min after a bolus IV injection of 0.2 mmol/
tion,12 systemic lupus erythematosus,13 and acute viral myo- R
kg of Gd-DTPA (DotaremV). The acquisition parameters were:
carditis.14 This study aimed at evaluating the usefulness of TE 5 1.5 ms; TR 5 5.3 ms; slice thickness 5 8 mm; acquisition
T1 mapping for the detection of myocardial alterations in matrix 5 224 3 160; spatial resolution 5 0.86 mm; and typical
CD, and describing their evolution after treatment. inversion time 5 250 ms. (3) T1 mapping before contrast and at
15 min after contrast injection, which was acquired on a mid-
Material and Methods ventricular short-axis slice using a Modified Look-Locker Inversion
Recovery (MOLLI) sequence. The MOLLI sequence was performed
Study Population using a 3(3)3(3)5 scheme to obtain 11 points on the recovery curve,
Eighteen Cushing’s syndrome patients were recruited between Sep-
within a single 17 heartbeat breath-hold. Imaging parameters were:
tember 2009 and March 2013 as previously reported.9 Inclusion
flip angle 5 35 8; TR 5 3.4 ms; TE 5 1.47 ms; number of
criteria were age between 18 and 60 years, and Cushing’s syndrome
excitations 5 0.5; matrix size 5 256 3 170; field of view 5 360 mm;
of endogenous origin newly diagnosed and untreated. Cushing’s
pixel spacing 5 1.25 3 1.25 mm; slice thickness 5 8 mm. (4) For dia-
syndrome was diagnosed according to the usual clinical and biolog-
stolic function assessment, phase contrast acquisitions of transmitral
ical criteria, including elevated urinary free cortisol (UFC) excre-
flow and basal myocardial velocities were acquired as described in a
tion, loss of the circadian plasma cortisol pattern, and lack of
previous study.15
cortisol suppression during the overnight 1-mg dexamethasone sup-
pression test. Among these 18 patients, the last 10 consecutive ANALYSIS OF LV MORPHOLOGY AND FUNCTION. Standard
patients with CD were included from February 2011 to March cine SSFP short-axis data were analyzed using a dedicated software
R
2013 in this analysis (age 36 6 15 years, 9 women), as they were (QMassV V7.0, Medis, Netherlands) which enabled a semi-
further evaluated by T1 mapping in MR. automated tracing of endocardial and epicardial contours for both
Ten healthy and asymptomatic volunteers aged 35 6 8 years end-systolic and end-diastolic phases and subsequently the estimation
(7 women) with no history of heart disease and no cardiovascular of LVM as well as end-systolic (ESV) and end-diastolic (EDV) vol-
risk factors were enrolled. Ten asymptomatic hypertensive volun- umes. Papillary muscles were excluded from left LVM and were
teers (HTN group) aged 39 6 9 (7 women) with no other cardio- included in LV volume. LV ejection fraction (LVEF) and stroke vol-
vascular risk factor were also included. They were similar in age, ume were calculated.
gender and heart rate to CD patients. CD patients underwent a For diastolic function assessment, phase contrast acquisitions
similar investigation protocol at inclusion and on average 5.8 6 3.2 of transmitral flow and basal myocardial velocities were analyzed as
months after the first scan. The biological remission was obtained described in Bollache et al,15 providing: E, early diastolic E-wave

Roux et al.: Effect of Cortisol on the Heart Using T1 Mapping
C
O
L
O
R
FIGURE 1: Cardiac MR images in a patient with Cushing’s disease. (a) Cine SSFP diastolic image. (b) Absence of late gadolinium
myocardial enhancement. (c) Illustration of the segmentation of a native T1 map: epicardial and endocardial borders traced on an
original grayscale MOLLI image with exclusion of left ventricular cavity and epicardial fat. (d) Superimposition of the contours
traced on the T1 map.
peak of transmitral flow; A, late diastolic atrial contraction A-wave map and corrected manually to exclude pixels from the LV cavity
peak; and e0 , early diastolic e0 peak of myocardial velocity. and epicardial fat (Fig. 1), resulting in mean myocardial T1 values. F1
This region of interest was then automatically divided into six seg-
MYOCARDIAL TISSUE CHARACTERIZATION. For myocardial ments starting at the superior right ventricular insertion point.
tissue characterization, an investigator with 5 years of experience in Raw images and error maps were examined for artifacts and corre-
cardiac MR reader blinded to clinical data performed a visual anal- sponding segments were excluded.
ysis of LGE images and noted presence or absence of delayed To take into account potential heart rate variations in patients
enhancement. In addition, MOLLI datasets were transferred for after treatment, T1 values were corrected as previously described by Lee
off-line analysis using a custom software that enabled T1 map esti- et al.17 Phantoms containing agarose gel were scanned using the previ-
mation, according to the previously described16 exponential fitting ously described MOLLI sequence with simulated heart rate. Then T1
for each pixel (y): y5A-B exp(-t/T1*), where t is the inversion values from phantoms at different heart rates were fitted using a second
time. Least square estimates of model parameters were obtained order polynomial function to ideal theoretical T1 values as obtained by
using the Levenberg-Marquardt algorithm. Finally, the corrected a spin echo-sequence. T1 values were then corrected according to heart
T1 was calculated as T15T1*(B/A-1). rate using the phantom results. A region of interest was positioned in the
For all subjects, myocardial borders were first manually LV cavity to estimate blood pool T1. This processing was performed on
drawn on the original grayscale MOLLI image acquired with the both native and postcontrast T1 maps resulting in myocardial and blood
inversion time providing the best contrast between myocardium pool T1 values used for the estimation of the partition coefficient (k),
and LV cavity. It was then superimposed on the corresponding T1 calculated using a previously described formula: k 5 (postcontrast
January 2017 149

R1myo – native R1myo)/(postcontrast R1LV-Cavity - native R1LV-cavity) with controls (Fig. 2). Native T1 times were not significantly cor-
R1 5 1/T1.18,19 T1 mapping analysis resulted in native T1 values related to LVM (P 5 0.981), LVMi (P 5 0.802), or LVM/
(native T1) as well as postcontrast T1 values (T1 after injection) and par- EDV (P 5 0.832) nor to age (P 5 0.907).
tition coefficient (k). These parameters were calculated for controls and At baseline, patients, HTN, and controls had similar
for CD patients before and after treatment. LVEDV while LVESV tended to be higher in patients
resulting in a significantly lower LVEF in CD patients than
All continuous variables are expressed as their mean and standard in controls and HTN (P 5 0.049; P 5 0.028, respectively).
deviation and dichotomous variables as frequencies and percentages Moreover, patients tended to have higher LVM than con-
when relevant. An unpaired nonparametric means comparison test trols, although this relationship was not statistically signifi-
(Mann-Whitney) was used for comparisons with controls. Non- cant (P 5 0.104). Hypertensive individuals and CD patients
parametric multiple means comparison test (Kruskal-Wallis) was were not statistically different for LVM (p 5 0.496). At
used for comparisons across groups, then for paired comparisons, baseline, LVMi and LVM/LVEDV were not significantly
we conducted a nonparametric Wilcoxon signed-rank test. different between patients, HTN, and controls (P 5 0.077
Association between variables was assessed and Spearman’s and P 5 0.516). No patient had regional wall motion
rank correlation coefficient was provided. A P value < 0.05 indicat- abnormalities at baseline, and decreased LVEF was related
ed statistical significance. Statistical analysis was performed using to global hypokinesia. No patient had LGE. MR diastolic
JMP 11 SAS software. parameters did not differ between patients, HTN, and con-
trols at baseline.
Results
Population Effects of Treatment and Cortisol Normalization
Controls, HTN, and CD patients’ clinical and biological After treatment, native T1 times and T1 times after injec-
T1 data at baseline and follow-up are summarized in Table 1. tion decreased significantly in CD patients (respectively,
Body mass index and surface area, and systolic and diastolic P 5 0.006, P 5 0.01), reaching values significantly lower
blood pressures were higher in CD patients at baseline than than those found in controls (respectively, P 5 0.038,
in controls as expected but did not differ from the HTN P 5 0.001). Moreover, k was significantly higher in the
group. At baseline, three CD patients met biochemical crite- patient group after treatment than in controls (P 5 0.049).
ria of diabetes. On oral glucose tolerance test, two patients There were no significant changes in patients in
had impaired glucose tolerance. Two patients were treated LVEDV or LVESV after treatment (P 5 0.275 and
with oral antidiabetic drugs. One patient had dyslipidemia P 5 0.769). However, stroke volume and LVEF tended to
and four patients had hypertension, among whom three be increased without reaching statistical significance
received antihypertensive treatment. After normalization of (P 5 0.193). However, cortisol normalization posttreatment
cortisol (the normal range of UFC excretion is 10–65 g/24 h was associated with a 26% reduction in LVM (P 5 0.004)
(27–179 nmol/24 h) in large cohort study20, correction of and with eccentric LV geometry as shown by decreased
glucose tolerance impairment was achieved in all patients LVM/LVEDV ratio in patients compared with controls
allowing interruption of antidiabetic treatments. Correction (P 5 0.023).
of blood pressures was observed in all patients leading to
Discussion
discontinuation of antihypertensive therapy in one patient.
This study evaluated myocardial tissue properties in Cush-
Effects of Cortisol on T1 Parameters ing’s disease using noninvasive T1 mapping in cardiac MR.
and LV Function Parameters The main finding of this study is that native myocardial T1
Cardiac morphological and functional parameters, native T1 time is increased in CD patients compared with healthy
T2 times, T1 times after injection, and k are summarized in controls, with marked modification of this tissue property
F2 Table 2 and Figure 2. after cortisol normalization. After treatment, native T1time
At baseline, native T1 times were higher in CD in patients becomes significantly lower than in controls.
patients compared with the other groups and were the only These differences remain significant with an incremental
parameter statistically different across all three groups trend when comparing CD patients to hypertensive individ-
(P 5 0.043). Pairwise post hoc tests showed that native T1 uals instead of healthy controls to adjust for the potential
times were significantly higher in patients at baseline versus effect of hypertension.
controls (P 5 0.023) and versus HTN (P 5 0.049), but In recent years, T1 mapping techniques have demon-
there was no significant difference in T1 times after injec- strated their value in several myocardial diseases.21 This
tion between controls and hypertensive individuals noninvasive and objective technique has been considered to
(P 5 0.650). No differences were found in terms of parti- quantify diffuse myocardial fibrosis or edema and to diag-
tion coefficient k between patients at baseline, HTN, and nose diseases such as acute myocarditis or amyloidosis. In

January 2017
TABLE 1. Clinical, Biological, and Hormonal Parameters for Controls, HTN, and CD Patients before and after Treatment
Controls HTN Cushing Cushing Pre Treat Pre Treat. Pre Treat.
Pre Treat. Post vs. vs. HTN vs. Post
Treat. Controls Treat.
(N 5 10) (N 5 10) (n 5 10) (n 5 10) (P) (P) (P)
Clinical parameters
Age (years) 35 6 8 39 6 9 36 6 15 37 6 14.5 0.622 0.3 -
Gender (female/male) 7/3 8/2 9/1 9/1 0.582 0.894 -
2 a
Body surface area (m ) 1.73 6 0.21 1.80 6 0.17 1.88 6 0.21 1.86 6 0.20 0.028 0.406 0.16
2 a
Body mass index (kg/m ) 23.8 6 5.4 25.3 6 3.9 28.7 6 5.5 28.3 6 6.0 0.028 0.151 0.084
a
Systolic blood 98 6 13 121 6 12 120 6 14 117 6 13 0.019 0.705 0.652
pressure (mmHg)
Diastolic blood 74 6 11 76 6 12 75 6 10 71 6 10 0.041a 0.94 0.57
pressure (mmHg)
Heart rate (bpm) 70 6 8 67 6 5 77 6 13 70 6 16 0.173 0.049a 0.08
Biochemical parameters
Fasting glucose (mmol/L) - - 5.24 6 0.80 4.2 6 0.47 - - 0.013a
Triglycerides (mmol/L) - - 1.00 6 0.35 1.37 6 0.70 - - 0.313
Total cholesterol (mmol/L) - - 5.05 6 0.83 5.62 6 0.72 - - 0.382
LDL cholesterol (mmol/L) - - 2.84 6 0.91 3.52 6 0.70 - - 0.148
HDL cholesterol (mmol/L) - - 1.76 6 0.29 1.19 6 0.40 - - 0.461
Hormonal parameters
UFC (mg/24 h) - - 665 6 950 13 6 14 - - 0.004a
a
Statistical significance if P < 0.05.
Stage:

151
Page: 151
152
TABLE 2. Cardiac MR Parameters for Controls, HTN, and CD Patients before and after Treatment
Controls HTN Cushing Cushing Pre Pre Pre Post

Pre Post Treat. vs. Treat. vs. Treat. vs. Treat. vs.
Treat. Treat. Controls HTN Post Treat. Controls
(n 5 10) (n 5 10) (n 5 10) (n 5 10) (P) (P) (P) (P)
T1 mapping
Native T1 time (ms) 929 6 80 952 6 51 1056 6 139 832 6 78 0.023a 0.049a 0.006a 0.038a
T1 time after 463 6 29 448 6 47 483 6 69 395 6 39 0.545 0.174 0.010a 0.001a
injection (ms)
(k) partition 41.2 6 4.2 41.6 6 6.5 44.7 6 11.6 48.3 6 7.0 0.88 0.821 0.148 0.049a
coefficient (%)
LV systolic function
EDV (ml) 121 6 43 137 6 28 129 6 30 143 6 32 0.64 0.197 0.275 0.199
EDV index (ml/m2) 70 6 21 76 6 15 69 6 16 77 6 17 0.94 0.297 0.322 0.289
ESV (ml) 47 6 21 51 6 11 57 6 25 59 6 21 0.364 0.24 0.77 0.199
2
ESV index (ml/m ) 27 6 11 29 6 6 30 6 12 31 6 10 0.406 0.406 0.56 0.406
a
Stroke volume (ml) 72 6 23 86 6 22 75 6 23 84 6 18 0.706 0.054 0.004 0.272
a a
Ejection fraction (%) 62 6 6 62 6 6 56 6 12 60 6 9 0.049 0.028 0.193 0.406
a
Myocardial mass (g) 93 6 19 116 6 26 111 6 22 82 6 22 0.104 0.207 0.004 0.597
2 a
Mass index (g/m ) 53.7 6 8.1 63.8 6 8.7 59.6 6 12.8 48.4 6 12.6 0.364 0.142 0.001 0.174
a
Mass/volume (g/ml) 0.82 6 0.21 0.86 6 0.20 0.87 6 0.14 0.65 6 0.21 0.151 0.690 0.004 0.023a
LV diastolic function
E (cm/s) 37.6 6 8.5 42.6 6 15.3 42.3 6 7.2 47.0 6 13.3 0.38 0.807 0.496 0.153
e’ (cm/s) 6.0 6 2.2 6.4 6 2.7 4.7 6 1.8 5.4 6 1.7 0.341 0.2 0.36 0.732
E/e’ (cm/s) 6.7 6 2.2 12.7 6 8 11.1 6 5.8 8.3 6 2.6 0.223 0.51 0.195 0.266
Stage:
a
Statistical significance if P < 0.05.

Volume 45, No. 1
Page: 152
FIGURE 2: Comparison of myocardial native T1 values (a), myocardial T1 values after injection (b), LVEF (c), stroke volume (d), LV
myocardial mass (e), and mass/EDV (f) in controls versus Pre Treat, HTN versus Pre Treat, and Pre Treat versus Post Treat. Note:
*P, statistical significance if P < 0.05.
such conditions, native T1 reaches a sensitivity of 92%14 atrial dysfunction.9 However, at the tissue level, the determi-
and 90%,11 respectively, to identify patients versus controls. nants of increased ventricular wall thickness and decreased con-
In addition to its potential diagnostic value, tissue character- tractility and relaxation of the myocardium exposed to
ization through T1 mapping could lead to refine manage- increased levels of cortisol remain unclear.23 These observations
ment of disease and therapy evaluation.22 are likely to be linked to complex changes in myocardial com-
Cardiovascular events are the main cause of death in position and metabolism. Cortisol is responsible for both min-
Cushing’s syndrome. Ischemic heart disease and complications eralocorticoid and glucocorticoid effects. The mineralocorticoid
related to cardiac hypertrophy4 are the main causes of morbidi- pathway increases collagen secretion by the activation of fibro-
ty and mortality. Currently, cardiac involvement in patients blasts24 which is why diffuse fibrosis is thought to be a poten-
with CD is assessed using Doppler-echocardiography and tial player in myocardial involvement in Cushing’s disease. In
mainly based on the identification of hypertrophic remodeling addition, the stimulation of MC receptors decreases myocyte
and ventricular dysfunction. Cardiac MR has recently refined contractility and stimulates mitosis, resulting in myocardial
and demonstrated the reversibility of the typical cardiac pheno- hypertrophy and dysfunction.25 Finally, the MC pathway also
type associated with hypercortisolism including significant promotes macrophage recruitment in the extracellular matrix
regional LV wall thickening and right ventricular as well as left and perivascular edema.26 These local effects promoting
January 2017 153

myocardial edema may be enhanced by the systemic hypervole- differential reversal of myocardial tissue content following
mia induced by renal MC action. cortisol suppression. Precise mechanisms remain unclear
Only one echo-based study attempted to assess the but, it has been shown that metabolic syndrome and diabe-
presence of myocardial fibrosis using integrated backscatter tes lead to myocardial steatosis,31 which could cause
in ultrasonography.8 This technique is an indirect way to decreased native myocardial T1 values.32 Lipid accumulation
characterize macroscopic myocardial structural features and within myocardial cells lead to decreased T1 because of the
may thus be very sensitive to potential confounding factors, very brief T1 relaxation time of fat as previously shown in
i.e., microscopic myocardial components such as diffuse Fabry disease32 or depend on off-resonance phenomena and
myocardial fibrosis or edema, also thought to be involved in balanced-SSFP readout.33 Myocardial fat fraction has been
CD pathophysiology.27 In our recent MR study, none of reported to vary within a small range from normals to meta-
the Cushing’s disease patients had focal myocardial fibrosis bolic diseases (diabetes, obesity).34,35 We postulate that
assessed by LGE.9 This shows that, despite higher cardiovas- myocardial steatosis may be present before treatment but
cular risk conferred by the metabolic syndrome induced by not detected as low T1 values, because of the overriding T1
CD, myocardial impairment cannot be explained by silent increase due to increased extracellular matrix (extracellular
infarcts. Furthermore, we found no regional dysfunction water and fibrosis). Indeed, native T1 reflects all compo-
relating to a coronary territory in any patient. nents of the myocardium, including myocytes and the inter-
Native T1 mapping is a very sensitive method to stitium. The increase in relative extracellular water content
detect infraclinical myocardial content alteration. However, could outweigh myocardial steatosis effect on native T1
it is not specific of a pathological or physiological process before treatment. However, radical treatment and the sup-
(i.e., myocardial fibrosis) and may reflect changes in T1- pression of cortisol can lead to a sharp decrease in myocar-
modifying contents such as global water (intra- and extracel- dial water content (highly labile) and, therefore, could
lular) and fat or protein density.28 We found native T1 unmask the presence of steatosis (less labile). Also, we can-
times to be significantly higher in patients versus controls not rule out that, after radical treatment, patients are affect-
and versus HTN at baseline. This significant increase in ed by mild adrenal insufficiency causing a decrease in
native myocardial T1 may be related to a significant increase extracellular myocardial water content.
in global myocardial water content. At baseline, we did not The complex and intricate mechanisms at play at the
observe any significant difference in postcontrast T1 and k tissue level cannot be completely individualized at the mac-
between patients and controls. We could have expected that roscopic level of in vivo cardiac MR using currently avail-
the increase in extracellular space consecutive to the presence able techniques that remain unspecific, but will be explored
of diffuse fibrosis should cause pooling of the contrast and in future studies.
thus reduce myocardial relaxation times, after injection. LVEF at baseline was significantly lower in patients
There are two possible explanations for this observation: (1) versus controls. Myocardial mass tended to be higher in
there are other tissue level changes (cofactors) that result in patients, although it did not reach statistical significance but
an increase of native myocardium T1 and no significant var- decreased after treatment. These findings are consistent with
iation in T1 after injection and k, such as increased intracel- our previous cardiac MR study.9
lular water content. Indeed, native T1 and partition fraction We found no significant difference across groups for
are not interchangeable. T1 after injection and partition diastolic function parameters. To account for the potential
fraction are extracellular markers22 and native T1 is global confounding effects of hypertension also present in CD
myocardial marker. (2) T1 mapping before injection has patients we included a group of matched hypertensive indi-
been shown to have a tight normal range within 2% from viduals in this study. Of interest, where LVM was not signifi-
the mean,29 allowing us to show differences in CD patients cantly different between HTN and CD groups, T1 values
despite our small population in a rare disease. However, it remained significantly increased in the CD group at baseline.
has been shown that, in contrast to native T1 estimates,
more than 20 patients are needed to detect a k change of Limitations
6.3% at 3 Tesla30 (T) due to small errors in each term nec- Our study has two main limitations. First is the small size of
essary for its calculation. Thus, this study may be under- our population, albeit in a rare disease. As shown above, pop-
powered to show such a k difference before and after ulation size is a critical factor for postcontrast enhancement
treatment. T1 mapping studies. It is likely that the reason for the inabil-
After treatment, we found a very significant decrease ity here to show differences in T1 and k after contrast
in native myocardial T1, becoming not only lower than at enhancement before treatment is related to lack of power in
baseline but much lower than in controls. As T1 values conjunction with variations in other potential tissue con-
before and after treatment were not explained by morpho- founders. The hematocrit values necessary to calculate the
logical changes, we postulate that this may be explained by extracellular volumes were not available in all subjects, hence,

not reported but k has been recently shown to correlate high- References
ly with ECV measured with or without the use of hemato- 1. Newell-Price J, Bertagna X, Grossman AB, Nieman LK. Cushing’s syn-
crit. Nevertheless, the striking results obtained for native T1 drome. The Lancet 2006;367:1605–1617.
highlights the potential value of native T1 in this disease as 2. Atkinson AB, Kennedy A, Wiggam MI, et al. Long-term remission
rates after pituitary surgery for Cushing’s disease: the need for long-
already demonstrated in other settings.14,32 The second limi- term surveillance. Clin Endocrinol (Oxf) 2005;63:549–559.
tation is the absence of direct histological proof to be com- 3. Sherlock M, Ayuk J, Tomlinson JW, et al. Mortality in Patients with
pared with T1 values. However, endomyocardial biopsy is Pituitary Disease. Endocr Rev 2010;31:301–342.
not currently recommended in this setting and is not free of 4. Muiesan ML, Lupia M, Salvetti M, et al. Left ventricular structural and
morbidity. Pathophysiological hypotheses, although theoreti- functional characteristics in Cushing’s syndrome. J Am Coll Cardiol
2003;41:2275–2279.
cally plausible, will require further studies including other
5. Pereira AM, Delgado V, Romijn JA, et al. Cardiac dysfunction is
MR techniques to investigate cardiac fat as well as histological reversed upon successful treatment of Cushing’s syndrome. Eur J
proof. However, this study is the first step in understanding Endocrinol 2010;162:331–340.
myocardial involvement in Cushing’s disease using MR. 6. € et al. Assessment of left ventricular dia-
Baykan M, Erem C, Gedikli O,
stolic function and Tei index by tissue doppler imaging in patients
This MR T1 mapping study explored myocardial tissue with Cushing’s syndrome: tissue Doppler imaging in Cushing’s syn-
changes in Cushing’s disease and the effect of radical treat- drome. Echocardiography 2007;25:182–190.
ment. Native myocardial T1 relaxation time is increased in 7. Toja PM, Branzi G, Ciambellotti F, et al. Clinical relevance of cardiac
CD patients in comparison to matched healthy and hyperten- structure and function abnormalities in patients with Cushing’s syn-
drome before and after cure: Cardiac alterations in Cushing’s syn-
sive individuals supporting the potential role of diffuse fibro- drome. Clin Endocrinol (Oxf) 2012;76:332–338.
sis and increased extracellular water in this disease. The 8. Yiu KH, Ajmone Marsan N, Delgado V, et al. Increased myocardial
drastic reduction in T1 times after treatment highlights the fibrosis and left ventricular dysfunction in Cushing’s syndrome. Eur J
Endocrinol 2011;166:27–34.
reversibility of this biomarker but also the complexity of
myocardial changes and the potential role of other variables 9. Kamenicky P, Redheuil A, Roux C, et al. Cardiac structure and func-
tion in Cushing’s syndrome: a cardiac magnetic resonance imaging
affecting T1 times. Further studies will be needed to deter- study. J Clin Endocrinol Metab 2014;99:E2144–E2154.
mine if T1 times and derivatives could act as new cardiovas- 10. Moon JC, Messroghli DR, Kellman P, et al. Myocardial T1 mapping and
cular endpoints to detect cardiac involvement at infraclinical extracellular volume quantification: a Society for Cardiovascular Magnetic
Resonance (SCMR) and CMR Working Group of the European Society of
stages of this disease and to assess the effectiveness of therapy. Cardiology consensus statement. J Cardiovasc Magn Reson 2013;15:92.
11. Karamitsos TD, Piechnik SK, Banypersad SM, et al. Noncontrast T1
Mapping for the Diagnosis of Cardiac Amyloidosis. JACC Cardiovasc
Acknowledgments Imaging 2013;6:488–497.
Contract grant sponsor: French Society of Radiology; Con- 12. Dall’Armellina E, Piechnik SK, Ferreira VM, et al. Cardiovascular mag-
tract grant sponsor: Assistance Publique-Hôpitaux de Paris netic resonance by non contrast T1-mapping allows assessment of
(AP-HP): CORTICOEUR Study severity of injury in acute myocardial infarction. J Cardiovasc Magn
Reson 2012;14:15.
C.R. performed imaging data analysis and quantification,
13. Puntmann VO, D’Cruz D, Smith Z, et al. Native myocardial T1 map-
statistical analysis and drafted the manuscript, N.K. contrib- ping by cardiovascular magnetic resonance imaging in subclinical car-
uted to quantitative image analysis methodology and manu- diomyopathy in patients with systemic lupus erythematosus. Circ
Cardiovasc Imaging 2013;6:295–301.
script revision, Z.R. performed imaging data analysis, E.M.
contributed to methodology, image acquisition supervision 14. Ferreira VM, Piechnik SK, Dall’Armellina E, et al. T1 Mapping for the
diagnosis of acute myocarditis using CMR. JACC Cardiovasc Imaging
and manuscript revision, J.Y. contributed to patient recruit- 2013;6:1048–1058.
ment and clinical data acquisition, M.G. contributed to 15. Bollache E, Redheuil A, Cl ement-Guinaudeau S, et al. Automated left
image acquisition methodology and MRI sequence optimi- ventricular diastolic function evaluation from phase-contrast cardiovas-
cular magnetic resonance and comparison with Doppler echocardiog-
zation, C.J. contributed to patient recruitment and clinical raphy. J Cardiovasc Magn Reson 2010;12:63.
data acquisition, P.C. contributed to manuscript revision,
16. Messroghli DR, Radjenovic A, Kozerke S, et al. Modified Look-Locker
P.C. contributed to clinical study supervision and manu- inversion recovery (MOLLI) for high-resolutionT1 mapping of the
script revision, P.K. contributed to study design and clinical heart. Magn Reson Med 2004;52:141–146.
study supervision, clinical data acquisition and manuscript 17. Lee JJ, Liu S, Nacif MS, et al. Myocardial T1 and extracellular volume
fraction mapping at 3 tesla. J Cardiovasc Magn Reson 2011;13:75.
revision, A.R. contributed to study design and supervision,
18. Kellman P, Wilson JR, Xue H, et al. Extracellular volume fraction map-
imaging data acquisition and analysis and manuscript revi- ping in the myocardium, part 1: evaluation of an automated method.
sion. C.R. received a research grant from the French Society J Cardiovasc Magn Reson 2012;14:63.
of Radiology (Societe Française de Radiologie, SFR). P.C., 19. Kellman P, Wilson JR, Xue H, et al. Extracellular volume fraction map-
P.K., and A.R. received an institutional academic hospital ping in the myocardium, part 2: initial clinical experience.
J Cardiovasc Magn Reson 2012;14:64.
research grant from Assistance Publique-Hôpitaux de Paris
20. Rosmalen JGM, Kema IP, W€ ust S, et al. 24h urinary free cortisol in
(AP-HP): CORTICOEUR Study. large-scale epidemiological studies: Short-term and long-term stability
and sources of variability. Psychoneuroendocrinology 2014;47:10–16.
January 2017 155

21. Puntmann VO, Voigt T, Chen Z, et al. Native T1 mapping in differenti- 29. Piechnik SK, Ferreira VM, Lewandowski AJ, et al. Normal variation of
ation of normal myocardium from diffuse disease in hypertrophic and magnetic resonance T1 relaxation times in the human population at
dilated cardiomyopathy. JACC Cardiovasc Imaging 2013;6:475–484. 1.5 T using ShMOLLI. J Cardiovasc Magn Reson 2013;15:13.
22. Lima JAC. The promise of myocardial fibrosis assessment by T1 map- 30. Liu S, Han J, Nacif MS, et al. Diffuse myocardial fibrosis evaluation
ping. JACC Cardiovasc Imaging 2013;6:485–487. using cardiac magnetic resonance T1 mapping: sample size consider-
ations for clinical trials. J Cardiovasc Magn Reson 2012;14:90.
23. Kim MY, Eiby YA, Lumbers ER, et al. Effects of glucocorticoid expo-
sure on growth and structural maturation of the heart of the preterm 31. Liu C-Y, Bluemke DA, Gerstenblith G, et al. Myocardial steatosis and
piglet. PLoS One 2014;9:e93407 its association with obesity and regional ventricular dysfunction: Evalu-
ated by magnetic resonance tagging and 1H spectroscopy in healthy
24. Fujisawa G, Dilley R, Fullerton MJ, Funder JW. Experimental cardiac African Americans. Int J Cardiol 2014;172:381–387.
fibrosis: differential time course of responses to mineralocorticoid-salt
administration. Endocrinology 2001;142:3625–3631. 32. Sado DM, White SK, Piechnik SK, et al. Identification and assessment
of Anderson-Fabry disease by cardiovascular magnetic resonance
25. Feng X, Reini SA, Richards E, et al. Cortisol stimulates proliferation noncontrast myocardial T1 mapping. Circ Cardiovasc Imaging 2013;6:
and apoptosis in the late gestation fetal heart: differential effects of 392–398.
mineralocorticoid and glucocorticoid receptors. AJP Regul Integr
Comp Physiol 2013;305:R343–R350. 33. Kellman P, Bandettini WP, Mancini C, et al. Characterization of myo-
cardial T1-mapping bias caused by intramyocardial fat in inversion
26. Young MJ, Morgan J, Brolin K, et al. Activation of mineralocorticoid recovery and saturation recovery techniques. J Cardiovasc Magn
receptors by exogenous glucocorticoids and the development of car- Reson 2015;17:33.
diovascular inflammatory responses in adrenalectomized rats. Endocri-
nology 2010;151:2622–2628. 34. Liu C-Y, Redheuil A, Ouwerkerk R, et al. Myocardial fat quantification
in humans: Evaluation by two-point water-fat imaging and localized
27. Funder J. Mineralocorticoids and cardiac fibrosis: the decade in proton spectroscopy. Magn Reson Med 2010;63:892–901.
review. Clin Exp Pharmacol Physiol 2001;28:1002–1006.
35. Szczepaniak LS, Victor RG, Orci L, Unger RH. Forgotten but not gone:
28. Kellman P, Hansen MS. T1-mapping in the heart: accuracy and preci- the rediscovery of fatty heart, the most common unrecognized dis-
sion. J Cardiovasc Magn Reson 2014;16:2. ease in America. Circ Res 2007;101:759–767.

ORIGINAL RESEARCH
Betel Quid Chewing Alters Functional

Connectivity in Frontal and Default
Networks: A Resting-State fMRI Study
Xiaojun Huang, MD,1 Zhening Liu, MD, PhD,1,2 Tumbwene E. Mwansisya, PhD,3
Weidan Pu, PhD,4 Li Zhou, MD,1 Chang Liu, MD,1 Xudong Chen, MD,1
Robert Rohrbaugh, MD,5 Carla Marienfeld, MD,5 Zhimin Xue, MD, PhD,1 and
Haihong Liu, MD, PhD6*
Purpose: To explore the acute effect of betel quid (BQ) use on functional network connectivity by comparing the global
functional brain networks and their subsets before and immediately after BQ chewing.
Materials and Methods: Resting-state functional magnetic resonance imaging (fMRI) was performed in 27 healthy male
participants before and just after chewing BQ on a 3.0T scanner with a gradient-echo echo planar imaging sequence.
Independent component analysis (ICA) was used to determine components that represent the brain’s functional net-
works and their spatial aspects of functional connectivity. A paired t-test was used for exploring the connectivity differ-
ences in each network before and after BQ chewing.
Results: Sixteen networks were identified by ICA. Nine of them showed connectivity differences before and after BQ
chewing (P < 0.05 false discovery rate corrected): (A) orbitofrontal, (B) left frontoparietal, (C) visual, (D) right frontopari-
etal, (E) anterior default mode, (F) medial frontal/anterior cingulate (G) frontotemporal, (H) occipital/parietal, (I) occipi-
tal/temporal/cerebellum. Moreover, networks A, B, C, D, G, H, and I showed increased connectivity, while networks E
and F showed decreased connectivity in participants after BQ chewing compared to before chewing.
Conclusion: The acute effects of BQ use appear to actively alter functional connectivity of frontal and default networks
that are known to play a key role in addictive behavior.
1990s.4 BQ has been associated with immediate psychologi-

B etel quid (BQ) is a mixture of ingredients that typically
includes areca nut, piper betel leaf, slaked lime (calcium
hydroxide), tobacco, and a Chinese herb (Acacia catechu) for
cal effects such as drunkenness, restfulness, euphoria, and a
sense of well-being, as well as physiological effects such as
flavor, but the ingredients vary considerably by region, tachycardia, salivation, miosis, and tremors.3,5 However,
country, ethnicity, and personal preference.1,2 It is most long-term BQ use may result in malignancies such as oral
commonly used in South and Southeast Asia and the Pacific squamous cell carcinoma,6 pregnancy complications,7 and
Islands and is ranked the fourth most frequently used psy- systemic diseases such as type 2 diabetes.8 BQ is a source of
choactive substance worldwide, just after alcohol, nicotine, significant and avoidable morbidity and mortality, yet the
and caffeine.3 Although there is little accessible data on the brain mechanism underlying BQ’s acute effect and its
nationwide frequency of BQ use in China, Zhang and dependence syndrome remain poorly understood.
Reichart reported that the prevalence of use was 64.5% to Like other abused substances, BQ use has been associ-
82.7% in Hunan Province from the late 1980s to early ated with a dependency syndrome that includes positive
*Address reprint requests to: H.L., Clinical Psychology Center of Xiangya Hospital, Central South University. 87 Xiangya Road, Changsha, Hunan, 410008,
PR China. E-mail: liuhaihong9@gmail.com.
From the 1Mental Health Institute of the Second Xiangya Hospital, Central South University, Changsha, Hunan, PR China; 2State Key Laboratory of Medical
Genetics, Central South University, Changsha, PR China; 3College of Health Sciences, University of Dodoma, Dodoma, Tanzania; 4Medical Psychological
Institute, Second Xiangya Hospital, Central South University, Changsha, PR China; 5Department of Psychiatry, Yale University School of Medicine,
New Haven, Connecticut, USA; and 6Clinical Psychology Center of Xiangya Hospital, Central South University, Changsha, Hunan, PR China

V 157

effects such as increased concentration, mild euphoria and pants met the following inclusion criteria: 1) 18–30 years of age;
relaxation, postprandial satisfaction, a sense of well-being, 2) Han Chinese ethnicity; 3) completed 9 or more years of educa-
heightened alertness, and increased capacity to work.9 Also, tion; 4) right-handed; 5) had not chewed BQ before, without fre-
BQ use has been associated with negative effects such as habit- quent consumption of caffeine and alcohol, or abuse of other
uation, addiction, and withdrawal. Its withdrawal syndrome psychoactive substances such as cigarettes, cocaine, heroin, amphet-
amines, etc. Participants were excluded if they had: i) a history of
includes insomnia, mood swings, irritability, and anxiety, the
neurological disorders or other serious physical illnesses; ii) a his-
severity of which can be compared to that of amphetamine
tory of any DSM IV Axis-I mental disorders; iii) used any psycho-
use.10 Dependency and withdrawal syndromes of BQ use may
active substance in 24 hours before imaging scan; or iv) a
be related to its physiological effects on the central and auto- contraindication to MRI. Moreover, the Beck Depression Inventory
nomic nervous system.9 Compared with other substances, and Beck Anxiety Inventory were administered before the first
however, few studies have explored the effect of BQ use on resting-state scan to assess the participants’ emotional status.
the neuroanatomy and function of the human brain.
Only one study using magnetic resonance imaging Images Acquisition and Preprocessing
(MRI) has explored the brain morphological changes in sub- All participants first underwent a resting-state functional (f )MRI
jects with BQ dependency.11 Chen et al11 found that indi- scan. They were then brought out of the scanner and given BQ in
viduals with BQ dependence exhibited significant gray the form of a half fruit in a dried state, including the husk. The
BQ was a final industrially packaged product without tobacco, pre-
matter volume loss in the midbrain, right rostral anterior
viously described by Zhang and Reichart.4 The participants were
cingulate cortex, bilateral dorsolateral prefrontal cortex, and
instructed to chew the nut and swallow the saliva as fast as possible
right superior temporal gyrus, as well as increased gray mat-
in 3 minutes. Then the residual BQ was disgorged, the participants
ter volume in the hippocampal and precuneus. That study were repositioned, and the second fMRI scan was started another 3
also found a negative relationship between the duration of minutes later.
BQ dependency and gray matter volumes in the dorsolateral Images were acquired on a Philips Gyroscan Achieva 3.0T
prefrontal cortex and rostral anterior cingulate cortex, as scanner (Best, Netherlands) in the axial direction. Gradient-echo
well as between the severity of BQ dependency and the echo planar imaging sequence was used with the following parame-
reduced gray matter volume in the midbrain.11 Interestingly, ters: repetition time 5 2000 msec, echo time 5 30 msec, flip
Chu12 found that BQ chewing increased both alpha and angle 5 908, matrix 5 64 3 64, slice thickness 5 4 mm, gap 5
beta activities but decreased theta activity in 52 BQ users, 0 mm, field of view 5 24 3 24 cm, number of slices 5 36. Foam
and the altered rhythms were restricted to the occipital areas pads were used to minimize head motion and earplugs to minimize
for alpha, but were widespread for both beta and theta scanner noise. Participants were instructed to lie supine and to
activity. These findings suggest that BQ chewing may cause remain at rest and motionless with their eyes closed. Each resting-
state fMRI scan lasted 500 seconds and a total of 250 timepoints
a state of arousal and a lesser degree of relaxation, symptoms
were obtained. The images of the initial 10 timepoints were dis-
that may be associated with the acute effects of BQ use on
carded for scanner calibration and for participants to get used to
the distributed functional brain networks. However, how
the circumstances. SPM5 software (http://www.fil.ion.ucl.ac.uk/
BQ impacts human brain networks remains unknown. spm) was used for further image preprocessing. The remaining
In the present study we tried to explore the global images were corrected for the acquisition delay between slices and
functional brain networks and find out their subsets that for head motion. Realignment allowing a maximum translation
may underlie the acute effects of BQ use. Since BQ, like and/or rotation met the following two criteria: 1) maximum dis-
other psychoactive drugs, is associated with habitual and placement in x, y, or z was less than 2 mm and 2) angular rotation
addictive action, we hypothesized that BQ would have acute about each axis was less than 28. Data were then spatially normal-
effects in the brain networks of learning and memory.13 The ized into Montreal Neurological Institute standardized space
frontal and default networks may be affected in particular resampled to 3 3 3 3 3 mm voxels and spatially smoothed with a
because of their key roles in executive control, reward, and Gaussian kernel (full width at half-maximum, 8 mm).
social cognition.14,15 The findings from this study contrib-
Independent Component Analysis
ute to a better understanding of the functional brain mecha-
The GIFT group ICA toolbox (http://icatb.sourceforge.net/, v.
nism involved in the habituation, addiction, and withdrawal 3.0a) was used to determine temporally distinct components in 54
symptoms of BQ chewing, which in turn may be helpful fMRI scans of 27 participants before and after BQ chewing. Modi-
for developing available interventions. fied minimum description length (MDL) algorithm16,17 that
accounts for spatial correlation17 was used to calculate the dimen-
Materials and Methods sion estimation to determine the number of components. The
Participants average number of independent components was 24, estimated
This study was approved by the Ethics Committee of the Second from data across all subjects. The stability of the derived networks
Xiangya Hospital of Central South University. Written informed was investigated using ICASSO software based on a random initia-
consent was obtained from all 27 healthy male participants. Partici- tion method.18 Then principal components analysis was used to

Huang et al.: Betel Quid Alters Network Connectivity
C
O
L
O
R
FIGURE 1: Independent component analysis non-noise resting-state networks from left to right (top row: posterior default mode,
orbitofrontal, left frontoparietal; 2nd row: visual, thalamic/basal ganglia, right frontoparietal; 3rd row: frontotemporal, anterior
default mode, medial frontal/anterior cingulate; 4th row: parietal, frontotemporal, occipital/parietal; 5th row: frontal, right tempo-
ral, occipital/temporal/cerebellum; bottom row: temporal-limbic).
reduce the data, followed by independent component estimation Statistical Analysis

with the infomax algorithm.19 A method based on principal com- A paired t-test was used to explore the functional connectivity dif-
ponents analysis compression and projection was used to back- ference in each component before and after BQ chewing by using
reconstruct the independent components’ spatial maps and time SPM5 (http://www.fil.ion.ucl.ac.uk/spm5). The significance level
courses for each subject20 and image distribution was centered to a was set at P < 0.05 (false discovery rate corrected). The final statis-
mode of zero.21 tical parametric map represented the functional connectivity differ-
ence in each network before and after BQ chewing.
Identifying Networks
To identify valid resting-state networks, standard methods of Results
rejecting artifactual independent components were employed. Demographic and Clinical Characteristics
Components were examined visually to eliminate obvious arti- The participants were 24.5 6 1.45 years old. The mean
facts, then spatially correlated to a priori probabilistic gray mat- weight of given BQ was 2.73 6 0.48 g. The mean interval
ter, white matter, and cerebrospinal fluid templates (in SPM5) between the beginning time of chewing BQ and the subse-
using multiple regressions. Components having low association
quent beginning time of image acquisition was 351.19 6
(jbetaj < 0.3) with gray matter template and high association
34.30 seconds. The scores on the Beck Depression Inven-
(jbetaj > 2) with white matter and cerebrospinal fluid template
tory (mean score 3.89 6 4.63, range 0 to 14) and Beck
were regarded as artifacts. Thus, eight components were dis-
carded as noise and 16 components were deemed valid networks Anxiety Inventory (mean score 23.19 6 2.66, range 21 to
for further analysis. A voxel-wise one-sample t-test was used for 32) showed that none of the participants had depression or
each remaining component at a group level. The statistical para- anxiety.
metric maps were obtained to examine the validity of compo-
F1 nents, illustrated in Fig. 1. The likely function of each network Network Connectivity Changes After BQ Chewing
was determined by referring to Laird et al22 and Khadka et al23 Sixteen resting-state functional networks were validated
T1 as shown in Table 1. (Table 1, Fig. 1). Functional connectivity changes after BQ
January 2017 159

160
TABLE 1. Identified Networks and Their Supposed Function and Connectivity Differences Before and After Betel Quid Chewing
Componet Network Network function

number Paired t-test
Before vs. Cluster size Region Brodmann MNI
after betel location (AAL) area
quid chewing
XYZ
1 Noise — — — — — —
2 Noise — — — — — —
3 Posterior default Theory of mind/social bBQ > aBQ Negative — — —
mode cognition
bBQ < aBQ Negative — — —
4 Orbitofrontal Olfaction, gustation, bBQ > aBQ Negative — — —
emotion
bBQ < aBQ 26 Cingulum_Ant_R 32 3 42 12
21 Frontal_Sup_Orb_R 11 27 42 -15
5 Left fronto- Language encoding and bBQ > aBQ Negative — — —
parietal recognition, visuospatial
processing and reasoning
bBQ < aBQ 332 Frontal_Mid_L, 9, 46 -39 9 36
Precentral_L,
Frontal_Inf_Oper_L
35 Occipital_Mid_L 19 -30 -78 33
6 Visual Visuospatial processing bBQ > aBQ Negative — — —
and reasoning
bBQ < aBQ 168 Lingual_L, Lingual_R, 30, 18, 17 -24 -81 3
Calcarine_L, Cal-
carine_R,
Cerebellum_4_5_R
7 Thalamic/basal Mental processes, most bBQ > aBQ Negative — — —
ganglia strongly to reward tasks,
nonpainful thermal stimu-
Stage:
lation, and interoceptive

functions

Volume 45, No. 1
Page: 160
TABLE 1: Continued
January 2017
quid chewing
XYZ
8 Right fronto- Attention, inhibition and bBQ > aBQ Negative — — —
parietal memory
bBQ < aBQ 53 Frontal_Mid_R 8 42 21 39
20 Frontal_Mid_R 10 36 60 6
9 Fronto-temporal Auditory processing, bBQ > aBQ Negative — — —
sensorimotor
10 Noise — — — — — —
11 Anterior default Theory of mind/social bBQ > aBQ 258 Cingulum_Mid_R, 32, 24, 9 -15 30 18
mode cognition Cingulum_Ant_L,
Frontal_Sup_Medial_L,
Cingulum_Mid_L,
Cingulum_Ant_R
12 Medial frontal/ Theory of mind and social bBQ > aBQ 41 Frontal_Sup_Medial_L 8 -9 42 45
anterior cingulate cognition
28 Frontal_Sup_Medial_L, 10 -15 48 12
Cingulum_Ant_L
13 Noise — — — — — —
14 Parietal Visuospatial processing bBQ > aBQ Negative — — —

and reasoning, multiple
cognitive processes
Stage:

15 Noise — — — — — —

161
Page: 161
162
TABLE 1: Continued

quid chewing
XYZ
16 Fronto-temporal Auditory processing, emo- bBQ > aBQ Negative — — —
tion/interoception
processing
bBQ < aBQ 26 Temporal_Pole_Sup_L 22 -51 12 -6
17 Occipital/parietal Visuospatial processing bBQ > aBQ Negative — — —

and reasoning
bBQ < aBQ 308 Precuneus_L, Precu- 7 -12 -72 51
neus_R, Parietal_-
Sup_L, Parietal_Sup_R,
Cuneus_R
24 Frontal_Sup_R 6 27 12 57
18 Frontal Cognitive control of visuo- bBQ > aBQ Negative — — —
motor timing and prepara-
tion of executed
movement.
19 Right temporal Viewing complex, often bBQ > aBQ Negative — — —
emotional, stimuli as well
as action observation, overt
picture naming, and visual
tracking of moving objects
20 Noise — — — — — —
21 Noise — — — — — —
22 occipital/tempo- Sensorimotor, autonomic, bBQ > aBQ Negative — — —
Stage:
ral/cerebellum and cognitive functions,

simple visual stimuli,
higher-level visual

processing
Volume 45, No. 1

Page: 162
January 2017
TABLE 1: Continued

quid chewing
XYZ
bBQ < aBQ 217 Cerebellum_Crus2_L, — -9 -87 -21
Cerebellum_Crus1_L,
Cerebellum_6_L,
Cerebellum_Crus2_R
47 Temporal_Inf_R, Cere- 19, 37 54 -66 -18
bellum_Crus1_R,
Fusiform_R
41 Cerebellum_6_R — 15 -69 -18
23 Temporal-limbic Sensorimotor, autonomic bBQ > aBQ Negative — — —
and cognitive functions,
linking cognition and
emotion/interoception
24 Noise — — — — — —
Paired t-test, P < 0.05 false discovery rate corrected. AAL, anatomical automatic labeling; MNI, Montreal Neurological Institute; bBQ, before betel quid chewing; aBQ, after betel quid
chewing; negative, no statistical significance; L/R, left/right; Ant, anterior; Sup, superior; Orb, orbital; Mid, middle; Inf, inferior; Oper, operculum.
Stage:

163
Page: 163
C
O
L
O
R
FIGURE 2: Differences in network connectivity in participants before and after BQ chewing, determined through paired t-tests (P
< 0.05 false discovery rate corrected). Increased connectivity networks after betel quid chewing from left to right (top row: orbito-
frontal, left frontoparietal; 2nd row: visual, right frontoparietal; 4th row: frontotemporal, occipital/parietal; bottom row: occipital/
temporal/cerebellum) and decreased networks from left to right (3rd row: anterior default mode, medial frontal/anterior cingu-
late). The color bars represent the range of t values.
chewing were found in nine networks, including increased been found to be associated with the acute or long-term
connectivity in the orbitofrontal, left and right frontoparie- effects of substance use.26 For example, acute effects of caf-
tal, frontotemporal, visual, occipital/parietal, and occipital/ feine, a well-known cognitive enhancer, increase activation
temporal/cerebellum networks, and decreased connectivity of parietal cortex, cerebellum, and other subcortical areas,
in the anterior default and medial frontal/anterior cingulate and decrease deactivation of posterior medial and left lateral
networks (paired t-test, P < 0.05 false discovery rate cor- cortex.27 Nicotine-dependent smokers show increased
F2 rected, Table 1, Fig. 2). resting-state functional connectivity in prefrontal and fron-
toparietal networks, which could facilitate drug-cue respond-
Discussion ing.28 These findings in combination provide a clearer
The present study aimed to identify which brain networks explanation of the brain regions, networks, and connectivity
were affected by BQ chewing and how functional connectiv- alterations that may be associated with the positive and neg-
ity was changed in healthy individuals. The final identified ative effects of BQ chewing.
resting-state networks were consistent with previous multi- Independent component analysis showed that BQ
variate fMRI data decompositions that used independent chewing increased the connectivity in most of the identified
component analysis.24,25 After BQ chewing, connectivity networks, especially the frontal networks. The orbitofrontal
changes were found in nine of these networks, including network is supposed to promote the ability to regulate com-
increased connectivity in the orbitofrontal, bilateral fronto- plex and flexible emotions and behavior according to likely
parietal, frontotemporal, visual, occipital/parietal, and occi- consequences, playing an important role in the processing of
pital/temporal/cerebellum networks and decreased rewards and punishments that is involved in the evolution
connectivity in the anterior default and medial frontal/ante- of drug addiction.14 Brain imaging studies showing that the
rior cingulate networks. These networks have previously orbitofrontal cortex is activated by cues or acute use of drug

in addicts29 support our findings of enhanced orbitofrontal global brain network rather than being limited to a priori
connectivity after BQ chewing. The orbitofrontal and fron- defined networks such as the default network.
toparietal areas were activated when deciding between Some limitations are worth mentioning. First, only
immediate or delayed rewards.30 Sober alcoholics and acute effects of BQ chewing on human brain were meas-
nonsubstance-abusing control subjects showed a tendency to ured; whether these effects are prolonged requires further
wait for a larger, delayed reward. This inclination correlates study of subjects with BQ dependence. Second, the partici-
directly with the magnitude of the bold oxygen level- pants of this study were young, healthy adults who were
dependent (BOLD) signal in the lateral orbitofrontal cortex, na€ıve to BQ. Since the effects of BQ chewing are known to
while immediate reward bias is directly associated with the be habit-related and dose-dependent, the na€ıve chewers may
magnitude of the BOLD signal in the frontoparietal areas.31 experience stronger acute effects of BQ chewing than habit-
Moreover, the frontoparietal network is involved in the ual chewers.9,36 On the other hand, a state of BQ craving
development of reasoning ability32 and cognitive control,33 in habitual chewers may complicate the human brain hemo-
language comprehension (along with the frontotemporal dynamic response. Third, since the BQ administered to par-
areas),34 and visuospatial processing (along with the visual ticipants in this study were final industrial products of half
and occipitoparietal areas).35 However, the aforementioned areca nut, varied BQ doses for different participants may
cognitive abilities and behaviors are claimed to be impaired cause variation in individuals’ network connectivity altera-
in most BQ chewers.12,36 In line with these findings, BQ tions in response to BQ chewing. Finally, like other studies
users experience euphoria, a sense of well-being, increased using resting-state fMRI, the effects of physiologic noises
alertness, and enhanced capacity to work,9 supporting the such as heart rhythm and respiration can be aliased into the
possibility that the observed increase in networks’ connectiv- low-frequency fluctuations. These effects can be reduced by
ity after BQ chewing may result in immediate enhancement
bandpass filtering of 0.01–0.08 Hz, but cannot be com-
of cognition.
pletely eliminated. In this study, subsequent independent
In this study BQ chewing decreased the connectivity
component analysis may have helped further estimate, segre-
in default networks, including in the anterior default mode
gate, or remove these physiological effects.
and medial frontal/anterior cingulate networks. Default net-
In conclusion, the acute effects of BQ chewing
works, which are active when the brain is “at rest,” have
increase functional connectivity in the orbitofrontal, fronto-
been associated with mind-wandering37 and social cogni-
parietal, frontotemporal, visual, occipital/parietal, and occi-
tion, including emotion perception, theory of mind, and
pital/temporal/cerebellum networks, and decrease
morality.15 Activity of the default networks is deactivated
connectivity in the anterior default and medial frontal/ante-
when individuals focus on the external environment or on
rior cingulate networks. Altered connectivity in these net-
performing goal-directed tasks, possibly by reducing goal-
works may be involved in BQ’s psychological and
irrelevant functions.38 Imaging studies on psychiatric ill-
physiological effects and play an important role in addictive
nesses have demonstrated that major depression is associated
with greater resting-state connectivity in the medial frontal/ behavior. Further studies are required to refine the specific
anterior cingulate cortex,39,40 and that antidepressants nor- role of each identified network and to explore the long-term
malize default network connectivity.41 Therefore, the present effects of BQ chewing.
study shows that BQ chewing, which immediately sup-
presses the connectivity of default networks, may result in Acknowledgments
reduced mind-wandering, enhanced focused attention, Contract grant sponsor: National Natural Science Founda-
diminished depression, and improved social cognition. tion of China; contract grant number: 81561168021;
Interestingly, the pharmacological effects of BQ have 81471362; 81401125; 81271485; 81000587
been found to have an onset within 2 minutes after chewing The authors thank Zhong He from the Department of
and peak within 4–6 minutes,9 a time span that coincides Radiology of Second Xiangya Hospital, Central South Uni-
with the time during which we measured the brain activity versity, for assistance in imaging data acquisition.
in our study. Our findings thus likely indicate the actual
brain networks that are influenced by the BQ pharmacologi-
cal effects. Thus, the present study provides preliminary evi-
dence of the association of BQ pharmacological effects with References
the set of large-scale functionally connected brain networks 1. Winstock A. Areca nut-abuse liability, dependence and public health.
that can be captured during resting-state fMRI. Addict Biol 2002;7:133–138.
This resting-state fMRI study demonstrates the acute 2. IARC Working Group on the Evaluation of Carcinogenic Risks to
Humans. Betel-quid and areca-nut chewing and some areca-nut
effects of BQ chewing on the human brain’s functional net- derived nitrosamines. IARC Monogr Eval Carcinog Risks Hum 2004;
works. It has an advantage of exploring all subsets of the 85:1–334.
January 2017 165

3. Boucher BJ, Mannan N. Metabolic effects of the consumption of 23. Khadka S, Meda SA, Stevens MC, et al. Is aberrant functional connec-
Areca catechu. Addict Biol 2002;7:103–110. tivity a psychosis endophenotype? A resting state functional magnetic
resonance imaging study. Biol Psychiatry 2013;74:458–466.
4. Zhang X, Reichart PA. A review of betel quid chewing, oral cancer
and precancer in Mainland China. Oral Oncol 2007;43:424–430. 24. Allen EA, Erhardt EB, Damaraju E, et al. A baseline for the multivariate
comparison of resting-state networks. Front Syst Neurosci 2011;5:2.
5. Chu NS. Effects of Betel chewing on the central and autonomic nerv-
ous systems. J Biomed Sci 2001;8:229–236. 25. Damoiseaux JS, Rombouts SA, Barkhof F, et al. Consistent resting-
state networks across healthy subjects. Proc Natl Acad Sci U S A
6. Trivedy CR, Craig G, Warnakulasuriya S. The oral health consequences
2006;103:13848–13853.
of chewing areca nut. Addict Biol 2002;7:115–125.
26. Tanabe J, Thompson L, Claus E, Dalwani M, Hutchison K, Banich MT.
7. Yang MS, Chang FT, Chen SS, Lee CH, Ko YC. Betel quid chewing
Prefrontal cortex activity is reduced in gambling and nongambling
and risk of adverse pregnancy outcomes among aborigines in south- substance users during decision-making. Hum Brain Mapp 2007;28:
ern Taiwan. Public Health 1999;113:189–192. 1276–1286.
8. Mannan N, Boucher BJ, Evans SJ. Increased waist size and weight in 27. Park CA, Kang CK, Son YD, et al. The effects of caffeine ingestion on
relation to consumption of Areca catechu (betel-nut); a risk factor for cortical areas: functional imaging study. Magn Reson Imaging 2014;
increased glycaemia in Asians in east London. Br J Nutr 2000;83:267– 32:366–371.
275.
28. Janes AC, Nickerson LD, Frederick Bde B, Kaufman MJ. Prefrontal
9. Chu NS. Neurological aspects of areca and betel chewing. Addict and limbic resting state brain network functional connectivity differs
Biol 2002;7:111–114. between nicotine-dependent smokers and non-smoking controls.
10. Giri S, Idle JR, Chen C, Zabriskie TM, Krausz KW, Gonzalez FJ. A Drug Alcohol Depend 2012;125:252–259.
metabolomic approach to the metabolism of the areca nut alkaloids 29. Schoenbaum G, Shaham Y. The role of orbitofrontal cortex in drug addic-
arecoline and arecaidine in the mouse. Chem Res Toxicol 2006;19: tion: a review of preclinical studies. Biol Psychiatry 2008;63:256–262.
818–827.
30. McClure SM, Ericson KM, Laibson DI, Loewenstein G, Cohen JD.
11. Chen F, Zhong Y, Zhang Z, et al. Gray matter abnormalities associ- Time discounting for primary rewards. J Neurosci 2007;27:5796–5804.
ated with betel quid dependence: a voxel-based morphometry study.
Am J Transl Res 2015;7:364–374. 31. Boettiger CA, Mitchell JM, Tavares VC, et al. Immediate reward bias in
humans: fronto-parietal networks and a role for the catechol-O-
12. Chu NS. Effects of betel chewing on electroencephalographic activity: methyltransferase 158(Val/Val) genotype. J Neurosci 2007;27:14383–14391.
spectral analysis and topographic mapping. J Formos Med Assoc
1994;93:167–169. 32. Wendelken C, Ferrer E, Whitaker KJ, Bunge SA. Fronto-parietal net-
work reconfiguration supports the development of reasoning ability.
13. Hyman SE. Addiction: a disease of learning and memory. Am J Psy- Cereb Cortex 201;26:2178–2190.
chiatry 2005;162:1414–1422.
33. Gulbinaite R, van Rijn H, Cohen MX. Fronto-parietal network oscilla-
14. Kringelbach ML, Rolls ET. The functional neuroanatomy of the human tions reveal relationship between working memory capacity and cog-
orbitofrontal cortex: evidence from neuroimaging and neuropsychol- nitive control. Front Hum Neurosci 2014;8:761.
ogy. Prog Neurobiol 2004;72:341–372.
34. Rodd JM, Longe OA, Randall B, Tyler LK. The functional organisation
15. Li W, Mai X, Liu C. The default mode network and social understand- of the fronto-temporal language system: evidence from syntactic and
ing of others: what do brain connectivity studies tell us. Front Hum semantic ambiguity. Neuropsychologia 2010;48:1324–1335.
Neurosci 2014;8:74.
35. de Graaf TA, Roebroeck A, Goebel R, Sack AT. Brain network dynam-
16. Jafri MJ, Pearlson GD, Stevens M, Calhoun VD. A method for func- ics underlying visuospatial judgment: an FMRI connectivity study.
tional network connectivity among spatially independent resting-state J Cogn Neurosci 2010;22:2012–2026.
components in schizophrenia. Neuroimage 2008;39:1666–1681.
36. Cawte J. Psychoactive substances of the South Seas: betel, kava and
17. Li YO, Adali T, Calhoun VD. Estimating the number of independent pituri. Aust N Z J Psychiatry 1985;19:83–87.
components for functional magnetic resonance imaging data. Hum
Brain Mapp 2007;28:1251–1266. 37. Mason MF, Norton MI, Van Horn JD, Wegner DM, Grafton ST,
Macrae CN. Wandering minds: the default network and stimulus-
18. Himberg J, Hyvarinen A, Esposito F. Validating the independent com- independent thought. Science 2007;315:393–395.
ponents of neuroimaging time series via clustering and visualization.
Neuroimage 2004;22:1214–1222. 38. Anticevic A, Cole MW, Murray JD, Corlett PR, Wang XJ, Krystal JH.
The role of default network deactivation in cognition and disease.
19. Calhoun VD, Adali T, Pearlson GD, Pekar JJ. Spatial and temporal Trends Cogn Sci 2012;16:584–592.
independent component analysis of functional MRI data containing a
pair of task-related waveforms. Hum Brain Mapp 2001;13:43–53. 39. Greicius MD, Flores BH, Menon V, et al. Resting-state functional connec-
tivity in major depression: abnormally increased contributions from sub-
20. Calhoun VD, Adali T, Pearlson GD, Pekar JJ. A method for making genual cingulate cortex and thalamus. Biol Psychiatry 2007;62:429–437.
group inferences from functional MRI data using independent compo-
40. Davey CG, Harrison BJ, Yucel M, Allen NB. Regionally specific altera-
nent analysis. Hum Brain Mapp 2001;14:140–151.
tions in functional connectivity of the anterior cingulate cortex in
21. Stevens MC, Kiehl KA, Pearlson GD, Calhoun VD. Brain network major depressive disorder. Psychol Med 2012;42:2071–2081.
dynamics during error commission. Hum Brain Mapp 2009;30:24–37.
41. Posner J, Hellerstein DJ, Gat I, et al. Antidepressants normalize the
22. Laird AR, Fox PM, Eickhoff SB, et al. Behavioral interpretations of default mode network in patients with dysthymia. JAMA Psychiatry
intrinsic connectivity networks. J Cogn Neurosci 2011;23:4022–4037. 2013;70:373–382.

ORIGINAL RESEARCH
Benign and Malignant Orbital

Lymphoproliferative Disorders:
Differentiating Using Multiparametric
MRI at 3.0T
Xiao-Quan Xu, MD,1 Hao Hu, MD,1 Hu Liu, MD, PhD,2 Jiang-Fen Wu, MD,3
Peng Cao, MD,3 Hai-Bin Shi, MD, PhD,1* and Fei-Yun Wu, MD, PhD1*
Purpose: To determine the optimal combination of parameters derived from 3T multiparametric (conventional magnetic
resonance imaging [MRI], diffusion-weighted [DW] and dynamic contrast-enhanced [DCE]) MRI for differentiating malig-
nant from benign orbital lymphoproliferative disorders (OLPDs).
Materials and Methods: Forty patients with OLPDs (18 benign and 22 malignant) underwent conventional 3.0T MR,
DW, and DCE-MRI examination for presurgery evaluation. Conventional MRI features (including tumor laterality, shape,
number of involved quadrants, signal intensity on T1-weighted imaging (WI) and T2WI, flow void sign on T2WI, and find-
ings suggestive of sinusitis) were reviewed, and multivariate logistic regression analysis was used to identify the most
significant conventional MRI features. Apparent diffusion coefficient (ADC) and DCE-MRI derived parameters (area
under curve [AUC], time to peak [TTP], maximum rise slope [Slopemax]) were measured and compared between two
groups. Receiver operating characteristic (ROC) curve analyses were used to determine the diagnostic ability of each
combination that was established based on identified qualitative and quantitative parameters.
Results: Multivariate logistic regression analysis showed that the presence of flow void sign on T2WI significantly associ-
ated with benign OLPDs (P 5 0.034). Malignant OLPDs demonstrated significantly lower ADC (P 5 0.001) and AUC
(P 5 0.002) than benign mimics. ROC analyses indicted that, ADC alone showed the optimal sensitivity (threshold value,
0.886 3 1023 mm2/s; sensitivity, 90.9%), while a combination of no presence of flow void sign on T2WI 1 ADC 0.886
3 1023 mm2/s 1 AUC 7.366 showed optimal specificity (88.9%) in differentiating benign from malignant OLPDs.
Conclusion: Multiparametric MRI can help to differentiate malignant from benign OLPDs. DWI offers optimal sensitivity,
while the combination of conventional MRI, DWI, and DCE-MRI offers optimal specificity.
O rbital lymphoproliferative disorders (OLPDs) represent

a broad spectrum of benign and malignant lesions,
including lymphoid hyperplasia, atypical lymphoid hyper-
Therefore, preoperative differentiation of benign and malig-
nant OLPDs has a treatment planning benefit. The clinical
differentiation between these two entities is limited.3 There-
plasia, lymphoma, and idiopathic inflammatory pseudotu- fore, there is a need for an effective method to differentiate
mor.1,2 Besides these, orbital IgG4-related disease (IgG4- these two entities.
RD) is becoming increasingly recognized and incorporated Magnetic resonance imaging (MRI) plays an impor-
into benign OLPDs groups based on recent surveillance.3 tant role in evaluating the extent of orbital lesions, and pro-
In terms of treatment, orbital lymphoma is amenable vides supplementing information beyond clinical
1,5,6
to low-dose radiation therapy, whereas benign mimics usu- examination. A prior study indicated that some specific
ally exhibit a positive response to corticosteroid therapy.3,4 imaging features such as tumor shape, presence of flow void
The first two authors contributed equally to this work.
*Address reprint request to: F.-Y.W. or H.-B.S., Department of Radiology, First Affiliated Hospital of Nanjing Medical University, No. 300, Guangzhou Rd.,
Nanjing, China. E-mail: wfy_njmu@163.com or shihb@njmu.edu.cn
From the 1Department of Radiology, First Affiliated Hospital of Nanjing Medical University, Nanjing, China; 2Department of Ophthalmology, First Affiliated
Hospital of Nanjing Medical University, Nanjing, China; and 3GE Healthcare, Shanghai, China

V 167

TABLE 1. Demographic and Histological Characteristics of OLPD Patients
Parameters Benign OLPD Malignant OLPD P value

Group (n 5 18) Group (n 5 22)
Mean age 50.4 6 14.4 62.0 6 13.6 0.022

Gender (F/M) 5 : 13 8 : 14 0.564
Histologic subtypes IIP (8) MALT lymphoma (17)
RLH (6) DLBCL (3)
IgG4-related disease (4) Follicular lymphoma (2)
F, female; M, male; MALT, mucosa-associated lymphoid tissue; DLBCL, diffuse large B-cell lymphoma; IIP, idiopathic inflammatory
pseudotumor; IgG4-RD, IgG4 related disease; RLH, reactive lymphoid hyperplasia. Data in parentheses indicate the number of the
corresponding patients in our study.
sign, and imaging findings suggestive of sinusitis might be Among these patients, 40 patients (27 men and 13 women;
potentially useful for differentiating benign from malignant mean age, 57.4 6 15.6 years; range, 22–79 years) were included in
OLPDs.3 However, the qualitative assessment of MRI fea- our study according to the following criteria: 1) the diameter of
tures is a subjective process with limited interreader repro- the lesions exceeded 1 cm; 2) no biopsy or corticosteroid therapy
ducibility, hence indicating a need for more objective was administered before the MRI scan; 3) both DW and DCE-
MRI were scanned; and 4) there was adequate imaging quality
methods to improve the diagnostic accuracy and confidence.
without obvious susceptibility artifacts.
To date, there have been a few investigations that used
The group pf 40 OLPDs patients was comprised of 18
functional MRI techniques to assess the orbital tumors,
benign (13 men and 5 women; mean age, 50.4 6 14.4 years;
including diffusion-weighted (DW) and dynamic contrast- range, 22–67 years) and 22 malignant (14 men and 8 women;
enhanced (DCE) MRI.7–14 One previous study by Hara- mean age, 62.0 6 13.6 years; range, 40–79 years) patients. Detailed
dome et al reported that the orbital lymphoma showed sig- demographic and histological information for the two groups is
nificantly lower apparent diffusion coefficient (ADC) value shown in Table 1. The final diagnosis was made based on the sur- T1
than benign OLPDs, which might be associated with the gical pathological results in 35 patients, and on the follow-up after
higher cellularity in lymphoma and interstitial edematous steroid treatment in five patients with presumed, inflammatory
changes in benign OLPDs.3 Besides diffusion characteristics, pseudotumors.
angiogenesis is also an important tumor hallmark that
MRI
deserves systematic assessment.15 All images were acquired with a 3.0T MRI system (Verio; Siemens,
DCE-MRI is an emerging imaging technique that Erlangen, Germany) with a 12-channel head coil. Conventional
allows quantification of various vascular biomarkers.16 It has unenhanced imaging protocols contained an unenhanced axial T1-
been reported to be useful for various applications, such as weighted imaging (repetition time [TR] / echo time [TE], 600/10
differentiating lung cancer from benign solitary pulmonary msec), axial T2-weighted imaging (WI) (TR/TE, 4700/79 msec)
lesions, assessing the different T stage of nasopharyngeal car- with fat saturation, and coronal T2WI (TR/TE, 3500/79 msec).
cinoma, noninvasively evaluating the immature mean vessel DWI was performed in the axial plane by using a single-shot
density of brain gliomas, and predicting the treatment spin-echo echo-planar imaging sequence for all patients. The fol-
lowing imaging parameters were used: TR/TE, 4000/85 msec; sec-
response for patients with multiple myeloma.16–20
tion thickness, 4 mm; intersection gap, 0 mm; flip angle (FA),
Therefore, the purpose of this study was to determine
1508; number of averages, 6; field of view (FOV), 200 3 200 mm;
the optimal combination of parameters derived from 3T matrix, 384 3 384; and number of sections, 10. The b values
multiparametric MRI (conventional MR, DW, and DCE) used were 0 and 800 sec/mm2. The total acquisition time of DWI
for differentiating malignant from benign OLPDs. was 4 minutes 14 seconds. ADC maps were automatically
generated.
DCE-MRI was performed in the axial plane using a 2D
MATERIALS AND METHODS
turbo fast low angle shot (FLASH) sequence with an integrated
Patients parallel acquisition technique (iPAT). A standard dose of 0.1
Our Institutional Review Board approved this retrospective study, mmol/kg of gadolinium-diethylene triamine pentaacetic acid (Mag-
and written informed consent was waived. A review of our hospital nevist; Bayer Schering Pharma, Berlin, Germany) was administered
database identified 59 clinically or histologically confirmed OLPDs at a rate of 4 mL/s. The bolus of contrast material was followed by
patients who underwent orbital MR examination for disease evalu- a 20-mL bolus of saline administered at the same injection rate.
ation between March 2013 and January 2016. Contrast was administered after five baseline dynamics (total: 95

Xu et al.: 3T Multiparametric MRI of OLPDs
dynamics). The dynamic acquisition was performed with a tempo- 3.1 cm3) were placed on all imaging sections and encompassed as
ral resolution of 3.3 seconds. The detailed parameters for the much as tumor area, excluding the large necrotic, cystic, and hem-
DCE-MRI were as follows: TR/TE, 474.66/1.43 msec; FA, 128; orrhagic areas and surrounding blood vessels referred to T2WI and
average, 1; FOV, 230 mm; matrix, 128 3 128; section thickness, contrast-enhanced T1WI. To minimize the effect of partial volume
4.5 mm; and number of sections, 7. The total acquisition time was averaging, the edges of lesions were avoided during ROI place-
5 minutes 15 seconds. After the DCE-MRI scan, postcontrast ment. If bilateral lesions were seen, only the relatively larger lesion
axial, coronal, and sagittal T1-weighted images were obtained. was included for analysis.
The quantitative assessment results of the two readers were
Imaging Analysis used to evaluate the interreader reproducibility. Additionally, to
Qualitative image assessments were performed by two radiologists
evaluate intrareader reproducibility, the MRI data were quantita-
(Reader 1: X.X. with 6 years of experience; Reader 2: H.H. with 4
tively assessed again by Reader 1, with a minimum washout period
years of experience) who were blinded to the clinical information,
of at least 1 month. The average of the two measurement results
histological results, and study design. If disagreement existed, con-
of Reader 1 was used for the statistical analysis.
sensus was reached by discussing the images with another radiolog-
ist (F.W. with 15 years of experience).
Our qualitative image assessment focused on the laterality, Statistical Analysis
shape, number of involved quadrants, internal signal architecture, Quantitative results are expressed as mean 6 standard deviation,
presence of flow void sign, and findings suggestive of sinusitis. The and we used the Kolmogorov–Smirnov test for normally distrib-
laterality was noted as either unilateral or bilateral. The shape was uted analysis.
classified as regular or irregular. Similar to a prior study, the orbit Univariate analysis was first performed in an attempt to
was divided into four quadrants.8 The number of involved quad- characterize the ability of each variable for predicting malignant
rants was recorded as 1 or ⭌2 quadrants involved. OLPDs. The frequency distribution of gender and each qualitative
Compared with that of the adjacent extraocular muscle, the MRI feature between the two groups was compared with Fisher’s
signal intensity of the tumor on precontrast T1- and T2WI were exact test. The differences in age between the two groups were
classified as hypointense, isointense, and hyperintense. The pres- compared with an unpaired t-test. Then multivariate logistic regres-
ence of a signal void from a vessel within the lesions on T2WI was sion analysis was used to identify the most valuable variables that
defined as presence of flow void sign. Findings suggestive of sinusi- were predictive of OLPDs.
tis were judged according to the criterion proposed by a previous The differences of mean ADC value and DCE-MRI-derived
study,3 including: 1) significant paranasal mucosal thickness parameters between the two groups were compared using an
(>4 mm); 2) fluid level; and 3) the presence of a retention cyst at unpaired t-test. Based on the identified qualitative MRI variables
each paranasal cavity. If bilateral lesions occurred, only the rela- and functional MRI parameters, we established different diagnostic
tively larger one was assessed. models. The receiver operating characteristic (ROC) curve was per-
During ADC value measurements, all DWI data were con- formed to determine the value of each diagnostic model in differ-
verted into Digital Imaging and Communication in Medicine entiating benign from malignant OLPDs. Sensitivity and specificity
(DICOM) format and postprocessed offline with in-house software were calculated with a threshold criterion determined as the value
(FireVoxel; CAI2R; New York University, NY).21 The mean ADC would maximize the Youden index.
value was calculated using the exponential fitting formula: The interreader reproducibility for the conventional MRI
ADC 5 – ln (Sb/S0) / b, where b represents the diffusion sensitivity features assessment was evaluated using kappa analysis. The inter-
coefficients, Sb and S0 represent the corresponding signal values of and intrareader reproducibility for DW and DCE-MRI parameters
the given region-of-interest (ROI). measurement was evaluated using the intraclass correlation coeffi-
The DCE-MRI data were processed using a dedicated post- cient (ICC) with 95% confidence intervals (CI) and applying a
processing software (Omni Kinetics; GE Healthcare, Milwaukee, two-way ICC with a random rater assumption. The ICC and
WI). The time intensity curve (TIC) was obtained for each mass, kappa value ranges 0–1.00, with values closer to 1.00 representing
and divided into persistent, plateau, and washout patterns accord- better reproducibility. They were interpreted as follows: (<0.40,
ing to a previous study.14 For each TIC, the signal intensity (SIpre, poor; 0.41–0.60, moderate; 0.61–0.80, good; and 0.81, excel-
and SImax) and time (Tpre, and Tpeak) were derived. SIpre was lent). All statistical analyses were performed with statistical software
defined as the precontrast signal intensity, and SImax was the signal (SPSS, v. 19.0, Chicago, IL; MedCalc, v. 9.0, MedCalc Software,
intensity at maximal contrast enhancement. Tpre and Tpeak were Mariakerke, Belgium). The statistical significance threshold was set
the time corresponding to the SIpre and SImax. Then quantitative at a two-sided P value below 0.05, while at a P value below 0.017
parameters including area under curve (AUC), time to peak (0.05/3) for the Bonferroni correction for comparison of the DCE-
(TTP), and maximum rise slope (Slopemax) were derived. AUC MRI parameters.
was defined as the area under the whole enhancement curve. TTP
was defined as the time from Tpre to Tpeak. Slopemax was calculated
using the following formula: Slopemax 5 (SImax2SIpre)/ [SIpre 3
RESULTS
(Tpeak – Tpre)] 3 100. There was a significant difference in the age (P 5 0.022),
During the DW and DCE-MRI assessment, ROIs were also while no difference in the gender distribution (P 5 0.564)
determined by the two radiologists mentioned above. ROIs (5.2 6 of patients between benign and malignant OLPDs groups.
January 2017 169

TABLE 2. Conventional MR Imaging Features of Benign And Malignant OLPDs Patients
Parameters Benign OLPD Malignant OLPD P value K value

Laterality 0.641 0.977

Unilateral 11 15
Bilateral 7 7
Shape 0.225 0.803
Irregular 8 14
Regular 10 8
Number of involved quadrants 0.266 0.719
1 5 3
⭌2 13 19
Signal intensity on T1WI 0.673 0.721
Low 17 20
Iso 1 2
Signal intensity on T2WI 0.781 0.758
Low 1 2
Iso 13 15
High 4 5
Presence of flow void sign on T2WI 0.003 0.833
Yes 16 9
No 2 13
Finding suggestive of sinusitis 0.009 0.862
Yes 14 8
No 4 14
The signal intensity on T1- and T2-weighted imaging was compared with that of extraocular muscle.
T2 Table 2 summarizes the frequency distribution of con- an AUC of 0.707, sensitivity of 63.6%, and specificity of
ventional MRI features of OLPDs patients. There were dif- 77.8%.
ferences in the presence of flow void sign on T2WI Table 4 summarizes the detailed comparisons of mean T4
(P 5 0.003) and findings suggestive of sinusitis (P 5 0.009), ADC value and DCE-MRI-derived parameters between the
while no significant difference in tumor laterality two groups. A significantly lower mean ADC value was
(P 5 0.641), shape (P 5 0.225), number of involved quad- found in malignant OLPDs compared with benign mimics
rants (P 5 0.266), signal intensity on T1WI (P 5 0.673), or (P 5 0.001) (Fig. 1a). Regarding the TIC pattern, benign F1
signal intensity on T2WI (P 5 0.781) was seen between OLPDs demonstrated a persistent pattern (n 5 15) or pla-
benign and malignant OLPDs groups. teau pattern (n 5 3), while malignant OLPDs demonstrated
Then age, presence of flow void sign on T2WI, and a washout pattern (n 5 17) or plateau pattern (n 5 5)
findings suggestive of sinusitis were adapted into multi- (Table 4). As to the DCE-MRI-derived parameters, malig-
variate logistic regression analysis. As a result, no presence nant OLPDs showed lower AUC (P 5 0.002) values than
of flow void sign on T2WI was identified as the most benign mimics, while no significant differences were found
important qualitative imaging feature that was indicative on TTP (P 5 0.026) and Slopemax (P 5 0.327) (Fig. 1b–d).
of malignant OLPDs. Detailed multivariate logistic The ROC curve results indicated that, setting AUC 7.366
T3 regression analysis results are summarized in Table 3. Set- as the cutoff value, the best diagnostic ability could be
ting no presence of flow void sign on T2WI as the diag- achieved (AUC, 0.795; sensitivity, 77.3%; specificity,
nostic criterion for malignant OLPDs, we could achieve 72.2%). When ADC 0.886 3 1023 mm2/s was set as the

TABLE 3. Multivariate Logistic Regression Results Using Conventional MRI Features to Differentiate Benign From
Malignant OLPDs
Parameters b Coefficient SE Odds ratio (95%CI) P value
Age 0.058 0.033 1.059 (0.994-1.129) 0.077

No presence of flow 21.992 0.942 0.136 (0.022-0.865) 0.034
void sign on T2WI
No finding suggestive of sinusitis 21.250 0.837 0.287 (0.056-1.479) 0.136
SE, standard error; CI, confidence interval.
threshold value, the best diagnostic ability could be achieved MRI features, while excellent inter- and intrareader agree-
(AUC, 0.779; sensitivity, 90.9%; specificity, 61.1%). ments were achieved during the measurement of ADC and
Based on no presence of flow void sign on T2WI, DCE-MRI derived quantitative parameters. Detailed kappa
AUC 7.366 and ADC 0.886 3 1023 mm2/s, we estab- values for qualitatively assessing the conventional MRI fea-
lished seven potential diagnostic combinations. ROC curves tures are listed in Table 2, and detailed ICCs for the mea-
analyses results indicated that a combination of no presence surement of ADC and DCE MR derived imaging
of flow void sign on T2WI 1 ADC 0.886 3 1023 mm2/ parameters are listed in Table 4.
s 1 AUC 7.366 showed the optimal specificity (88.9%)
F2 (Fig. 2a), while the ADC value alone showed the optimal
sensitivity (90.9%) in predicting orbital malignant OLPDs DISCUSSION
(Fig. 2b). Detailed diagnostic performances of each noted Our study found that flow void sign on T2WI was the most
T5 parameter and diagnostic combination are summarized in significant qualitative conventional MRI feature that may be
F3 Table 5. Figures 3 and 4 show the multiparametric MR helpful in differentiating malignant OLPDs from benign
F4 images of the representative patients with orbital mucosa- mimics. With the addition of DW and DCE-MRI, optimal
associated lymphoid tissue lymphoma and idiopathic specificity could be achieved in the differentiation of benign
inflammatory pseudotumor. and malignant OLPDs, while optimal sensitivity could be
Good or excellent interreader agreements were obtained in differentiating these two entities when DWI was
achieved during the qualitative assessment of conventional used alone.
TABLE 4. Differences of DW and DCE-MRI-Derived Parameters Between Benign and Malignant OLPD Groups
Parameters Benign OLPD Malignant OLPD P value ICC

Inter Intra
DWI
ADCmean 0.927 6 0.225 0.711 6 0.152 0.001 0.886 0.902
DCE-MRI
TIC pattern – – –
Persistent 15 –
Plateau 3 5
Washout – 17
TTP 2.958 6 1.047 2.340 6 0.617 0.026 0.851 0.883
AUC 8.520 6 1.820 6.662 6 1.774 0.002 0.849 0.872
Slopemax 8.321 6 3.279 7.307 6 3.159 0.327 0.855 0.891
-3 2
Except for the P values, data are reported as mean 6 standard deviation. Unit for ADC value is 3 10 mm /s. ADC, apparent diffu-
sion coefficient; ADCmean, mean ADC value; TIC, time intensity curve; TTP, time to peak; AUC, area under curve; Slopemax, maxi-
mum rise slope.
January 2017 171

FIGURE 1: Box-and-whisker plots show ADC (a), AUC (b), Slopemax (c), and TTP (d) calculated for benign and malignant OLPDs
group. The line in the box is the median, the height of the box represents interquartile range, whiskers are the lowest and highest
data points still within the 1.5 interquartile range, and circles indicate outliers.
The flow void sign on T2WI was the most significant This feature was found more frequently in benign OLPDs
qualitative conventional MRI feature that would be helpful than in malignant mimics, which might be due to the
in differentiating malignant OLPDs from benign mimics. hypervascular nature of benign OLPDs.3 Besides that, the
C
O
L
O
R
FIGURE 2: ROC curves of using T21ADC1AUC (a) and ADC (b) to differentiate orbital benign from malignant OLPDs. (a) Combina-
tion of “no presence of flow void sign on T2WI” 1 ADC £ 0.886 3 1023 mm2/s 1 AUC £ 7.366 shows optimal specificity (88.9%),
(b) while the ADC value alone shows optimal sensitivity (90.9%).

TABLE 5. Diagnostic Performance of Each Imaging Variable and Their Combinations
Model Threshold value AUC Sensitivity Specificity
T2WI – 0.707 (0.542-0.840) 63.6% (40.7%-82.8%) 77.8% (52.4%-93.6%)

a
ADC 0.886 0.779 (0.620-0.895) 90.9% (70.8%-98.9%) 61.1% (35.7%-82.7%)
AUC 7.366 0.795 (0.638-0.906) 77.3% (54.6%-92.2%) 72.2% (46.5%-90.3%)
T2WI1ADC – 0.735 (0.572-0.862) 63.6% (40.7%-82.8%) 83.3% (58.6%-96.4%)
T2WI1AUC – 0.712 (0.547-0.844) 59.1% (36.4%-79.3%) 83.3% (58.6%-96.4%)
ADC1AUC – 0.758 (0.596-0.879) 68.2% (45.1%-86.1%) 83.3% (58.6%-96.4%)
T2WI1ADC1AUC – 0.740 (0.577-0.865) 59.1% (36.4%-79.3%) 88.9%b (65.3%-98.6%)
ADC, apparent diffusion coefficient; AUC, area under curve; unit for ADC value is 10-3 mm2/s. Data in parentheses is the 95% con-
fidence interval.
a
ADC value demonstrated highest sensitivity.
b
Combination of T2, ADC, and AUC values demonstrated highest specificity.
patients with malignant OLPDs showed significantly older extension of inflammatory changes to the mucosa of nasal
age than the patients with benign OLPDs in univariate and paranasal cavities could occur.3 Therefore, benign
analysis, which was similar to a prior study.3 This result OLPDs in our study showed more frequent imaging find-
indicated that the patients’ age should be considered in ings indicative of sinusitis than malignant mimics, which is
future clinical differentiation work. In addition, for the similar to a prior study.3 The imaging finding indicative of
benign OLPDs, particularly the IgG4-related diseases, an sinusitis also can provide potential supplemental
C
O
L
O
R
FIGURE 3: Mucosa-associated lymphoid tissue lymphoma in a 54-year-old man. (a) Axial and (b) coronal T2WI show a mass located
in the medial-inferior temporal quadrant of the left orbit. (c) The mass appears hypointense on the ADC map, the ADC value is
0.741 3 1023 mm2/s. (d) ROI drawn around the mass. (e) Obtained TIC shows as a washout pattern. (f) Color map for AUC is
obtained, and the mean AUC value of this patient is 8.073.
January 2017 173

C
O
L
O
R
FIGURE 4: Idiopathic inflammatory pseudotumor in a 37-year-old man. (a) Axial T2WI shows a mass located in the medial-inferior
temporal quadrant of the left orbit. (b) Note the flow void signal (arrow) and the imaging findings suggestive of sinusitis on the
coronal T2WI. (c) ADC value of this mass is 1.003 3 1023 mm2/s. (d) ROI drawn around the mass. (e) Obtained TIC shows as a pla-
teau pattern. (f) Color map for AUC is obtained, and the mean AUC value of this patient is 8.579.
information for differentiation work in clinical practice. mal threshold value in the present study.3 The reasons for this
Our study found no difference in tumor shape between the discrepancy might be due to the variance in the pathological
two groups, which differed from a prior study.3 In our characteristics of the included lesions. Also, the magnetic field
opinion, this contradiction might be due to the low repro- strength (1.5T or 3.0T) and b value (b 5 800 or 1000 s/mm2)
ducibility of qualitative MRI assessment, and this problem would affect the ADC measurements.3
seems to be an important clinical obstacle. Unfortunately, the diagnostic specificity of DWI in
In agreement with previous studies, we found that the our study was limited. This might be associated with the
mean ADC value of malignant OLPDs was significantly lower inclusion of IgG4-related orbital diseases in our study.
than that of the benign mimics.3,6–12,22 This may be attributed IgG4-RD has been noted as a special benign OLPDs. It is
to the enlarged nuclei and hypercellularity of the malignant also known as a dense lymphoplasmacytic infiltrate that is
OLPDs.3,8,10 These typical histological characteristics may act to similar to lymphoma.22 Restricted diffusion in IgG4-RD
reduce the diffusion space in the extracellular and intracellular has been reported in the urethra and the pancreas.23,24 The
dimension with a resultant decrease in ADC value. Moreover, ADC value of orbital IgG4-RD in our study was also rela-
the interstitial changes in benign OLPDs would lead to an tively lower than that of other benign OLPDs, and over-
increased ADC value, which could also contribute to a difference lapped partially with that of the malignant OLPDs. The
in ADC value between benign and malignant OLPDs. There- lower ADC values do not certainly indicate the diagnosis of
fore, DWI and derived ADC values could serve as a promising malignant OLPDs, and IgG4-RD should also be considered,
imaging biomarker for differentiating benign and malignant thus the diagnostic specificity would be influenced. There-
OLPDs. However, the detailed ADC threshold value and diag- fore, the differentiation work between orbital IgG4-RD and
nostic accuracy in the present study was not consistent with those malignant OLPDs based on ADC measurement should be
of previous studies.3,12 Haradome et al reported that an ADC done carefully because of the potential overlap.
value of less than 0.612 3 1023 mm2/s was optimal for diagnos- The application of DCE-MRI in the orbit has been
ing orbital lymphoma, while 0.886 3 1023 mm2/s was the opti- reported with both 1.5T and 3.0T scanners.5,14 Compared

with previous studies using a 1.5T MR scanner, higher tem- tional MRI feature that might help to diagnose malignant
poral resolution could be achieved when a 3.0T MR scanner OLPDs. The specificity of differentiating work could be sig-
was used for DCE-MRI in the present study.5,14 High tem- nificantly improved by adding DWI and DCE-MRI to con-
poral resolution is a critical component of high-quality ventional MRI alone. Meanwhile, DWI alone demonstrated
DCE-MRI. It allows a more accurate assessment of the the optimal sensitivity in differentiating orbital benign and
hemodynamic process as gadolinium contrast agent passes malignant OLPDs.
through the microvasculature in the tissue of interest.16 The
AUC parameter derived from DCE-MRI in benign OLPDs REFERENCES
was significantly higher than that of the malignant OLPDs 1. Tailor TD, Gupta D, Dalley RW, Keene CD, Anzai Y. Orbital neo-
plasms in adults: clinical, radiologic, and pathologic review. Radio-
in the present study. The TIC pattern of these two entities
graphics 2013;33:1739–1758.
could help us to explain this finding. In our study, the TIC
2. Priego G, Majos C, Climent F, Muntane A. Orbital lymphoma: imaging
pattern of benign OLPDs mostly showed as Type I (persis- features and differential diagnosis. Insights Imaging 2012;3:337–344.
tent pattern), while the malignant ones mostly showed as 3. Haradome K, Haradome H, Usui Y, et al. Orbital lymphoproliferative
Type III (washout pattern), which was similar to a previous disorders (OLPDs): value of MR imaging for differentiating orbital lym-
phoma from benign OPLDs. AJNR Am J Neuroradiol 2014;35:1976–
study.14 Considering the similar TTP and Slopemax, natu- 1982.
rally the AUC would demonstrate a significant difference
4. Garner A. Orbital lymphoproliferative disorders. Br J Ophthalmol
between the two groups. However, despite that a significant 1992;76:47–48.
difference was found, the AUC showed considerable overlap 5. Xian J, Zhang Z, Wang Z, et al. Value of MR imaging in the differen-
between the two entitles, and thus the diagnostic ability was tiation of benign and malignant orbital tumors in adults. Eur Radiol
2010;20:1692–1702.
limited, and further attempts to employ multiparametric
6. Ben Simon GJ, Annunziata CC, Fink J, Villablanca P, McCann JD,
MRI techniques are needed.
Goldberg RA. Rethinking orbital imaging establishing guidelines for
With the combination of conventional qualitative interpreting orbital imaging studies and evaluating their predictive
image assessment, DWI, and DCE-MRI techniques, optimal value in patients with orbital tumors. Ophthalmology 2005;112:2196–
2207.
specificity could be achieved in differentiating benign from
7. Ro SR, Asbach P, Siebert E, Bertelmann E, Hamm B, Erb-Eigner K.
malignant OLPDs. In the clinical setting, the high specific- Characterization of orbital masses by multiparametric MRI. Eur J
ity could be valuable in enhancing our diagnostic determi- Radiol 2016;85:324–336.
nation for potential malignancy, which is crucial for doctor– 8. Xu XQ, Hu H, Su GY, et al. Orbital indeterminate lesions in adults:
patient communication and determination of a therapy combined magnetic resonance morphometry and histogram analysis
of apparent diffusion coefficient maps for predicting malignancy.
plan. However, the drawback of the combination of multi- Acad Radiol 2016;23:200–208.
parametric MRI was the low sensitivity, which was unfavora- 9. Sepahdari AR, Kapur R, Aakalu VK, Villablanca JP, Mafee MF.
ble for identifying suspicious malignant OLPDs patients. By Diffusion-weighted imaging of malignant ocular masses: initial results
and directions for further study. AJNR Am J Neuroradiol 2012;33:
contrast, when using the ADC alone as the differentiating
314–319.
index, optimal sensitivity could be achieved. Therefore, we
10. Xu XQ, Hu H, Su GY, et al. Utility of histogram analysis of ADC maps
suggest using DWI as the filtering metric to identify for differentiating orbital tumors. Diagn Interv Radiol 2016;22:
patients with suspicious malignant OLPDs, and using the 161–167.
combination of conventional MR, DW, and DCE-MRI as 11. Razek AA, Elkhamary S, Mousa A. Differentiation between benign and
malignant orbital tumors at 3-T diffusion MR imaging. Neuroradiology
the differentiating metric for diagnosing malignant OLPDs. 2011;53:517–522.
We acknowledge that this study has some limitations.
12. Sepahdari AR, Aakalu VK, Setabutr P, Shiehmorteza M, Naheedy JH,
First, it was a single-center retrospective study with a rela- Mafee MF. Indeterminate orbital masses: restricted diffusion at MR
tively small cohort, but we believe that our results are a imaging with echo-planar diffusion-weighted imaging predicts malig-
nancy. Radiology 2010;256:554–564.
strong basis for larger prospective studies. Second, the
13. Zhang Z, Shi J, Guo J, Yan F, Fu L, Xian J. Value of MR imaging in
model-based parameters derived from DCE-MRI are prefer- differentiation between solitary fibrous tumor and schwannoma in the
able, as they provide greater insight about microvasculature orbit. AJNR Am J Neuroradiol 2013;34:1067–1071.
physiology and tumor biology. A previous study pointed out 14. Yuan Y, Kuai XP, Chen XS, Tao XF. Assessment of dynamic contrast-
that the calculation of model-based parameters showed poor enhanced magnetic resonance imaging in the differentiation of malig-
nant from benign orbital masses. Eur J Radiol 2013;82:1506–1511.
reproducibility, which might be associated with the inter-
15. Carmeliet P, Jain RK. Angiogenesis in cancer and other diseases.
reader variability in arterial input function measurement.25 Nature 2000;407:249–257.
We chose model-free parameters because we wanted to min-
16. Gaddikeri S, Gaddikeri RS, Tailor T, Anzai Y. Dynamic contrast-
imize the dependence on complex pharmacokinetic model- enhanced MR imaging in head and neck cancer: techniques and clini-
ing, and model-free parameters have also been proven to be cal applications. AJNR Am J Neuroradiol 2016;37:588–595.
robust indicators of tumor vascular characteristics.26 17. Yuan M, Zhang YD, Zhu C, et al. Comparison of intravoxel incoherent
motion diffusion-weighted MR imaging with dynamic contrast-
Despite the limitations, our study showed no presence enhanced MRI for differentiating lung cancer from benign solitary pul-
of flow void sign on T2WI as the most significant conven- monary lesions. J Magn Reson Imaging 2016;43:669–679.
January 2017 175

18. Huang B, Wong CS, Whitcher B, et al. Dynamic contrast-enhanced sensitised driven-equilibrium preparation technique. Eur Radiol 2014;
magnetic resonance imaging for characterising nasopharyngeal carci- 24:581–586.
noma: comparison of semiquantitative and quantitative parameters
23. Choi JW, Kim SY, Moon KC, Cho JY, Kim SH. Immunoglobulin G4-
and correlation with tumour stage. Eur Radiol 2013;23:1495–1502.
related sclerosing disease involving the urethra: case report. Korean J
19. Jia ZZ, Gu HM, Zhou XJ, et al. The assessment of immature microvas- Radiol 2012;13:803–807.
cular density in brain gliomas with dynamic contrast-enhanced mag-
24. Taniguchi T, Kobayashi H, Nishikawa K, et al. Diffusion-weighted
netic resonance imaging. Eur J Radiol 2015;84:1805–1809.
magnetic resonance imaging in autoimmune pancreatitis. Jpn J Radiol
20. Dutoit JC, Claus E, Offner F, Noens L, Delanghe J, Verstraete KL. 2009;27:138–142.
Combined evaluation of conventional MRI, dynamic contrast-enhanced
25. Koh MJ, Kim HS, Choi CG, Kim SJ. Which is the best advanced MR
MRI and diffusion weighted imaging for response evaluation of patients
imaging protocol for predicting recurrent metastatic brain tumor fol-
with multiple myeloma. Eur J Radiol 2016;85:373–382.
lowing gamma-knife radiosurgery: focused on perfusion method.
21. Wu CJ, Wang Q, Li H, et al. DWI-associated entire-tumor histogram analy- Neuroradiology 2015;57:367–376.
sis for the differentiation of low-grade prostate cancer from intermediate-
26. Boxerman JL, Schmainda KM, Weisskoff RM. Relative cerebral blood
high-grade prostate cancer. Abdom Imaging 2015;40:3214–3221.
volume maps corrected for contrast agent extravasation significantly
22. Hiwatashi A, Yoshiura T, Togao O, et al. Diffusivity of intraorbital lym- correlate with glioma tumor grade, whereas uncorrected maps do
phoma vs. IgG4-related disease: 3D turbo field echo with diffusion- not. AJNR Am J Neuroradiol 2006;27:859–867.

ORIGINAL RESEARCH
Intrinsic Brain Network Abnormalities

in Codeine-Containing Cough
Syrup-Dependent Male Individuals
Revealed in Resting-State fMRI
Ying-wei Qiu, MD, PhD,1,2,3* Huan-Huan Su, MD,2 Xiao-fei Lv, MD, PhD,4
Xiao-fen Ma, MD, PhD,2 Gui-hua Jiang, MD, PhD,2* and Jun-zhang Tian, MD, PhD2*
Purpose: To identify codeine-containing cough syrups (CCS)-related modulations of intrinsic connectivity network (ICN)
and to investigate whether these changes of ICN can be related to duration of CCS use and to impulsivity behavior in
CCS-dependent individuals.
Materials and Methods: Resting-state functional magnetic resonance imaging (fMRI) data in 41 CCS-dependent individ-
uals and 34 healthy controls (HC) were scanned at 1.5T and analyzed using independent component analysis (ICA), in
combination with a “dual-regression” technique to identify the group differences of three important resting-state net-
works, the default mode network (DMN), the executive control network (ECN), and the salience network (SN) between
the CCS-dependent individuals and HC.
Results: Compared with the HC, CCS-dependent individuals had aberrant intrinsic connectivity within the DMN, ECN,
and SN (P < 0.05, AlphaSim corrected). Furthermore, a longer duration of CCS use was associated with greater abnor-
malities in the intrinsic network functional connectivity (FC, P < 0.05, Bonferroni correction). Intrinsic network FC also
correlated with higher impulsivity in CCS-dependent individuals (P < 0.05, AlphaSim corrected).
Conclusion: Our findings revealed aberrant DMN, ECN, and SN connectivity patterns in CCS-dependent patients, which
may provide new insight into how neuronal communication and information integration are disrupted among DMN,
ECN, and SN key structures due to long duration of CCS use.
C hronic drug abuse is a major threat to public health

and the social cohesion of modern societies. Drug
addiction is conceptualized as a syndrome of impaired
epidemiological study demonstrated that CCS abuse would
lead to mild forms of physical and psychiatric disorders, and
the withdrawal from CCS occurred in a manner similar to
response inhibition and salience attribution, characterized by heroin withdrawal, although in a comparatively milder
a compulsive drive to take drugs despite serious negative form.5 However, the neural biological mechanism underlying
consequences.1 Scientific advances over the last quarter cen- the CCS dependence in these young adults is still not clear.
tury have established that drug addiction is a chronic brain Recently, resting-state functional magnetic resonance
disease.2 According to the World Drug Report of 2012 and imaging (fMRI) has enabled evaluation of intrinsic brain
2013,3,4 the misuse of prescription or over-the-counter networks (ICNs) on the basis of baseline energy expenditure
(OTC) drugs, such as codeine-containing cough syrup in awake and resting-states of the brain.6 Spontaneous fluc-
(CCS), has been increasing rapidly worldwide. An tuations in the low-frequency (<0.1 Hz) blood oxygen
Received Dec 7, 2015, Accepted for publication Jun 6, 2016.
*Correspondence to: Y.-w.Q. or G.-h.J., J.-z.T, Department of Medical Imaging, Guangdong No. 2 Provincial People’s Hospital,
Guangzhou 510317, P.R. China. E-mail: qiuyw1201@gmail.com or yingwei.qiu@duke-nus.edu.sg; or jghimage@gmail.com; jiangguihua1970@163.com
From the 1Department of Medical Imaging, Zhongshan Ophthalmic Center, Sun Yat-sen University, Guangzhou, P.R. China; 2Department of Medical
Imaging, Guangdong No. 2 Provincial People’s Hospital, Guangzhou, P.R. China; 3Center for Cognitive Neuroscience, Neuroscience and Behavioral
Disorders Program, Duke-National University of Singapore Graduate Medical School, Singapore; and 4Sun Yat-sen University Cancer Center, State Key
Laboratory of Oncology in South China, Collaborative Innovation Center for Cancer Medicine, Guangzhou, P.R. China.
Additional Supporting Information may be found in the online version of this article.

V 177

TABLE 1. Demographic and Clinical Characteristics of the Chronic Codeine-Containing Cough Syrups (CCS)-
Dependent Individuals and Controls
Characteristic CCS-dependent Controls P value

individuals (n 5 41) (n 5 34)
Age (years) 24.85 (2.877) 24.00(2.387) 0.114

Education (years) 13.24 (2.672) 12.216 (3.391) 0.593
Nicotine ((no. cigarette/day)) 14.88 (8.931) 13.68 (8.466) 0.115
Head motion (mean FD_Jenkinson) 0.141 (0.131) 0.116 (0.113) 0.361
Cough syrups use (years) 5.41 (range:1–12) N/A —
Age of first use cough syrups 20.30 (range:12–30) N/A —
Mean dose (ml/d) 477.58 (range:60–1800) N/A —
Total BIS scores 71.56 (4.336) 56.76(5.336) 0.000*
Attentional impulsivity 17.88 (2.848) 15.56(1.691) 0.000*
Motor impulsivity 30.66 (2.456) 22.12(3.346) 0.000*
Nonplan impulsivity 23.02 (1.837) 19.26(2.799) 0.000*
Unless otherwise indicated, data are means 6 standard deviations. NA, not applicable. *P < 0.05.
level-dependent signal have been recognized to reliably these ICNs is unknown. Research on ICNs in CCS-
organize into spatially independent networks, which are dependent individuals might provide useful data to confirm
believed to represent the maintenance of baseline human the network impairment hypothesis. We hypothesized that
cognition and metabolic equilibrium (ie, oxygen balance).7 the DMN, ECN, and SN connectivity patterns in CCS-
These independent networks were also named resting-state dependent individuals would be also altered.
networks (RSNs), which include the functional architecture In the present study, we mainly aim to verify 1)
of memory networks, language, attention, visual, auditory whether the functional connectivity (FC) of ICNs, particular
and somatomotor, and so on.8 the DMN, ECN, and SN, can be altered in CCS-
Of the multiple ICNs that exist,9 three are particularly dependent individuals, and 2) if so, whether these changes
relevant to the study of substances addiction because their are related to the measured clinical scales. The results of this
addicted-related changes and associated addictive behaviors research may employ a new manner to investigate the neu-
have been replicated in multiple studies.10–13 The three rological and neurophysiological mechanisms of CCS-
ICNS are: the default mode network (DMN), the executive dependent individuals.
control network (ECN), and the salience network (SN).14
The DMN, including areas of the medial prefrontal cortex
(MPFC), the posterior cingulate cortex (PCC), and the pre-
cuneus, is associated with internally oriented mentation and Participants
This prospective study was approved by the local Institutional
autobiographical memory,7,15 while the ECN, mainly
Review Board. Written informed consent was obtained from all
involving the dorsolateral prefrontal and the parietal regions,
subjects. Seventy-five right-handed male participants, including 34
is engaged in higher-order cognitive processes and externally
control subjects and 41 CCS-dependent individuals participated in
oriented attention,16 and the SN, comprising paralimbic this study (see Table 1 for the demographic information). All T1
structures—most prominently the anterior insular (AI) and CCS-dependent individuals were screened based on the DSM-IV
medial frontal areas such as the anterior cingulate cortex criteria from the medical history, along with a urine test and an
(ACC) and presupplementary motor area (pre-SMA), is interview. All patients were naive for any type of illicit drug and
involved in switch between the DMN and task-related net- alcohol use. Details for the criterions for the healthy controls are
works in cognitive control.17 All of the DMN, ECN, and presented in Supplementary Methods.
SN are related to cognition, and potentially relate to addic-
tive behaviors. Previous resting-state fMRI studies have Impulsivity Assessment
reported abnormalities of DMN, ECN, and SN connectiv- Impulsivity was assessed using the Barratt Impulsiveness Scale
ity patterns in other substance-dependent individuals.10–13 (BIS.11).18 The details for the BIS.11 assessment are presented in
Whether CCS dependency can also lead to dysfunction in the Supplementary Methods.

Qiu et al.: Network Abnormalities in CCS-Dependent Individuals
TABLE 2. Regions That Showed Significant Changes in Resting State FC Between CCS-Dependent Individuals
and Controls Within DMN, ECN, and SN
MNI, mm
Brain regions
RSN (Brodmann Area) Side Voxels x y z Peak values
DMN STR/PUT/CN/NAcc R 125 9 15 -3 -4.323

PCUN/SOG/CUN (BA 19) L 133 -24 -66 33 3.6027
ECN SFG/MFG (BA 9/10/46) R 109 15 54 18 -3.2046
IFG/MFG (BA9/45/46) L 137 -54 18 27 -3.5832
SFGmed (BA 6/8) R/L 87 -6 36 45 -3.4997
ANG/PCUN/IPL (BA 19) L 62 42 -75 40 -3.856
SN IPL/STG/SMAR (BA40) R 148 58 -29 21 -3.9223
MCC/SMA (BA 24) R/L 254 -2 -6 49 -3.7248
ROL/STG/PreCG (BA6/22) R 154 57 -2 7 -3.4493
CCS, codeine-containing cough syrups; FC, functional connectivity; RSN, resting state network; DMN, default mode network; ECN,
executive control network; SN, salience network; MNI, Montreal Neurological Institute; STR, striatum; PUT, putamen; CN, caudate
nucleus; NAcc, Nucleus Accumbens; PCUN, precuneus; SOG, superior occipital gyrus; CUN, cuneus; SFG, superior frontal gyrus;
MFG, middle frontal gyrus; IFG, inferior frontal gyrus; SFGmed, medial superior frontal gyrus; ANG, angular gyrus; IPL, inferior
parietal lobe; STG, superior temporal gyrus; SMAR, supramarginal gyrus; MCC, middle cingulate cortex; SMA, supplementary motor
area; ROL, ronlandic; PreCG, precentral gyrus; BA, Brodmann area; L, left; R, right.
MRI Scanning normalized to the standard structural MRI template in the Mon-
MRI data were obtained on a Philips (Best, Netherlands) Achieva treal Neurologic Institute space using nonlinear transformation.
1.5T Nova dual MR scanner using a 16-channel NV coil. Each par- The same transformation parameters were applied to the fMRI
ticipant underwent a T1-weighted structural MRI and a resting-state images. To remove the sources of possible spurious variance from
fMRI in the same session. Sagittal structural images (160 sagittal sli- each voxel’s fMRI time series, we 1) removed linear trends, 2)
ces, repetition time [TR] 5 25 msec, echo time [TE] 5 4.1 msec, removed nuisance signals (white matter, cerebrospinal fluid signals,
thickness 5 1.0 mm, no gap, in-plane resolution 5 231 3 232, six head-motion parameters), and 3) applied temporal bandpass fil-
field of view [FOV] 5 230 3 230 mm2, flip angle 5 308) were tering (0.01–0.08 Hz). The fMRI data was resampled at a resolu-
acquired using a fast field echo (FFE) 3D T1-weighted sequence. tion of 3 3 3 3 3 mm3 and spatially smoothed fMRI data with a
Resting-state fMRI scans were performed by an echo planar imaging 6-mm full-width at half-maximum isotropic Gaussian kernel.
(EPI) sequence with scan parameters of TR 5 2000 msec, TE 5
INDEPENDENT COMPONENT ANALYSIS (ICA) AND IDENTIFI-
50 msec, flip angle 5 908, matrix 5 64 3 64, FOV 5 230 3 230
CATION OF RSNS. Group spatial ICA was conducted for all the
mm2, slice thickness 5 4.5 mm and slice gap 5 0 mm. Each brain
75 subjects using the infomax algorithm within the GIFT software
volume comprised 22 axial slices and each functional run contained
(http://icatb.sourceforge.net/), which incorporates temporal concate-
240 volumes (8 min). During resting state fMRI scanning, subjects
nation plus back-reconstruction (produces subject-specific images).20
were instructed to close their eyes and keep as still as possible, and
This enables a comparison of both the time courses and the images
not to think of anything systematically or fall asleep. After the scan,
for one group or multiple groups.20 A detailed review of group ICA
all the participants were asked questions to verify the degree of their
fMRI analyses can be found in Calhoun et al.21 In this study, the
cooperation; details are presented in Supplementary Methods.
images were reduced to 40 dimensions using principal component
Data Analysis analysis, and the optimal number of components was found to be
34 using the minimum description length criteria modified to
DATA PREPROCESSING. Functional image preprocessing was account for spatial correlation.22 We applied the ICASSO algorithm
performed using the Data Processing & Analysis of Brain Imaging implemented in GIFT to increase the robustness of our independent
(DPABI 1.2)19 with statistical parametric mapping (SPM8, http:// components to initial algorithm conditions by repeating the ICA
www.fil.ion.ucl.ac.uk/spm/), following previous approaches.19 The estimation 100 times. Single subjects’ spatial maps and correspond-
first 10 images were discarded to allow for signal stabilization and ing time courses were then computed and converted to z-scores for
subject adaptation with the remaining images corrected for slice display and use in further statistical analyses. Each voxel in the brain
time differences and head motion (a least-squares approach and a has a z-score representing the strength of its contribution to the
six-parameter spatial transformation). Subjects with head motion component’s time course. We selected DMN, ECN, and SN com-
exceeding 1.0 mm in any dimension through the resting-state run ponents using an automated and objective method. In detail, we
were discarded from further analysis. Individual functional images chose from the network template23 (http://findlab.stanford.edu/func-
were then coregistered to T1-weighted MR images that were later tional_ROIs.html) representing subsystems of SN, DMN, and
segmented (gray matter, white matter, and cerebrospinal fluid) and ECN, and performed multiple spatial regression analyses of our 34
January 2017 179

C
O
L
O
R
FIGURE 1: Group level DMN (A), ECN (B), and SN (C) connectivity in CCS-dependent individuals (left) and controls (right) (P <
0.01, FDR-corrected). CCS 5 codeine-containing cough syrups; DMN 5 default mode network; ECN 5 executive control network;
SN 5 salience network. Color bar represents t value.
independent components’ spatial maps on these templates. We use, initial age of CCS use in CCS-dependent individuals, we per-
selected components with the highest correlation coefficient with the formed correlation analyses between the aberrant RSNs FC
templates (details can be found in the Supplementary Materials). In (revealed by group comparison) and the duration of CCS use, ini-
addition, all selected components were also visually inspected by two tial age of CCS use in SPSS. First, for each CCS-dependent indi-
researchers to ensure the selected DMN, ECN, and SN corre- vidual, the mean Z values of regions showing group differences
sponded to the cerebral components with the largest spatial correla- were extracted separately using DPABI.19 Pearson’s correlations
tions with the network templates.24 were then computed between the mean Z values of each CCS-
dependent individual and the duration of CCS use and initial age
SECOND-LEVEL ANALYSIS OF THE RSNS. The components in
of CCS use. The results are reported at a significance level of P <
line with three RSNs were extracted from all subjects. One-sample
t-tests were then calculated in each group to gather the spatial maps 0.05 with Bonferroni correction for multiple comparisons (eight
of each RSNs. Thresholds were set at P < 0.01 (false discovery rate brain regions as showed in Table 2). T2
[FDR]-corrected) (see details in the Supplementary Materials). To
compare the changes of the RSNs between CCS-dependent individ- ASSOCIATIONS BETWEEN RSNS AND IMPULSIVITY. To evalu-
uals and controls, two-sample t-tests were calculated. Age, education, ate the relationship between the RSNs FC and impulsivity meas-
cigarette smoking, and head motion were entered as covariates. urements, we performed a voxel-based correlation analysis of RSNs
Thresholds were set at P < 0.05 (AlphaSim-corrected). By using FC against four impulsivity indices (BIS total scores, attentional
this program, a corrected significance level of P < 0.05 was impulsivity, motor impulsivity, and nonplanning impulsivity) in
obtained by clusters with a minimum volume of 999 mm3 at an CCS-dependent individuals and controls individually. Age, educa-
uncorrected individual voxel height threshold of P < 0.01. This tion, cigarette smoking, and head motion were entered as covari-
enabled the identification of significant changes in FC in CCS- ates. Thresholds were set at P < 0.05 (AlphaSim corrected).
dependent individuals compared with healthy control subjects.
(Parameters were individual voxel P value of 0.01, 1000 simulations,
Effects of Head Motion
estimated smoothness, and gray matter mask.). The group compari-
Head motion has been considered a confounding factor on resting-
sons were masked to the voxels within corresponding RSNs. The
state FC in the recent literature.25 In this study, we discarded sub-
mask was created by combining the regions of corresponding RSNs
jects with head motion exceeding 1.0 mm of maximal translation
in both CCS-dependent individuals and controls, which were
(in any direction of x, y, or z) or 1.08 of maximal rotation through
obtained from one-sample t-tests results (P < 0.01, FDR-corrected).
the resting state run. In addition, we also performed a two-sample
Correlation Analyses t-test on the mean absolute estimated movement parameters (shift
and rotation) on all three axes to examine between-group differen-
ASSOCIATIONS BETWEEN ABERRANT RSNS FC AND DURA- ces in the degree of head motion. We also performed statistical
TION OF CCS USE, INITIAL AGE OF CCS USE. To evaluate the analysis with head motion as an extra covariate in the multiple
associations between aberrant RSNs FC and the duration of CCS regression models.

C
O
L
O
R
FIGURE 2: Group comparison maps of the DMN (A), ECN (B), and SN (C) contrasting CCS-dependent individuals versus controls
(P < 0.05, AlphaSim-corrected). Hot color denotes increased FC in CCS-dependent individuals, cold color denotes decreased FC in
CCS-dependent individuals. The coordinates of significantly different regions are shown in Table 2. CCS, codeine-containing cough
syrups; DMN, default mode network; ECN, executive control network; SN, salience network. Color bar represents t value.
January 2017 181

TABLE 3. Correlation Between FC of Abnormal Brain Regions and Duration of CCS Use in CCS-Dependent
Individuals
MNI coordinate
Correlation
RSN Brain areas x y z coefficient P values
DMN STR/PUT/CN/NAcc 9 15 -3 -0.471 0.002**

PCUN/SOG/CUN -24 -66 33 0.099 0.540
ECN SFG/MFG 15 54 18 -0.352 0.024*
IFG/MFG -54 18 27 -0.342 0.029*
SFGmed -6 36 45 -0.431 0.005**
ANG/PCUN/IPL 42 -75 40 -0.411 0.008**
SN IPL/STG/SMAR 58 -29 21 -0.336 0.032*
MCC/SMA -2 -6 49 -0.341 0.029*
ROL/STG/PreCG 57 -2 7 -0.333 0.033*
MNI, Montreal Neurological Institute. *P < 0.05 (without multiple comparison correction). **P < 0.05 (Bonferroni correction).
CCS, codeine-containing cough syrups; FC, functional connectivity; RSN, resting state network; DMN, default mode network; ECN,
executive control network; SN, salience network; MNI, Montreal Neurological Institute; STR, striatum; PUT, putamen; CN, caudate
nucleus; NAcc, Nucleus Accumbens; PCUN, precuneus; SOG, superior occipital gyrus; CUN, cuneus; SFG, superior frontal gyrus;
MFG, middle frontal gyrus; IFG, inferior frontal gyrus; SFGmed, medial superior frontal gyrus; ANG, angular gyrus; IPL, inferior
parietal lobe; STG, superior temporal gyrus; SMAR, supramarginal gyrus; MCC, middle cingulate cortex; SMA, supplementary motor
area; ROL, ronlandic; PreCG, precentral gyrus.
Results individuals had weaker intranetwork connectivity in the bilat-

Demographic and Behavioral Measures eral dorsal lateral PFC (dlPFC), the bilateral medial frontal
There were no significant differences in age, formal years of gyrus and right angular for the ECN, the right putamen for
education, or number of cigarettes smoked per day between the DMN, the bilateral middle cingulate cortex (MCC) and
the chronic CCS-dependent individuals and controls. The the right temporal cortex for the SN, and enhanced FC
mean duration of CCS use was 5.41 years. Mean age of first within the left precuneus for the DMN.
use of CCS was 20.30 years. Mean dose of CCS used was
477.58 ml/d in CCS users (Table 1). None of the partici- Weaker FC Within DMN, ECN, and SN Correlated
pants was found to have excessive movement (translation With CCS Duration in CCS-Dependent Individuals
exceeded 1.0 mm or rotation exceeded 1.08). There were no Longer duration of CCS use correlated with weaker stria-
significant differences in head motion during the resting tum, medial frontal gyrus, and angular FC within the
state fMRI between these two groups (Table 1). DMN and ECN separately, and the trend correlated with
A comparison of average BIS.11 scores from the SN connectivity (Table 3). The relationship between these T3
chronic CCS-dependent individuals and control groups is FC abnormalities and the initial age of drug use were also
showed in Table 1. CCS-dependent individuals had signifi- checked. No results exceeded the threshold.
cantly higher attentional impulsivity, motor impulsivity,
nonplan impulsivity, and total scores than the control group Impulsivity Behavior Correlates With RSNs FC in
(P < 0.05). CCS-Dependent Individuals
BIS total scores correlated with the FC of the bilateral stria-
Spatial Distribution of DMN, ECN, and SN tum within the DMN and left dorsal lateral prefrontal cor-
Spatial distributions of DMN, ECN, and SN for both
tex (DLPFC) within the ECN; while the attentional
F1 groups are shown in Fig. 1. (Details can be found in Sup-
impulsivity correlated with the right inferior parietal lobe
plementary Materials). The spatial distribution of the three
(IPL) within the ECN and bilateral middle cingulate cortex
particular networks is the same as found by previous
(MCC) within the SN in CCS-dependent individuals (Fig.
studies.26,27
3). Moreover, we found that the regions (bilateral striatum F3
Aberrant DMN, ECN, and SN in CCS-Dependent and right IPL) showed correlations with the impulsivity in
Individuals CCS-dependent individuals and overlay the regions (right
Groups comparisons demonstrated significant differences striatum and right IPL) showed group FC differences (Sup-
between CCS-dependent individuals and controls (P < 0.05, plementary Fig. 1). We did not find any correlations
F2 AlphaSim-corrected) (Fig. 2, Table 2). CCS-dependent between the impulsivity and RSNs FC in healthy controls.

C
O
L
O
R
FIGURE 3: BIS total scores correlated with the functional connectivity of the bilateral striatum within the DMN (A) and left dorsal
lateral prefrontal cortex (DLPFC) within the ECN (B); while the attentional impulsivity correlated with the right inferior parietal
lobe (IPL) within the ECN (C) and bilateral middle cingulate cortex (MCC) within the SN (D) in CCS-dependent individuals. DMN,
default mode network; ECN, executive control network; SN, salience network. Color bar represents r value.
Discussion aberrant DMN, ECN, SN FC and higher impulsivity behav-

Given that the DMN, ECN, and SN are particularly relevant ior, as well as CCS duration, provide an additional important
to addicted-related changes and associated addictive behav- clue to the pathophysiology underlying CCS dependent.
iors,10–13 we studied the resting-state FC of these three par- Our findings are consistent with previous resting-state
ticular networks and their associations with the clinical FC studies investigating illicit drug dependency, showing
parameters in CCS-dependent individuals. As hypothesized, abnormalities in connectivity in networks involving the
CCS-dependent individuals had aberrant FC within the striatum, MCC, bilateral dorsal lateral prefrontal cortex, and
DMN, ECN, and SN. Longer duration of CCS use was asso- bilateral medial prefrontal areas.12,28 These brain regions
ciated with greater abnormalities of FC within the DMN, have also been implicated in task studies investigating illicit
ECN, and SN. Moreover, higher impulsivity correlated with drug addiction.29,30
lower FC within the DMN, ECN, and SN in CCS- The DMN is engaged in the maintenance of the base-
dependent individuals. The observed relationships between line brain activities related to cognition of self-awareness,
January 2017 183

episodic memory, and interactive modulation between the study of drug abuse and relapse: attending to and assessing
internal mind activities and external tasks.15 Brain regions situations, recognizing potential consequences, making deci-
of the DMN are known to be involved in different high- sions, planning and coordinating behavior, and inhibiting
level cognitive functions; the deficits in the DMN suggest inappropriate behavior.39 Thus, FC deficits in the bilateral
higher cognitive impairment in CCS-dependent individuals, dorsolateral and bilateral medial frontal cortices may impact
such as autobiographical episodic memory, self-referential the executive system in users of CCS-dependent individuals.
processes, and social cognitive processes. Our findings Previous studies have suggested the primary role of the
revealed disassociated changes within the DMN FC in SN, which may operate to dynamically control changes of
CCS-dependent individuals, with decreased FC in the right activity of other brain networks.17 Specifically, the SN plays
striatum and increased FC in the left precuneus. Previous a primary role in the integration of sensations, internally
functional neuroimaging studies have suggested that the generated thoughts, and information about goals and plans
striatum is a key region involved in the reward circuit,31 in order to update expectations about the internal or an
and the dysfunction of left striatum in CCS-dependent indi- external milieu and, if necessary, to allow action to be initi-
viduals may be responsible for the dysregulation of the ated or modified.40 The deficits in the SN might suggest
reward circuit, which may result in higher impulsivity. “salience” processing impairment in CCS-dependent indi-
The precuneus is a brain region in the posteromedial viduals, such as task control, and switch between the DMN
cortex of the parietal lobe and plays an important role in and task-related networks in cognitive control. We found
fundamental cognitive functioning.32 The precuneus has decreased FC in middle cingulate gyrus and right temporal
been proposed to be involved in episodic memory retrieval, cortex within SN in CCS-dependent individuals. Functional
visual-spatial imagery, self-processing, and consciousness.32 neuroimaging studies have consistently implicated the activ-
The increased FC in the precuneus observed in present ity of the dorsal and posterior portions of the medial pre-
study is in line with the hyperfunction of the precuneus in frontal cortex, specifically the mid-cingulate cortex (MCC),
other addictions. Regarding online gaming addiction, in detecting preresponse conflict between intended actions
researchers reported increased regional homogeneity and and their outcomes. The decreased FC in MCC in CCS-
amplitude of low-frequency fluctuation of the spontaneous dependent individuals is not surprising, and potentially
neuronal activity in the left precuneus in online-gaming could underlie their addiction-related deficits in adaptive
addicts compared with controls.33 Moreover, one study decision making and self-control, especially when the deci-
showed that the precuneus was associated with a gaming sion to stop drug use conflicts with the drive to keep
urge, craving, and severity of online gaming addicts, and using.30
suggested that the precuneus activates to process the gaming Interestingly, most of the regions revealed by group
cue, integrate retrieved memory, and contribute to cue- comparison, except for the left precuneus, were negatively
induced craving for online gaming.34 Our previous study of correlated (or trend related) with duration of CCS use. This
heroin addicts also found increased amplitude of low- finding suggests that chronic CCS consumption has a
frequency fluctuation of the spontaneous neuronal activity cumulative effect—the longer the CCS use, the lower the
in the left precuneus in heroin addicts when compared with resting-state FC in these regions. Our previous studies of
controls. In addition, we found that the hyperactivity of the the white matter microstructure and gray matter volume in
precuneus was related to methadone treatment.35 Thus, the CCS-dependent individuals also demonstrated this
increased FC of the left precuneus may be either related to effect.41,42 This phenomenon is not surprising in illicit drug
the craving or related to the reimbursement mechanism, addiction.43 Yuan et al44 found that the gray matter density
which needs to be verified in future study. of most brain areas (prefrontal, temporal, and insular) show-
The executive control network (ECN) plays a critical ing group differences were negatively correlated with the
role in the active maintenance and manipulation of infor- duration of heroin use in heroin-dependent individuals.
mation in working memory, and in judgment and decision Monterosso et al45 showed a negative correlation between
making in the context of goal-directed behavior.36 Previous duration of heroin use and performance in a stop-signal
neuroimaging studies of drug addiction consistently reveal task. Together with previous studies, we demonstrated that
cortical abnormal activation across regions of the ECN,37,38 chronic CCS consumption can impair brain function, as
and dysregulation of this network is hypothesized to account well as the structure, progressively as other illicit drugs do.
for the lack of inhibitory control that is a hallmark of the Hence, early intervention may be particularly important for
addictive phenotype. Our study also found the decreased the treatment of CCS addiction.
FC of the bilateral dorsal lateral PFC, bilateral medial fron- Impulsivity behavior correlated with the bilateral stria-
tal gyrus, and right angular within the ECN in CCS- tum, left DLPFC, right IPL, and bilateral MCC FC within
dependent individuals compared with controls. Frontal and the DMN, ECN, and SN separately in CCS-dependent
prefrontal areas mediate executive functions relevant in the individuals. Importantly, some of the regions showed

correlation with impulsivity behavior overlaid with the

Foundation of Jiangxi Province, China; contract grant
regions showing group differences (Supplementary Fig. 2). number: 20151BAB205049; Contract grant sponsor:
These findings support the idea that the impairment of Planned Science and Technology Project of Guangdong
these ICNs FC underlay the higher impulsivity behavior in Province, China; contract grant number: 2013B021800172.
drug addicts. Thus, the ICNs FC could be potential bio- We thank LetPub (www.letpub.com) for linguistic assistance
markers when exploring the brain function. during the preparation of the article.
Some limitations in our study are worth mentioning.
First, as a case–control cross-sectional study, we can only
observe the abnormal FC within the DMN, ECN, and SN
in the CCS-dependent individuals. However, we cannot abso- References
lutely assert the causation of the CCS abuse and the abnor- 1. Volkow ND, Li TK. Drug addiction: the neurobiology of behaviour
gone awry. Nat Rev Neurosci 2004;5:963–970.
malities of resting-state FC. Second, the results of disrupted
2. Leshner AI. Addiction is a brain disease, and it matters. Science 1997;
DMN, ECN, and SN connectivity can be caused by struc- 278:45–47.
tural connectivity abnormalities (eg, axonal injury), which 3. UNODC. World Drug Report 2013. United Nations Publication,
were not investigated in the current study. Future studies Vienna; 2013.
designed to explore the relationship between structural and 4. UNODC. World Drug Report 2012. United Nations Publication,
functional network deficits within DMN, ECN, and SN Vienna; 2012.
nodes will be valuable. Third, we only enrolled male patients, 5. Wairagkar NS, Das J, Kumar S, et al. Codeine containing cough syrup
addiction in Assam and Nagaland. Indian J Psychiatry 1994;36:129.
so females must be included to fully understand the intrinsic
6. Damoiseaux JS, Rombouts SA, Barkhof F, et al. Consistent resting-
network deficits related to CCS abuse in future study. state networks across healthy subjects. Proc Natl Acad Sci U S A
Fourth, the 1.5T MR scanner used in the present study has a 2006;103:13848–13853.
relatively lower signal-to-noise ratio and functional contrast- 7. Gusnard DA, Raichle ME, Raichle ME. Searching for a baseline: func-
to-noise ratios when compared to a 3.T MR scanner, which tional imaging and the resting human brain. Nat rev Neurosci 2001;2:
685–694.
may be another potential limitation.46 Another potential
8. Fox MD, Corbetta M, Snyder AZ, Vincent JL, Raichle ME. Spontane-
limitation is that the group differences in the present study ous neuronal activity distinguishes human dorsal and ventral attention
only survived AlphaSim correction for multiple comparisons systems. Proc Natl Acad Sci U S A 2006;103:10046–10051.
(did not survived FDR correction); therefore, the result 9. Smith SM, Fox PT, Miller KL, et al. Correspondence of the brain’s
functional architecture during activation and rest. Proc Natl Acad Sci
should be interpreted with caution. Finally, we did not evalu- U S A 2009;106:13040–13045.
ate the withdrawal symptoms in the present study, and evi-
10. Wang L, Zou F, Zhai T, et al. Abnormal gray matter volume and
dence from a previous study showed that the severity of resting-state functional connectivity in former heroin-dependent indi-
opioid withdrawal correlated with the local and remote corti- viduals abstinent for multiple years. Addict Biol 2016;21:646–656.
cal functional connectivity47; the potential effects of with- 11. Zhang Y, Gong J, Xie C, et al. Alterations in brain connectivity in
three sub-regions of the anterior cingulate cortex in heroin-
drawal symptom may also exist in the present study. dependent individuals: Evidence from resting state fMRI. Neuro-
In conclusion, we applied an ICA approach to analyze science 2015;284:998–1010.
resting-state fluctuations in CCS-dependent individuals to 12. Ma N, Liu Y, Li N, et al. Addiction related alteration in resting-state
brain connectivity. NeuroImage 2010;49:738–744.
gain insight into the RSNs’ abnormalities associated with
their impulsivity. The findings showed direct evidence of 13. Ma X, Qiu Y, Tian J, et al. Aberrant default-mode functional and
structural connectivity in heroin-dependent individuals. PLoS One
brain FC dysfunction within the DMN, ECN, and SN. 2015;10:e0120861.
Moreover, longer duration of CCS use was associated with 14. Menon V. Large-scale brain networks and psychopathology: a unifying
greater abnormalities in FC. Greater abnormalities in FC, triple network model. Trends Cogn Sci 2011;15:483–506.
especially for the ECN, were associated with higher impul- 15. Fox MD, Snyder AZ, Vincent JL, Corbetta M, Van Essen DC, Raichle ME.
The human brain is intrinsically organized into dynamic, anticorrelated
sivity. This study advanced our understanding of CCS functional networks. Proc Natl Acad Sci U S A 2005;102:9673–9678.
dependency towards a functional integral point of view of
16. Bressler SL, Menon V. Large-scale brain networks in cognition: emerg-
large-scale brain networks. The observed relationships ing methods and principles. Trends Cogn Sci 2010;14:277–290.
between DMN, ECN, SN disruption and impulsivity, as 17. Sridharan D, Levitin DJ, Menon V. A critical role for the right fronto-
well as clinical parameters, provide an additional important insular cortex in switching between central-executive and default-
mode networks. Proc Natl Acad Sci U S A 2008;105:12569–12574.
clue to the pathophysiology underlying CCS dependency.
18. Patton JH, Stanford MS, Barratt ES. Factor structure of the Barratt
impulsiveness scale. J Clin Psychol 1995;51:768–774.
Acknowledgment 19. Yan CG, Zang YF. DPARSF: A MATLAB toolbox for “pipeline” data
analysis of resting-state fMRI. Front Syst Neurosci 2010;4:13.
Contract grant sponsor: Natural Scientific Foundation of
20. Calhoun VD, Adali T, Pearlson GD, Pekar JJ. A method for making
China; contract grant numbers: 81201084, 81560283, group inferences from functional MRI data using independent compo-
81471639; Contract grant sponsor: Natural Scientific nent analysis. Hum Brain Mapp 2001;14:140–151.
January 2017 185

21. Calhoun VD, Liu J, Adali T. A review of group ICA for fMRI data and Internet gaming addiction and in remitted subjects. Addict Biol 2013;
ICA for joint inference of imaging, genetic, and ERP data. Neuro- 18:559–569.
Image 2009;45(1 Suppl):S163–172.
35. Jiang GH, Qiu YW, Zhang XL, et al. Amplitude low-frequency oscilla-
22. Li YO, Adali T, Calhoun VD. Estimating the number of independent tion abnormalities in the heroin users: a resting state fMRI study. Neu-
components for functional magnetic resonance imaging data. Hum roImage 2011;57:149–154.
Brain Mapp 2007;28:1251–1266.
36. Bunge SA, Ochsner KN, Desmond JE, Glover GH, Gabrieli JD. Pre-
23. Shirer W, Ryali S, Rykhlevskaia E, Menon V, Greicius M. Decoding frontal regions involved in keeping information in and out of mind.
subject-driven cognitive states with whole-brain connectivity patterns. Brain 2001;124(Pt 10):2074–2086.
Cereb Cortex 2012;22:158–165.
37. Droutman V, Read SJ, Bechara A. Revisiting the role of the insula in
24. van de Ven VG, Formisano E, Prvulovic D, Roeder CH, Linden DE. addiction. Trends Cogn Sci 2015;19:414–420.
Functional connectivity as revealed by spatial independent compo-
38. Qiu Y-w, Jiang G-h, Su H-h, et al. Short-term UROD treatment on cer-
nent analysis of fMRI measurements during rest. Hum Brain Mapp
ebral function in codeine-containing cough syrups dependent male
2004;22:165–178.
individuals. Eur Radiol 2015:1–10.
25. Yan CG, Cheung B, Kelly C, et al. A comprehensive assessment of
39. Aron JL, Paulus MP. Location, location: using functional magnetic res-
regional variation in the impact of head micromovements on func-
onance imaging to pinpoint brain differences relevant to stimulant
tional connectomics. NeuroImage 2013;76:183–201.
use. Addiction 2007;102(Suppl 1):33–43.
26. Raichle ME, MacLeod AM, Snyder AZ, Powers WJ, Gusnard DA,
40. Palaniyappan L, Liddle PF. Does the salience network play a cardinal
Shulman GL. A default mode of brain function. Proc Natl Acad Sci U
role in psychosis? An emerging hypothesis of insular dysfunction.
S A 2001;98:676–682.
J Psychiatry Neurosci 2012;37:17–27.
27. Greicius MD, Srivastava G, Reiss AL, Menon V. Default-mode network
41. Qiu YW, Su HH, Lv XF, Jiang GH. Abnormal white matter integrity in
activity distinguishes Alzheimer’s disease from healthy aging: evidence
chronic users of codeine-containing cough syrups: a tract-based spa-
from functional MRI. Proc Natl Acad Sci U S A 2004;101:4637–4642.
tial statistics study. AJNR Am J Neuroradiol 2015;36:50–56.
28. Qiu Yw, Jiang Gh, Ma Xf, Su HH, Lv Xf, Zhuo Fz. Aberrant interhemi-
42. Qiu YW, Lv XF, Jiang GH, et al. Reduced ventral medial prefrontal
spheric functional and structural connectivity in heroin-dependent
cortex (vmPFC) volume and impaired vmPFC-default mode network
individuals. Addict Biol 2016 [Epub ahead of print].
integration in codeine-containing cough syrups users. Drug Alcohol
29. Sjoerds Z, de Wit S, van den Brink W, et al. Behavioral and neuroi- Depend 2014;134:314–321.
maging evidence for overreliance on habit learning in alcohol-
43. Qiu YW, Jiang GH, Su HH, et al. The impulsivity behavior is corre-
dependent patients. Transl Psychiatry 2013;3:e337.
lated with prefrontal cortex gray matter volume reduction in heroin-
30. Goldstein RZ, Volkow ND. Dysfunction of the prefrontal cortex in dependent individuals. Neurosci Lett 2013;538:43–48.
addiction: neuroimaging findings and clinical implications. Nat Rev
44. Yuan Y, Zhu Z, Shi J, et al. Gray matter density negatively correlates
Neurosci 2011;12:652–669.
with duration of heroin use in young lifetime heroin-dependent indi-
31. Haruno M, Kawato M. Different neural correlates of reward expecta- viduals. Brain Cogn 2009;71:223–228.
tion and reward expectation error in the putamen and caudate
45. Monterosso JR, Aron AR, Cordova X, Xu J, London ED. Deficits in
nucleus during stimulus-action-reward association learning.
response inhibition associated with chronic methamphetamine abuse.
J Neurophysiol 2006;95:948–959.
Drug Alcohol Depend 2005;79:273–277.
32. Cavanna AE, Trimble MR. The precuneus: a review of its functional
46. Voss HU, Zevin JD, McCandliss BD. Functional MR imaging at 3.0 T
anatomy and behavioural correlates. Brain 2006;129(Pt 3):564–583.
versus 1.5 T: a practical review. Neuroimaging Clin N Am 2006;16:
33. Liu J, Gao XP, Osunde I, et al. Increased regional homogeneity in 285–297.
internet addiction disorder: a resting state functional magnetic reso-
47. Fingelkurts AA, Fingelkurts AA, Kivisaari R, et al. Opioid
nance imaging study. Chin Med J 2010;123:1904–1908.
withdrawal results in an increased local and remote functional connec-
34. Ko CH, Liu GC, Yen JY, Chen CY, Yen CF, Chen CS. Brain correlates tivity at EEG alpha and beta frequency bands. Neurosci Res 2007;58:
of craving for online gaming under cue exposure in subjects with 40–49.

ORIGINAL RESEARCH
Reproducibility Measurement of
Glutathione, GABA, and Glutamate:
Towards In Vivo Neurochemical Profiling
of Multiple Sclerosis With MR
Spectroscopy at 7T
Hetty Prinsen, PhD,1* Robin A. de Graaf, PhD,1,2 Graeme F. Mason, PhD,1,3
Daniel Pelletier, MD,4 and Christoph Juchem, PhD1,4
Purpose: To determine the reproducibility of a comprehensive single-session measurement of glutathione (GSH), c-

aminobutyric acid (GABA), glutamate, and other biochemicals implicated in the pathophysiology of multiple sclerosis
(MS) in the human brain with 1H magnetic resonance spectroscopy (MRS).
Materials and Methods: Five healthy subjects were studied twice in separate 1-hour sessions at 7T. One MS patient
was also scanned once. GSH and GABA were measured with J-difference editing using a semilocalized by adiabatic
selective refocusing sequence (semi-LASER, TE 5 72 msec). A stimulated echo acquisition mode sequence (STEAM,
TE 5 10 msec) was used to detect glutamate along with the overall biochemical profile. Spectra were quantified with
LCModel. Quantification accuracy was assessed through Cramer-Rao lower bounds (CRLB). Reproducibility of the
metabolite quantification was tested using coefficients of variation (CoV).
Results: CRLB were 7% for GSH, GABA, and glutamate and average CoV of 7.8 6 3.2%, 9.5 6 7.0%, and 3.2 6 1.7%
were achieved, respectively. The average test/retest concentration differences at this measurement reproducibility and
quantification accuracy were smaller for GABA and glutamate than intersubject variations in metabolite content with
CoV ratios of 0.6 and 0.8, respectively. As proof of principle, GSH, GABA, and glutamate were also detected in an MS
patient.
Conclusion: GSH, GABA, glutamate, and other metabolites relevant in MS can be quantified at 7T with high accuracy
and reproducibility in a single 1-hour session. This methodology might serve as a clinical research tool to investigate
biochemical markers associated with MS.
M ultiple Sclerosis (MS) is a chronic disorder of the cen-

tral nervous system that leads to demyelination and
neurodegeneration. MS disease-modifying therapies can
progress, the cause of MS and the underlying neurobiologi-
cal mechanisms leading to progressive impairment remain
unclear.
minimize the number of relapses and associated neurological Clinical symptoms may worsen despite stable magnetic
deficits during the predominantly inflammation-mediated resonance imaging (MRI) lesion burden, which suggests that
relapsing-remitting phase (RR-MS).1 To date, however, only processes other than acute inflammation and its immediate
partially effective treatments have been developed to prevent consequences must contribute to cell loss, brain atrophy,
progressive disability and functional decline observed and ongoing clinical decline.2 Magnetic resonance spectros-
throughout the disease course.1 Despite significant research copy (MRS) enables the assessment of pathological changes

*Address reprint requests to: H.P., Department of Radiology and Biomedical Imaging, Yale University School of Medicine, Magnetic Resonance Research
Center (MRRC), the Anlyan Center (TAC), 300 Cedar St., New Haven, CT 06520-8043. E-mail: hetty.prinsen@yale.edu
From the 1Department of Radiology and Biomedical Imaging, Yale University, New Haven, Connecticut, USA; 2Department of Biomedical Engineering,
Yale University, New Haven, Connecticut, USA; 3Department of Psychiatry, Yale University, New Haven, Connecticut, USA; and 4Department of Neurology,
Yale University, New Haven, Connecticut, USA.

V 187

and brain damage noninvasively from the earliest stage of GSH and GABA,22 for instance, limit the detection sensitiv-
the disease.3 Previous MRS studies largely focused on struc- ity; moreover, their resonances are obscured by much stron-
tural and cellular integrity and therefore mainly aimed to ger signals from creatine and macromolecules. Specialized
quantify N-acetylaspartate (NAA),4 choline,5 and myo-inosi- MRS techniques like J-difference editing (JDE) must there-
tol.6 Besides these established biomarkers, there is increasing fore be applied to isolate GSH and GABA.23 JDE is a sub-
evidence that glutathione (GSH) is involved in MS patholo- traction technique that extracts the metabolite of interest
gy. For example, reactive oxygen species (ROS) are generat- from the rest of the spectrum by exploiting a selected intra-
ed by activated macrophages during inflammation and have molecular coupling. Another difficulty is the individual
been linked to damage of myelin, oligodendrocytes, and quantification of overlapping resonances from glutamate
mitochondria.7 GSH is assumed to play a neuroprotective and glutamine. Increasing magnetic field strength improves
role through direct interaction with ROS or as part of enzy- the overall MRS detection sensitivity and enhances spectral
matic redox cycling by which ROS are rendered innocuous.8 dispersion, which allows the separation of glutamate and
Largely based on investigations in vitro and in animal mod- glutamine at 7T.24
els, it has been hypothesized that dysfunction of GSH The aim of this study was to determine the accuracy
metabolism impairs the brain’s autoprotection against oxida- and reproducibility of a comprehensive single-session mea-
tive stress, thereby promoting demyelination, neurotoxicity, surement of GSH, GABA, glutamate, and other biochemi-
and cell death9–12 as a basis for functional decline in MS.13 cals implicated in the pathophysiology of MS in the human
However, to what extent the neurochemistry of GSH is defi- brain with MRS to explore their potential as a clinical
cient in MS remains unknown.
research tool.
GSH is by no means independent from its biochemi-
cal environment. Rates of GSH synthesis depend on the
extracellular glutamate concentrations.14 Glutamate is the
major excitatory neurotransmitter in the brain, a precursor Five healthy subjects (three males and two females, aged 24–40
years, mean age 32 6 8 years) were studied twice in different ses-
of c-aminobutyric acid (GABA), and with cysteine and gly-
sions to assess the measurement accuracy and reproducibility of the
cine one of the three amino acids comprising the GSH tri-
MRS methods. Test–retest sessions were between 1 hour and 1
peptide. Glutamate is closely linked to glutamine by the
month apart. As proof of principle, one RR-MS patient (female,
glutamate-glutamine cycle,15 a process that depends on the 41 years old) was studied once. In addition, a sixth subject (female,
neurotransmission rate.16 As such, the link between GSH 28 years old) was scanned twice in different sessions, but was on a
and other metabolites is neither negligible nor is it purely medication known to impact GABA concentrations and therefore
chemical in nature, but is dynamically influenced by brain not included in the general analyses. The total session duration did
function and neurotransmitter homeostasis. not exceed 1 hour in any case. All participants provided written
Several metabolites involved in GSH metabolism play informed consent, and the measurements were conducted in accor-
a role in MS pathology themselves. Excess extracellular glu- dance with the Institutional Review Board guidelines for research
tamate has been reported to cause calcium-mediated apopto- on human subjects.
sis in an in vitro model of MS and a lack of Experiments were carried out with a 7T head-only MR sys-
oligodendrocytic glutamate transporters has been speculated tem (Agilent, Santa Clara, CA) interfaced to a DirectDrive spec-
as a cause for excitotoxicity.17 Moreover, recent studies have trometer operating at 298.1 MHz for protons. The MR scanner
strongly tied alterations in genes associated with glutamate was equipped with custom-designed actively shielded gradients
metabolism to MS markers of injury.18 The immediate role (Magnex Scientific, Oxford, UK) and operated with Vnmrj 2.3A
of the inhibitory neurotransmitter GABA in MS is still software (Varian, Santa Clara, CA). An 8-channel transmit-receive
unknown, but its anti-inflammatory potential has been sug- array was used for spin handling and signal reception.
T1-weighted anatomical images were obtained with a cus-
gested based on the concurrent observations of increased
tomized inversion-recovery prepared MRI sequence (field of view
GABAergic activity and reduced inflammation in animals
[FOV] 200 3 220 3 78 mm3, matrix size 256 3 256 3 39, reso-
with experimental autoimmune encephalomyelitis (EAE), a
lution 0.78 3 0.86 3 2.00 mm3, inversion time [TI] 1000 msec,
murine model of brain inflammation.11,19
repetition time [TR] 3000 msec, echo time [TE] 6 msec, acquisi-
Despite ongoing efforts,20,21 neither the link between tion time 5 min). The anatomical images were used in the first ses-
GSH and related substances nor the relevance of these bio- sion to position the spectroscopy voxel on the midline occipital
chemicals themselves to MS pathophysiology has been fully cortex (Fig. 1A). The voxel was visually repositioned in the second F1
explored, partly due to constraints to noninvasively measure session at an estimated 1 mm error. Brain extraction of the ana-
GSH metabolism in the human brain in a limited time tomical images was performed using the Brain Extraction Tool
frame. Several important challenges need to be overcome to (BET)27 of the Oxford center for functional MRI of the brain
achieve reliable quantification of GSH and related metabo- (FMRIB) Software Library (FSL) v. 5.0 prior to brain segmenta-
lites by MRS. The relatively low in vivo concentrations of tion. Gray matter, white matter, and cerebrospinal fluid (CSF)

Prinsen et al.: Neurochemical Profiling of MS With MRS
C
O
L
O
R
FIGURE 1: Anatomical image in the axial plane (A) used for both voxel positioning on the midline occipital cortex and for brain
segmentation (B).
segmentation was applied using FMRIB’s Automated Segmentation (NAAG), creatine, phosphocreatine, ascorbate, aspartate, scyllo-
Tool (FAST)28 (Fig. 1B). inositol, and taurine (voxel size 2 3 2 3 2 cm3, TR 3 sec, TE 10
B1 phase shimming was achieved through in-house- msec, mixing time 50 msec, 96 averages, acquisition time 5 min).
developed MR methods and software (IMAGO). The B0 magnetic MM resonances were suppressed by a nonselective inversion pulse
field of the unshimmed brain was derived from five single-echo (TI 320 msec) optimized to minimize the majority of the MM sig-
gradient-echo images (FOV 220 3 220 3 60 mm3, matrix size nals. Water suppression was based on CHESS and outer-volume
126 3 64 3 20, TE 3.8 msec, TR 1.3 sec) at additional TE suppression was used for improved localization specificity.33 An 8-
delays of 0/0.2/0.5/1.5/3 msec. B0 shimming included zero- average water reference was acquired and used for both eddy cur-
through third-order spherical harmonic terms and was calculated rent correction and as internal reference for absolute metabolite
with customized software (B0DETOX)29. quantification.
After shimming, a single-voxel version of the semilocalized Spectra from individual acquisitions and all eight receivers
by adiabatic selective refocusing (semi-LASER) JDE sequence26 were stored separately. Spectral processing was done in MatLab
was applied for GABA editing (voxel size 3 3 3 3 3 cm3, TR 3 (MathWorks, Natick, MA) with a customized software package
sec, TE 72 msec, 128 averages per JDE condition, acquisition time (INSPECTOR). First, eddy current phase correction34 and
13 min) similar to Andreychenko et al.30 A multifrequency pulse sensitivity-weighted summation of the receive channels were
(Gaussian, 8 msec) was used for simultaneous spectral editing at applied. Spectra were phase and frequency aligned individually
1.89 ppm (J-coupled to the 2CH2 GABA resonance at 3.01 ppm) using a least-squares fit and a cross-correlation algorithm, respec-
and water suppression. The frequency position of the noninverted tively. Finally, the phase- and frequency-aligned spectra from indi-
condition was mirrored around water (7.51 ppm). A global vidual acquisitions were averaged. The achieved spectral quality
inversion-preparation was applied to minimize potential macromol- was assessed as full width at half maximum (FWHM) of the NAA
ecule (MM) contributions and to prevent MM coediting due to its resonance at 2.01 ppm in the noninverted JDE condition. With
coupling partner at 1.7 ppm. The optimal TI (300 msec) was STEAM, FWHM of creatine at 3.03 ppm and NAA at 2.01 ppm
determined experimentally and applied for the minimization of sig- were obtained.
nals from the 3.01 ppm MM resonance. Metabolite quantification was achieved with LCModel v.
Subsequently, the same JDE semi-LASER sequence was also 6.3-1K35 and was done relative to unsuppressed water. Basis spec-
used for GSH editing (number of averages 64 per JDE condition, tra were calculated in SpinWizard36 using density matrix simula-
acquisition time 7 min). The editing pulse was applied at 4.56 ppm tions based on the spectral characteristics summarized in
to select the J-coupled CH2 signal from the 2.95 ppm GSH cyste- Govindaraju et al.37 The STEAM basis set included simulated
ine moiety and far away (5 kHz offset) for the nonedited condi- spectra of ascorbate, aspartate, creatine, GABA, glutamate, gluta-
tion. Here, inversion preparation was not necessary due to the lack mine, GSH, myo-inositol, NAAG, NAA, phosphocholine, phos-
of potential MM coediting. Water suppression was achieved with phocreatine, scyllo-inositol, and taurine. Simulated spectra of
CHESS in both the GSH and GABA editing experiments.31 Water GABA, glutamate, glutamine, and NAA were needed for GABA
references were acquired with both GSH and GABA spectra for JDE signal fitting. For the LCModel analysis of GSH JDE signals,
eddy current correction. metabolite basis functions of only GSH and NAA were used
Lastly, the short echo-time stimulated echo acquisition mode because except for GSH and coedited NAA signals, all other peaks
(STEAM) sequence32 was used to detect GSH, GABA, glutamate, are subtracted out. The editing efficiencies of GABA and GSH
glutamine, choline, myo-inositol, NAA, N-acetylaspartylglutamate were determined theoretically for the applied JDE sequence
January 2017 189

above 50% were considered not detected.35 Reproducibility of the

metabolite measurements was evaluated based on coefficients of
variation (CoV) calculated in each subject for each individual
metabolite by dividing the standard deviation of the two measure-
ments by their concentration mean.
Results
The primary target substances GSH, GABA, and glutamate
as well as glutamine, choline, myo-inositol, NAA, NAAG,
creatine, phosphocreatine, ascorbate, aspartate, scyllo-
inositol, and taurine were detectable in all 10 sessions of
this reproducibility study.
Neither baseline nor first-order phase correction was
necessary with any spectrum. Narrow spectral lines with an
average width of 12.2 6 1.2 Hz were found for the
2.01 ppm NAA resonance of the noninverted condition of
the JDE measurement of GSH (Fig. 2A) and 11.7 6 0.9 Hz F2
for the noninverted JDE condition of GABA (Fig. 3A). F3
JDE of GSH allowed the isolation of the CH2 signal from
the cysteine moiety at 2.95 ppm (Fig. 2B). With JDE of
C GABA, the metabolite content was accessible via its
O
L 3.01 ppm CH2 resonance (Fig. 3B). With STEAM, average
O
R
FIGURE 2: JDE of GSH consisting of an edited condition (A,
red) and a noninverted reference (A, black) was performed
using a semi-LASER sequence (voxel size 3 3 3 3 3 cm3, TR 3
sec, TE 72 msec, 64 averages per JDE condition, acquisition
time 7 min). The difference spectrum (B, scaling factor 3.3)
exhibits expected coedited NAA at 2.49 and 2.67 ppm and
allows the isolation of the GSH CH2 signal of the cysteine moie-
ty at 2.95 ppm (dotted vertical line).
compared to a pulse-acquire experiment and were found to be

41.5% and 18.5% (at 10 Hz linewidth), respectively. Metabolite
concentrations were corrected for tissue water and CSF contribu-
tions to the MRS voxel based on brain segmentation results. Tissue
and CSF fractions were confirmed by a spectroscopic T2 measure-
ment in the volume of interest. To this end, a series of fully relaxed
MRS acquisitions at varying TE (10–250 msec) and long TR (15
sec) was acquired to disentangle tissue water from CSF based on dif-
ferences in T2. The T2 relaxation time of CSF had been measured
(190 msec, data not shown) and was included in this calculation as
prior knowledge. Water concentrations of 43,300 mM for gray mat-
ter, 35,880 mM for white matter, and 55,556 mM for CSF were
assumed.35 The water T2 relaxation time correction factor was set to
0.81, based on a TE of 10 msec and a T2 relaxation time of 47 C
msec for water.35,38 In addition, every JDE experiment was accom- O
panied with an explicit efficiency measurement which was executed L
O
to correct the obtained GSH and GABA concentrations. The water R
signal with and without the MEGA editing pulse applied to the FIGURE 3: JDE of GABA consisting of an edited condition (A,
water position was used to confirm the performance of the editing red) and a noninverted reference (A, black) was performed
RF pulse comprising both frequency and power adjustments. using a semi-LASER sequence (voxel size 3 3 3 3 3 cm3, TR 3
time 13 min). The difference spectrum (B, scaling factor 1.1)
Statistical Analysis exhibits expected coedited glutamate and glutamine at
LCModel quantification accuracy was assessed through Cramer- 3.74 ppm and allows the isolation of the GABA 2CH2 signal at
Rao lower bounds (CRLB).39 Metabolites quantified with a CRLB 3.01 ppm (dotted vertical line).

reproducible (Fig. 6, superposition) exhibited by the mean F6

CoV of 3.2 6 1.7% for glutamate (Fig. 7). The mean CoVs F7
of GSH and GABA based on STEAM were 11.6 6 5.1%
and 10.9 6 6.0%, respectively. The difference amplitudes of
the test/retest STEAM spectra were small and resembled the
overall noise floor (Fig. 6, difference), thereby demonstrat-
ing the level of reproducibility. The same holds true for the
JDE measurements, as visible in the minimal difference
between GSH and GABA difference spectra acquired during
different sessions in the same subject (Fig. 8). The high F8
reproducibility of the JDE measurements is exhibited by
mean CoVs of 7.8 6 3.2% and 9.5 6 7.0% for GSH and
GABA, respectively (Fig. 7). The average test/retest concen-
tration differences at this measurement reproducibility and
quantification accuracy were smaller for GABA and gluta-
mate than intersubject variations in metabolite content with
CoV ratios of 0.6 and 0.8, respectively. The obtained con-
C
O
centrations of glutamine, choline, myo-inositol, NAA, crea-
L tine, phosphocreatine, and ascorbate were highly
O reproducible between the first and second measurements
R
FIGURE 4: 1H MRS of target metabolites glutamate and gluta- (Table 1, Fig. 7) with mean CoVs well below 10%. The
mine along with additional metabolites choline, myo-inositol, average metabolite CoVs of NAAG, scyllo-inositol, and tau-
NAA, creatine, ascorbate, aspartate, scyllo-inositol, and taurine rine were between 10 and 20%, and aspartate had a mean
measured with STEAM (A, voxel size 2 3 2 3 2 cm3, TR 3 sec,
TE 10 msec, mixing time 50 msec, 96 averages, acquisition CoV of 21%.
time 5 min). The 4CH2 group of glutamate at 2.34/2.35 ppm Note that Figures 5, 6, and 8 include the results of a
could clearly be separated from both the 4CH2 group of gluta- sixth subject that have not been considered in the CoV anal-
mine at 2.43/2.46 ppm and the 3CH2 group of NAA at
2.49 ppm (B). The overall spectral quality supports the observa-
ysis above. The measurements and metabolite quantification
tion of downfield resonances, for instance, the phenylalanine were also highly reproducible between the two sessions with
peak at 7.37 ppm (C). CoVs of 5.4%, 0.1%, and 0.1% for GSH (JDE), GABA
(JDE), and glutamate, respectively, but the results were not
line widths of 11.8 6 1.5 Hz and 11.2 6 1.7 Hz were found included in the overall CoV analysis, as this subject was on
for the CH3 protons of creatine at 3.03 ppm and NAA at a medication known to impact GABA concentrations.
F4 2.01 ppm, respectively (Fig. 4A). Glutamate could be clearly As proof of principle, GSH, GABA, and glutamate
separated from glutamine and NAA (Fig. 4B). were detected along with the overall biochemical profile in
Metabolite quantification accuracy was high in both the occipital cortex of an MS patient in a similar single, 1-
sessions of the five participants. LCModel fits resulted in hour session (Fig. 9). F9
T1 minimal residuals and small quantification errors (Table 1,
mean CRLB (%) per metabolite). More specifically, CRLB Discussion
of 3% for GSH and 7% for GABA were obtained based Comprehensive single-voxel 1H MRS methods at 7T have
on JDE measurements. STEAM spectra resulted in CRLB been presented for the noninvasive, single-session measure-
of 7% for GSH, 14% for GABA, 2% for glutamate, ment of GSH and other neurochemicals key to MS patholo-
4% for glutamine, and <10% for most additional metab- gy. Single-voxel MRS was used, as B0 and B1 conditions can
olites (<24% in all cases). be highly optimized and reasonably short individual acquisi-
The mean GSH and GABA concentrations over all 10 tion times can be realized, allowing the execution of multi-
measurements of this study were respectively 2.15 6 ple scans in a single session. GSH, GABA, and glutamate
0.16 mM and 1.33 6 0.23 mM obtained through STEAM, were quantified along with the overall biochemical profile in
and 1.34 6 0.13 mM and 1.38 6 0.26 mM obtained the human brain with high accuracy and reproducibility in
through editing. The average glutamate concentration was a single 1-hour session.
10.56 6 0.48 mM. GSH, GABA, and glutamate concentra- Previous JDE of GSH was based on the point resolved
tions per subject and for both sessions are presented in spectroscopy (PRESS) sequence.25 Here, a semi-LASER
F5 Table 1 and Figure 5 (GSH and GABA concentrations sequence was applied for GSH editing, which was not only
based on editing measurements). The test and retest advantageous with respect to signal formation40 but also
STEAM spectra from the same subject were highly enabled excellent spatial definition of the MRS voxel. Note
January 2017 191

192
TABLE 1. Metabolite Concentrations (mM) Measured Twice in Different Sessions (#1 and #2) in Five Healthy Subjects and Mean Quantification Errors (CRLB,
%) Per Metabolite
CRLB
Subject 1 Subject 2 Subject 3 Subject 4 Subject 5
#1 #2 #1 #2 #1 #2 #1 #2 #1 #2 Mean
JDE GSH 2.32 2.08 2.09 1.97 1.99 2.17 2.41 2.02 2.36 2.09 3.0 Target
GABA 1.08 1.14 1.62 1.68 1.60 1.30 1.26 1.11 1.09 1.41 4.3
STEAM GSH 1.43 1.30 1.38 1.28 1.50 1.18 1.42 1.17 1.53 1.24 5.7
GABA 1.54 1.31 1.53 1.56 1.03 0.90 1.75 1.39 1.55 1.25 9.3
Glu 10.94 11.26 10.98 10.17 9.78 10.15 10.55 10.34 11.05 10.37 1.6
Gln 3.45 3.57 3.26 3.38 4.20 3.91 4.24 3.96 4.13 3.94 3.1
Cho 1.56 1.47 1.24 1.22 1.22 1.27 1.34 1.23 1.46 1.35 2.0 Established
mIns 8.77 8.28 7.70 7.48 7.32 7.31 6.56 6.40 7.93 7.23 1.7
NAA 11.84 10.92 12.40 12.17 10.68 10.85 11.10 10.85 12.74 11.62 1.0
NAAG 0.79 0.99 0.70 1.00 0.97 0.91 0.99 0.97 1.00 0.85 8.7
tNAA 12.64 11.91 13.10 13.17 11.65 11.75 12.09 11.82 13.73 12.46 1.0
Cr 5.73 5.74 6.13 5.78 5.40 5.71 5.19 5.83 6.06 5.68 3.0 Additional
PCr 4.19 4.01 4.11 4.02 3.45 3.35 4.01 3.46 4.13 3.77 4.1
tCr 9.92 9.75 10.24 9.80 8.85 9.06 9.20 9.29 10.19 9.46 1.0
Asc 2.52 2.02 1.92 1.96 1.92 1.81 1.93 2.29 2.24 2.32 7.7
Asp 2.49 2.30 1.10 1.89 0.97 0.95 1.94 1.14 2.19 1.62 14.6
sIns 0.30 0.31 0.26 0.19 0.16 0.19 0.25 0.28 0.31 0.22 12.7
Tau 2.32 2.02 1.99 1.52 1.93 1.77 2.37 2.42 2.94 2.09 4.8
Stage:

Volume 45, No. 1
Page: 192
T1 relaxation times, precluding the perfect cancellation of

all MM resonances. A TI of 320 msec optimized to mini-
mize the majority of the MM signals was used here.
C
O
L
O
R
FIGURE 5: GSH, GABA, and glutamate concentrations (mM) in
the occipital cortex of five healthy subjects studied twice in dif-
ferent sessions. Metabolite concentrations were also measured
twice in a subject on medication known to impact GABA con-
centrations. GSH and GABA were measured using semi-LASER
JDE, and STEAM was used to detect glutamate.
that the application of the editing scheme as proposed by

Andreychenko et al30 allowed 100% editing efficiency over
the entire echo time. The functionality of the editing pulses
themselves was confirmed experimentally in every subject by
application of the MEGA scheme for water suppression
(instead of JDE).
In contrast to GSH editing, JDE of GABA is vulnera-
ble to MM coediting. One way to handle potential MM
contamination is by applying the editing pulse around the
1.7 ppm MM resonance in the two steps of the editing
scheme (1.89 and 1.51 ppm) introduced by Henry et al.41
This subtraction, however, results in complete elimination of
the MM signal only under the assumption that both pulses
are perfectly positioned and the 1.7 ppm MM resonance is
fully symmetric. Potential system drifts therefore lead to sub-
optimal MM subtraction and a systematic over- or underes-
timation of GABA levels. Field variations due to the
respiratory cycle pose a second stochastic challenge to the
MM subtraction scheme. One way to avoid the coedited
MM signal is by applying an inversion-recovery preparation
as used by Behar et al.42 They demonstrated that MM sig-
nals can be fully suppressed by an inversion-recovery prepa-
ration if the delay between the inversion pulse and the MRS
sequence itself is chosen to minimize (“null”) the specific
MM at hand. Note that perfect signal elimination can be
achieved for a single MM resonance with inversion prepara-
tion and that an experimentally determined optimal TI of
300 msec was applied here for the 3 ppm MM resonance.
The resultant additional 50% reduction in GABA signal is
C
substantial but has been accepted in this work to ensure reli- O
able results. L
O
MM signals are also a problem for short TE STEAM, R
as they are known to cause a wavy baseline and to thereby FIGURE 6: Superposition of test–retest STEAM spectra (voxel
substantially complicate metabolite quantification.42,43 The size 2 3 2 3 2 cm3, TR 3 sec, TE 10 msec, mixing time 50
STEAM acquisition was therefore preceded by an inversion- msec, 96 averages, acquisition time 5 min) acquired during the
two study sessions (#1 and #2, left) and their difference (right).
recovery preparation to minimize MM contributions. This Extra: Corresponding spectra of a subject on medication
technique is, however, limited by the MM-specific spread of known to impact GABA concentrations.
January 2017 193

FIGURE 7: Mean CoV (%) of the target metabolites (GSH,

GABA, glutamate, and glutamine), the established MS bio-
markers (choline, myo-inositol, N-acetylaspartate/N-acetylaspar-
tylglutamate), and additional metabolites (creatine,
phosphocreatine, ascorbate, aspartate, scyllo-inositol, and tau-
rine) quantified twice during different sessions in five healthy
subjects. Error bars represent standard errors of the mean.
GSH and GABA were measured using semi-LASER JDE, and
STEAM was used to detect glutamate, glutamine, the estab-
lished MS biomarkers, and additional metabolites.
Dedicated measurements using highly optimized MRS

methods were performed for each of the metabolites of pri-
mary interest in clinical MS research, namely, GSH, GABA,
and glutamate. High-quality spectra characterized by a flat
baseline, narrow spectral lines, excellent water suppression,
and optimal MM suppression were obtained. With
STEAM, a clear separation of the glutamate and glutamine
(plus NAA) resonances at 2.35 ppm and 2.45 ppm, respec-
tively, was achieved. While not the focus of this research,
the overall spectral quality also supported the observation of
downfield resonances.44
CRLB were calculated to assess the quantification
accuracy of the presented methods. The mean CRLB was
low for the target metabolites GSH (JDE), GABA (JDE),
and glutamate, indicating high quantification accuracy.
Superposition of the spectra acquired in the same subject in
different sessions showed virtually perfect overlap, which is a
C
direct measure of high reproducibility. The reproducibility O
of the methods was confirmed by low mean CoV values for L
GSH (JDE), GABA (JDE), and for glutamate. The consis- O
R
tently elevated GABA levels obtained from the medicated FIGURE 8: Superposition of test–retest difference spectra for
subject underline the reproducibility of the measurements JDE of GSH acquired using a semi-LASER sequence (voxel size
and the sensitivity of the methods to detect abnormal 3 3 3 3 3 cm3, TR 3 sec, TE 72 msec, 64 averages per JDE con-
dition, acquisition time 7 min) during the two study sessions
metabolite levels. As proof of principle, GSH, GABA, and (#1 and #2, left) and their difference (right). Extra: Superposi-
glutamate were also detected in an MS patient. The pre- tion of test–retest difference spectra for JDE of GABA acquired
sented study design allowed the reliable quantification of using a semi-LASER sequence (voxel size 3 3 3 3 3 cm3, TR 3
the primary target metabolites, other established biomarkers
time 13 min) during the two study sessions (#1 and #2, left)
of MS including choline, myo-inositol, and NAA, and addi- and their difference (right) of a subject on medication known
tional metabolites of the biochemical profile including to impact GABA concentrations.

used a short TE STEAM sequence in the posterior cingulate

cortex.
Multiple studies presented GABA concentrations mea-
sured in the human occipital cortex at 7T previously with a
concentration range of 0.2 to 2 mM.30,46,50,52,53 The
GABA concentration obtained through JDE reported in the
present study is within this range. The reproducibility of
GABA quantification in different areas of the human brain
at 7T by means of CoV values has been assessed previous-
ly.50,51,54,55 The reproducibility of the GABA JDE measure-
ment in the current study was better than in the studies of
Terpstra et al,50 van de Bank et al,51 and Stephenson et al54
(mean CoV range of 16–22%), which used a STEAM or
semi-LASER sequence. In the study by Wijtenburg et al,55
the reproducibility of GABA quantification was addressed in
the anterior cingulate (AC) and dorsolateral prefrontal cor-
tex (DLPFC) for both short TE STEAM and Mescher-
FIGURE 9: Proof of principle: GSH (A), GABA (B), and gluta- Garwood PRESS (MEGA-PRESS) editing. In the editing
mate (C, Glu) detection in an MS patient. experiment, MM minimization was achieved by applying
the editing pulse around the 1.7 ppm MM resonance. The
creatine, phosphocreatine, and ascorbate were obtained in a mean CoV for GABA was lower in the present study than
single 1-hour session. that reported by Wijtenburg et al55 using the editing tech-
Comparison of GSH concentrations, quantification nique (13.6% in the AC and 13.4% in the DLPFC) and
accuracy, and reproducibility between STEAM and JDE using STEAM in the DLPFC (16.2%), but it was higher
measurements in the current study suggests the advantages than the mean CoV using STEAM in the AC (3.5%). The
of an editing over a nonediting technique. STEAM spectra GABA quantification accuracy expressed as CRLB was,
at short TE suffer from significant spectral overlap with however, better in the current study compared to that of the
MM signals, which limits the achievable quantification anal- STEAM experiments in the study by Wijtenburg et al55
ysis and accuracy. In addition, the GSH signal can acciden- (CRLB 10.5–11.0% in the DLPFC and 8.3–9.3% in the
tally be confounded with other, overlapping resonances like AC). Wijtenburg et al55 quantified GABA by peak integra-
creatine or GABA, leading to an over- or underestimation tion, and no quantification errors were reported.
of the GSH concentration. JDE of GSH reduces the num- Different sequences, including STEAM, PRESS,
ber of spectral components to GSH and NAA only and (semi-)LASER and SPECIAL,56 have been used to detect
simplifies the quantification due to the absence of spectral glutamate in the human occipital cortex at 7T. The concen-
overlap. In the present study the average GSH concentration tration of glutamate reported in the present study is within
obtained through STEAM was lower than that obtained the concentration range as reported in these studies (8.6–
through JDE. The mean CRLB of GSH quantified based 11.3 mM).38,46,57,58 The reproducibility of glutamate quan-
on editing was lower than that based on STEAM; the mean tification at 7T by means of CoV values has been explored
CoV was also lower. Additionally, the mean CRLB and previously in different areas of the human brain.49–51,54 The
CoV were lower for JDE of GABA than for GABA reproducibility of the glutamate measurement is better in
obtained through STEAM. These results advocate for the the present study than that of Stephenson et al54 (mean
measurement of GSH and GABA individually using JDE CoV of 8% in both the anterior cingulate cortex and the
instead of as part of, for example, a STEAM acquisition. insula), who also used a STEAM sequence. The mean CoV
Reported GSH concentrations determined in the for glutamate is also lower in the current study than in the
human occipital cortex at 7T range from 0.5– study of Lally et al49 (6.5–8.0% in the prefrontal cortex),
2.3 mM.38,45–47 The GSH concentration obtained through who used a TE optimized PRESS sequence in combination
JDE presented in the current study falls into the upper end with a J-suppression pulse to attenuate NAA at 2.49 ppm.
of this range. The reproducibility of GSH quantification at The study of van de Bank et al,51 which used a semi-
7T by means of CoV values has been reported previously in LASER sequence, and the present study show comparable
different areas of the human brain.48–51 These studies reproducibility of glutamate quantification (mean CoV
reported mean CoVs ranging from 5 to 14.4%. The mean 3.1% in the posterior cingulate cortex and 3.4% in the
CoV for JDE of GSH in the present study was slightly corona radiata). The reproducibility of the glutamate in the
higher than the 5% reported by Deelchand et al,48 who study of Terpstra et al,50 also using a semi-LASER sequence,
January 2017 195

is even better, with a mean CoV in the posterior cingulate considered small. Another reason for potential test/retest
of 2.5% (presented as bar graphs) and 3% in the variation might have been subject movement, including
cerebellum. both breathing and voluntary movements, throughout either
CRLB describe the lowest possible standard deviation the test and/or the retest session. In addition, physiological
of the spectral decomposition; in other words, it is a mea- variations like circadian rhythm, exercise level, or food
sure of maximum confidence in the numerical procedure at intake cannot be excluded. The high level of reproducibility
hand. The CoV summarizes the reproducibility of metabo- achieved in this study suggests, however, that the cumulative
lite concentrations irrespective of the way they were derived. impact of the above effects was small.
Both parameters provide means for the identification of lim- In conclusion, this study shows that GSH, GABA, glu-
ited data quality if CRLB and CoV are large. They do not tamate, and other MS-related metabolites can be quantified
represent unique measures of data quality and metabolite at 7T with high accuracy and reproducibility in a single 1-
quantification, however, as they lack the sensitivity to detect hour session. This brain metabolomics-type approach allows
systematic but reproducible errors. Such errors not only the comprehensive characterization of specific metabolites in
include extreme scenarios like erroneous voxel location (eg, vivo, and metabolic partners, implicated in MS pathogene-
left versus right coordinate inversion), but also more subtle sis. The methodology is expected to serve as a clinical
effects like coedited MM signals in GABA JDE experiments research tool to target the biochemistry of oxidative stress,
or the misinterpretation of spectral signals in the quantifica- tissue injury, and repair in neurodegenerative diseases such
tion process, eg, in case of short-TE STEAM, where GSH as MS.
and GABA are largely overlapping with much stronger sig-
nals from creatine and MM. MM concentrations are poten-
tially elevated in MS59 and thus the risk of erroneous
metabolite quantification is further enhanced for GABA Acknowledgment
JDE and STEAM investigations of MS pathology. There is Contract grant sponsor: National Multiple Sclerosis Society;
arguably no ground truth available for in vivo MRS, and contract grant numbers: RG 4319 and PP 3356; Contract
the positive proof of accurate metabolite quantification grant sponsor: Nancy Davis Foundation; Contract grant
remains inherently impossible. We therefore applied the best sponsor: National Institutes of Health (NIH); contract
available MRS methods to detect the key metabolites of MS grant numbers: UL1 TR000142, R01 NS062885, and P30
pathology and optimized them to minimize the risk of sys- NS052519.
tematic errors. The use of inversion-recovery preparation for We thank all the subjects who participated in this study.
the minimization of MM signals came at the cost of a sub-
stantial signal-to-noise ratio reduction. This choice was con-
sidered justified in this study, however, to achieve the most
reliable metabolite quantification and to thereby set the References
1. Goldenberg MM. Multiple sclerosis review. P T 2012;37:175–184.
stage for future clinical MS research.
Next to GSH, GABA, and glutamate, the concentra- 2. Lublin FD, Reingold SC, Cohen JA, et al. Defining the clinical course
of multiple sclerosis: the 2013 revisions. Neurology 2014;83:278–286.
tions of glutamine, choline, myo-inositol, NAA, NAAG,
3. De Stefano N, Filippi M. MR spectroscopy in multiple sclerosis.
creatine, phosphocreatine, aspartate, scyllo-inositol, and tau- J Neuroimaging 2007;17(Suppl 1):31S–35S.
rine were within previously reported ranges.22 The concen- 4. Arnold DL, Matthews PM, Francis G, Antel J. Proton magnetic reso-
tration of ascorbate was slightly higher in the present study nance spectroscopy of human brain in vivo in the evaluation of multi-
compared to literature values (0.5–1.5 mM).22 Differences ple sclerosis: assessment of the load of disease. Magn Reson Med
1990;14:154–159.
in metabolite concentrations between the current study and
5. Davie CA, Hawkins CP, Barker GJ, et al. Serial proton magnetic reso-
previous reports could be the result of variation of TE or nance spectroscopy in acute multiple sclerosis lesions. Brain 1994;
the application of T2 correction schemes. 117(Pt 1):49–58.
The study involved five healthy volunteers and one 6. Helms G, Stawiarz L, Kivisakk P, Link H. Regression analysis of metab-
MS patient to demonstrate the feasibility of neurochemical olite concentrations estimated from localized proton MR spectra of
active and chronic multiple sclerosis lesions. Magn Reson Med 2000;
profiling in a 1-hour experiment and to assess the achievable 43:102–110.
reproducibility. Neither did the study aim to compare 7. Paling D, Golay X, Wheeler-Kingshott C, Kapoor R, Miller D. Energy
metabolite concentrations between healthy subjects and MS failure in multiple sclerosis and its investigation using MR techniques.
J Neurol 2011;258:2113–2127.
patients, nor were the group sizes suitable for such analysis.
Voxel repositioning between test/retest sessions was achieved 8. Lu SC. Glutathione synthesis. Biochim Biophys Acta 2013;1830:3143–
3153.
at an estimated error of less than 1 mm in all three direc-
9. Calabrese V, Raffaele R, Cosentino E, Rizza V. Changes in cerebrospi-
tions. While a minute impact on the apparent metabolite nal fluid levels of malondialdehyde and glutathione reductase activity
content cannot be excluded, these imperfections are in multiple sclerosis. Int J Clin Pharmacol Res 1994;14:119–123.

10. Ljubisavljevic S, Stojanovic I, Pavlovic D, Sokolovic D, Stevanovic I. 32. Frahm J, Bruhn H, Gyngell ML, Merboldt KD, Hanicke W, Sauter R.
Aminoguanidine and N-acetyl-cysteine supress oxidative and nitrosa- Localized high-resolution proton NMR spectroscopy using stimulated
tive stress in EAE rat brains. Redox Rep 2011;16:166–172. echoes: initial applications to human brain in vivo. Magn Reson Med
1989;9:79–93.
11. Mohamed A, Shoker A, Bendjelloul F, et al. Improvement of experi-
mental allergic encephalomyelitis (EAE) by thymoquinone; an oxida- 33. Juchem C, Logothetis NK, Pfeuffer J. 1H-MRS of the macaque mon-
tive stress inhibitor. Biomed Sci Instrum 2003;39:440–445. key primary visual cortex at 7 T: strategies and pitfalls of shimming at
the brain surface. Magn Reson Imaging 2007;25:902–912.
12. Hayes JD, Strange RC. Potential contribution of the glutathione S-
transferase supergene family to resistance to oxidative stress. Free 34. Klose U. In vivo proton spectroscopy in presence of eddy currents.
Radic Res 1995;22:193–207. Magn Reson Med 1990;14:26–30.
13. Mann CL, Davies MB, Boggild MD, et al. Glutathione S-transferase 35. Provencher SW. Estimation of metabolite concentrations from local-
polymorphisms in MS: their relationship to disability. Neurology 2000; ized in vivo proton NMR spectra. Magn Reson Med 1993;30:672–679.
54:552–557.
36. de Graaf RA, Prinsen H, Giannini C, Caprio S, Herzog RI. Quantifica-
14. Frade J, Pope S, Schmidt M, et al. Glutamate induces release of glu- tion of H NMR Spectra from Human Plasma. Metabolomics 2015;11:
tathione from cultured rat astrocytes—a possible neuroprotective 1702–1707.
mechanism? J Neurochem 2008;105:1144–1152.
37. Govindaraju V, Young K, Maudsley AA. Proton NMR chemical shifts
15. Erecinska M, Silver IA. Metabolism and role of glutamate in mammali- and coupling constants for brain metabolites. NMR Biomed 2000;13:
an brain. Prog Neurobiol 1990;35:245–296. 129–153.
16. Rothman DL, De Feyter HM, de Graaf RA, Mason GF, Behar KL. 13C 38. Marjanska M, Auerbach EJ, Valabregue R, Van de Moortele PF,
MRS studies of neuroenergetics and neurotransmitter cycling in Adriany G, Garwood M. Localized 1H NMR spectroscopy in differ-
humans. NMR Biomed 2011;24:943–957. ent regions of human brain in vivo at 7 T: T2 relaxation times and
concentrations of cerebral metabolites. NMR Biomed 2012;25:332–
17. Pitt D, Nagelmeier IE, Wilson HC, Raine CS. Glutamate uptake by oli-
339.
godendrocytes: Implications for excitotoxicity in multiple sclerosis.
Neurology 2003;61:1113–1120. 39. Cavassila S, Deval S, Huegen C, van Ormondt D, Graveron-Demilly D.
Cramer-Rao bound expressions for parametric estimation of overlap-
18. Baranzini SE, Srinivasan R, Khankhanian P, et al. Genetic variation
ping peaks: influence of prior knowledge. J Magn Reson 2000;143:
influences glutamate concentrations in brains of patients with multiple
311–320.
sclerosis. Brain 2010;133:2603–2611.
40. Kaiser LG, Young K, Matson GB. Numerical simulations of localized
19. Bhat R, Axtell R, Mitra A, et al. Inhibitory role for GABA in autoim-
high field 1H MR spectroscopy. J Magn Reson 2008;195:67–75.
mune inflammation. Proc Natl Acad Sci U S A 2010;107:2580–2585.
41. Henry PG, Dautry C, Hantraye P, Bloch G. Brain GABA editing without
20. Choi IY, Lee P. Doubly selective multiple quantum chemical shift
macromolecule contamination. Magn Reson Med 2001;45:517–520.
imaging and T(1) relaxation time measurement of glutathione (GSH)
in the human brain in vivo. NMR Biomed 2013;26:28–34. 42. Behar KL, Rothman DL, Spencer DD, Petroff OA. Analysis of macro-
molecule resonances in 1H NMR spectra of human brain. Magn Reson
21. Srinivasan R, Ratiney H, Hammond Rosenbluth K, Pelletier D, Nelson
Med 1994;32:294–302.
S. MR spectroscopic imaging of glutathione in the white and gray
matter at 7 T with an application to multiple sclerosis. Magn Reson 43. de Graaf RA, Brown PB, McIntyre S, Nixon TW, Behar KL,
Imaging 2010;28:163–170. Rothman DL. High magnetic field water and metabolite proton T1
and T2 relaxation in rat brain in vivo. Magn Reson Med 2006;56:386–
22. de Graaf RA. In vivo NMR spectroscopy: principles and techniques.
394.
London: John Wiley & Sons; 2008.
44. Kreis R, Pietz J, Penzien J, Herschkowitz N, Boesch C. Identification
23. Rothman DL, Petroff OA, Behar KL, Mattson RH. Localized 1H NMR
and quantitation of phenylalanine in the brain of patients with phenyl-
measurements of gamma-aminobutyric acid in human brain in vivo.
ketonuria by means of localized in vivo 1H magnetic-resonance spec-
Proc Natl Acad Sci U S A 1993;90:5662–5666.
troscopy. J Magn Reson B 1995;107:242–251.
24. Tkac I, Andersen P, Adriany G, Merkle H, Ugurbil K, Gruetter R. In
45. Emir UE, Auerbach EJ, Van De Moortele PF, et al. Regional neuro-
vivo 1H NMR spectroscopy of the human brain at 7 T. Magn Reson
chemical profiles in the human brain measured by (1)H MRS at 7 T
Med 2001;46:451–456.
using local B(1) shimming. NMR Biomed 2012;25:152–160.
25. Bottomley PA. Selective volume method for performing localized
46. Lin Y, Stephenson MC, Xin L, Napolitano A, Morris PG. Investigating
NMR spectroscopy. US Patent 4 1984:480/228.
the metabolic changes due to visual stimulation using functional pro-
26. Scheenen TW, Klomp DW, Wijnen JP, Heerschap A. Short echo time ton magnetic resonance spectroscopy at 7 T. J Cereb Blood Flow
1H-MRSI of the human brain at 3T with minimal chemical shift dis- Metab 2012;32:1484–1495.
placement errors using adiabatic refocusing pulses. Magn Reson Med
47. Terpstra M, Ugurbil K, Tkac I. Noninvasive quantification of human
2008;59:1–6.
brain ascorbate concentration using 1H NMR spectroscopy at 7 T.
27. Smith SM. Fast robust automated brain extraction. Hum Brain Mapp NMR Biomed 2010;23:227–232.
2002;17:143–155.
48. Deelchand DK, Marjanska M, Hodges JS, Terpstra M. Sensitivity and
28. Zhang Y, Brady M, Smith S. Segmentation of brain MR specificity of human brain glutathione concentrations measured using
images through a hidden Markov random field model and the short-TE H MRS at 7 T. NMR Biomed 2016;29:600–606.
expectation-maximization algorithm. IEEE Trans Med Imaging 2001;
49. Lally N, An L, Banerjee D, et al. Reliability of 7T (1) H-MRS measured
20:45–57.
human prefrontal cortex glutamate, glutamine, and glutathione sig-
29. Juchem C, Umesh Rudrapatna S, Nixon TW, de Graaf RA. nals using an adapted echo time optimized PRESS sequence: A
Dynamicmulti-coil technique (DYNAMITE) shimming for echo- between- and within-sessions investigation. J Magn Reson Imaging
planar imaging of the human brain at 7 Tesla. Neuroimage 2015; 2016;43:88–98.
105:462–472.
50. Terpstra M, Cheong I, Lyu T, et al. Test-retest reproducibility of neu-
30. Andreychenko A, Boer VO, Arteaga de Castro CS, Luijten PR, Klomp rochemical profiles with short-echo, single-voxel MR spectroscopy at
DW. Efficient spectral editing at 7 T: GABA detection with MEGA- 3T and 7T. Magn Reson Med 2015 [Epub ahead of print].
sLASER. Magn Reson Med 2012;68:1018–1025.
51. van de Bank BL, Emir UE, Boer VO, et al. Multi-center reproducibility
31. Haase A, Frahm J, Hanicke W, Matthaei D. 1H NMR chemical shift of neurochemical profiles in the human brain at 7 T. NMR Biomed
selective (CHESS) imaging. Phys Med Biol 1985;30:341–344. 2015;28:306–316.
January 2017 197

52. Marsman A, Mandl RC, Klomp DW, et al. GABA and glutamate in 56. Mlynarik V, Gambarota G, Frenkel H, Gruetter R. Localized short-
schizophrenia: a 7 T (1)H-MRS study. Neuroimage Clin 2014;6:398–407. echo-time proton MR spectroscopy with full signal-intensity acquisi-
tion. Magn Reson Med 2006;56:965–970.
53. Schaller B, Mekle R, Xin L, Kunz N, Gruetter R. Net increase of lactate
and glutamate concentration in activated human visual cortex 57. Bednarik P, Tkac I, Giove F, et al. Neurochemical and BOLD
detected with magnetic resonance spectroscopy at 7 Tesla. responses during neuronal activation measured in the human visual
J Neurosci Res 2013;91:1076–1083. cortex at 7 Tesla. J Cereb Blood Flow Metab 2015;35:601–610.
54. Stephenson MC, Gunner F, Napolitano A, et al. Applications of multi- 58. Mekle R, Mlynarik V, Gambarota G, Hergt M, Krueger G, Gruetter R.
nuclear magnetic resonance spectroscopy at 7T. World J Radiol 2011; MR spectroscopy of the human brain with enhanced signal intensity
3:105–113. at ultrashort echo times on a clinical platform at 3T and 7T. Magn
Reson Med 2009;61:1279–1285.
55. Wijtenburg SA, Rowland LM, Edden RA, Barker PB. Reproducibility
of brain spectroscopy at 7T using conventional localization and 59. Mader I, Seeger U, Weissert R, et al. Proton MR spectroscopy with
spectral editing techniques. J Magn Reson Imaging 2013;38: metabolite-nulling reveals elevated macromolecules in acute multiple
460–467. sclerosis. Brain 2001;124:953–961.

ORIGINAL RESEARCH
Differential Diagnosis of Mitochondrial

Encephalopathy With Lactic Acidosis and
Stroke-Like Episodes (MELAS) and
Ischemic Stroke Using 3D
Pseudocontinuous Arterial Spin Labeling
Rui Li, MD,1,2 Hua-feng Xiao, MD, PhD,1,3 Jin-hao Lyu, MD,1
Danny J.J. Wang, PhD,4 Lin Ma, MD, PhD,1* and Xin Lou, MD, PhD1,4*
Purpose: To evaluate the efficacy of 3D pseudocontinuous arterial spin labeling (3D pCASL) in the differential diagnosis
between mitochondrial encephalopathy with lactic acidosis and stroke-like episodes (MELAS) and acute ischemic stroke (AIS).
Materials and Methods: Conventional magnetic resonance imaging (MRI) including T2-weighted imaging (T2WI) and
diffusion-weighted imaging (DWI), and 3D pCASL perfusion data were obtained on a 3.0T MR scanner in 16 newly
appearing lesions in nine patients with MELAS and 14 acute lesions in 12 patients with AIS. A postlabeling delay (PLD)
time of 2025 msec was applied. The cerebral blood flow (CBF) values were measured in the central part and the periph-
eral part of the lesions and the CBF values were compared between MELAS and AIS patients.
Results: The lesions of both MELAS and AIS showed high signal intensity on T2WI and demonstrated hyperintensity on
DWI. Compared with the perfusion defects or hypoperfusion in all AIS, hyperperfusion was revealed in 16 acute MELAS
lesions, especially in the peripheral part of the lesions. The CBF values of 16 lesions in MELAS were 11.20–73.11 ml/
100g/min in the central part and 65.33–169.11 ml/100g/min in the peripheral part. The CBF values of 14 lesions in AIS
were 12.32–19.94 ml/100g/min in the central part and 11.66–18.37 ml/100g/min in the peripheral part. The CBF value of
the peripheral part (119.80 6 35.41) in MELAS was significantly higher than that (14.66 6 2.61) in AIS (P 5 0.0001).
Conclusion: The whole-brain 3D pCASL technique might be useful in differentiating MELAS from AIS.
in about 90% of MELAS individuals.2 The classic MELAS

M itochondrial encephalopathy is a multisystem disorder
caused by defects or dysfunction of mitochondrial
enzymes. MELAS (mitochondrial encephalopathy with lactic
triad includes stroke-like episodes, lactic acidosis, and seizures,
which partly overlaps with ischemic stroke in children and
acidosis and stroke-like episodes) is one of the most domi- young adults.5–7 The incidence rate of ischemic stroke in young
nant subtypes,1 which involves multiple systems with vari- adults has increased recently, and approximately 10–14% of all
able clinical features and recurrent episodes.2 ischemic stroke occur in young adults aged 18–45 years.8,9
The peak onset age for patients with MELAS is in child- The imaging appearances of MELAS and acute ische-
hood and early adulthood. Only 5–8% patients present before mic stroke (AIS) in childhood and early adulthood can be
the age of 2 years and 1–6% after the age of 40 years.2,3 Mul- similar on conventional magnetic resonance imaging (MRI),
tiple organs could be involved in MELAS patients, with a both presenting hyperintensity on T2-weighted images
diversity of clinical manifestations.3,4 Stroke-like episode (T2WI) involving multifocal cortical and subcortical white
symptoms are one of the predominant presentations that occur matter areas, and both appearing as hyperintensity on
Received Feb 16, 2016, Accepted for publication Jun 9, 2016.
*Address reprint requests to: X.L. or L.M., Department of Radiology, Chinese PLA (People’s Liberation Army) General Hospital, 28 Fuxing Road, Beijing
100853, China. E-mail: louxin301@gmail.com or cjr.malin@vip.163.com
From the 1Department of Radiology, Chinese PLA General Hospital, Beijing, China; 2School of Medicine, Nankai University, Tianjin, China; 3Department of
Radiology, Chinese PLA 302 Hospital, Beijing, China; and 4Department of Neurology, University of California, Los Angeles, California, USA.

V 199

and within 1 month after onset in all MELAS patients. Follow-up

3D pCASL examinations were obtained in one MELAS patient. The
demographic information and lactate levels in plasma and/or cerebral
spinal fluid were acquired for all MELAS cases.
The diagnostic flowchart was: when emergency clinical mani-
festations (eg, stroke-like episodes, seizure, etc.) occurred in a
young patient. The routine neuroimaging including computed
tomography (CT), conventional MRI, and DWI were initially
applied and other diseases including tumor, infection, hemorrhage,
etc. were excluded. When hyperintensity on DWI was present, AIS
or other diseases including MELAS should be considered and MR
perfusion imaging (3D pCASL) is used to determine whether it is
caused by ischemic stroke or MELAS. If hypoperfusion is
observed, ischemic stroke is tentatively diagnosed, which is finally
diagnosed by using vascular imaging (digital subtraction angiogra-
phy and/or MR angiography) and clinical follow-up. On the con-
trary, if hyperperfusion is found, MELAS is highly suspected. The
FIGURE 1: The diagnostic flowchart for patients with recent
onset of stroke-like episodes. a: Abnormal high signal involved final diagnosis of MELAS is based on muscle biopsy and/or a
cortex and/or subcortical white matter. b: Vascular imaging genetic test. For all suspected MELAS patients, lactate level in
(MRA or DSA) should be performed. c: Lactate levels in plasma plasma and/or cerebrospinal fluid is also examined. The stepwise
and/or cerebral spinal fluid. diagnostic flowchart is summarized in Figure 1. F1
The patient information was anonymized at image analyses
diffusion-weighted imaging (DWI) during a recent onset of to protect the subject ’s privacy.
stroke-like episodes.
The overlaps of clinical neurological features and neu- MRI and Data Processing
roimaging findings (including routine MRI and DWI) All MRI data were obtained on a 3.0T MR scanner (Discovery
make the differential diagnosis difficult between MELAS 750, GE Healthcare, Milwaukee, WI) with a 32-channel head coil.
and AIS in childhood and early adulthood, yet the differen- Vacuum cushions were applied to reduce the noise and movement
artifact.
tiation between the two conditions is crucial for the choice
Conventional MR sequences included axial T1-weighted
of therapeutic strategies. For patients with AIS, thrombolytic
image (T1WI), axial T2WI, coronal T2WI, and DWI. The T1WI
therapy should be given, whereas individuals with MELAS
sequence was performed using T1-FLAIR (fluid attenuated inver-
should be treated by improving mitochondria function, sion recovery) (repetition time / echo time / inversion time [TR/
ameliorating stroke-like episode symptoms, and reducing TE/TI] 5 1750/24/780 msec, slice thickness 5 5.0 mm, slice
the frequency and severity of these episodes.2 gap 5 1.5 mm, field of view [FOV] 5 24 3 24 cm, matrix 5 320
Recently, 3D pseudocontinuous arterial spin labeling 3 320, NEX 5 1). Axial T2WI was performed using a fast spin
(3D pCASL), as a novel noninvasive quantitative evaluation echo (FSE) sequence (TR/TE 5 4252/103.7 msec, slice thick-
of brain perfusion using labeled blood as an endogenous ness 5 5.0 mm, slice gap 5 1.5 mm, echo train length 5 32,
tracer without exogenous injection of contrast agent, has FOV 5 24 3 24 cm, matrix 5 192 3 192, NEX 5 1.5). Coronal
aroused extensive concerns in the diagnosis and evaluation T2WI were performed using T2-FLAIR (TR/TE/TI 5 8500/163/
2100 msec, slice thickness 5 5.0 mm, slice spacing 5 1.5 mm,
of stroke and other brain disorders.10–14
FOV 5 24 3 24 cm, matrix 5 288 3 224, NEX 5 1). DWI was
The purpose of this study was to quantitatively evalu-
performed using fat-suppressed single-shot spin echo echo-planar
ate the efficacy of 3D pCASL in differentiating MELAS
imaging (TR/TE 5 6000/65.7 msec, slice thickness 5 5.0 mm, slice
from AIS. gap 5 1.5 mm, FOV 5 24 3 24 cm, matrix 5 192 3 192,
NEX 5 2) with b 5 1000 s/mm2 applied in the x, y, and z direc-
Materials and Methods tions, and b 5 0 s/mm2 without motion-probing gradients, fol-
Subjects lowed by automatic generation of isotropic DWI.
Twenty-seven consecutive patients with recent onset of stroke-like epi- Perfusion-weighted images were obtained using a 3D pCASL
sodes were retrospectively analyzed from February 2013 to December technique.15 pCASL was performed using a background-suppressed
2015. Six patients with MELAS were excluded because 3D pCASL 3D spiral FSE technique. The parameters were as follows: 8 arms
was not performed. Nine MELAS patients (seven males and two were used with 512 sampling points for each arm, spatial resolu-
females, age range from 12–45 years, mean 29.7 years) and 12 AIS tion 5 3.64 mm, bandwidth 5 6 62.5 Hz, TR 5 4,844 msec,
patients (eight males and four females, age range from 22–53 years, TE 5 10.5 msec, postlabeling delay (PLD) 5 2025 msec, slice
mean 36.7 years) were finally included. They all underwent conven- thickness 5 4.0 mm, number of slices 5 36, FOV 5 24 3 24 cm,
tional MRI, DWI, and 3D pCASL. The 3D pCASL examinations matrix 5 512 3 512, NEX 5 3, and 36 pairs of tag and control
were performed within 48 hours after onset in all the AIS patients, images were acquired in 4 minutes 41 seconds. Spiral data were

Li et al.: Differential Diagnosis of MELAS
TABLE 1. Summary of the Clinical Features of Nine Consecutive Patients With MELAS
Patient no. Age (year) Sex Clinical presentation Lactate Diagnosisa
1 17 M Headache, seizure, short stature, fatigue 1 MELAS

intolerance
2 28 F Headache, paroxysmal seizure, unconscious- 1 MELAS
ness, right hand myoclonus
3 45 F Speech disturbance, memory impairment, 1 MELAS
hearing loss, fatigue intolerance
4 29 M Speech disturbance, weakness of right N/A MELAS
extremities, seizure, diabetes mellitus
5 12 M Intermittent fever, seizure 1 MELAS
6 41 M Headache, upper myoclonus, memory 1 MELAS
impairment, focal seizure
7 39 M Speech disturbance, weakness of right 1 MELAS
extremities
8 29 M Headache, bilateral hearing loss, blurred 1 MELAS
vision, weakness of lower extremities,
unconsciousness
9 27 M Headache, bilateral hearing loss, blurred 1 MELAS
vision, weakness of right extremities, short
stature
M 5 male, F 5 female, N/A 5 not available.
a
Finally diagnosed by genetic test and/or muscle biopsy.
regridded to a matrix of 512 3 512 3 36 for reconstruction. The between MELAS and AIS at both the central and the peripheral
scan position extended from the vertex to the base of the skull. part of the lesion was conducted using two independent sample t-
By empirical visual assessment, the brain regions with high test. The significance level was set at P < 0.05.
signal (compared with the relative normal brain) were regarded as
hyperintensity and the regions with low signal (compared with the Results
relative normal brain) were regarded as hypointensity on T2WI Clinical Characteristics
and DWI. Similarly, by visual observations, the brain regions with The clinical information and lactate level for patients with
darker colors (compared with the relatively normal brain) were MELAS are summarized in Table 1. All patients with T1
regarded as hyperperfusion regions, and the regions with lighter MELAS were proved by muscle biopsy and/or genetic test.
colors (compared with the relative normal brain) were regarded as
All patients with AIS were finally diagnosed with cerebral
hypoperfusion regions on 3D pCASL.
artery stenosis confirmed by vascular imaging (digital sub-
Quantitative cerebral blood flow (CBF) maps derived from
traction angiography and/or MR angiography) and clinical
3D pCASL were generated using Functool software (GE Health-
care) on an ADW 4.5 workstation.
follow-up.
In this study, each lesion was divided into the peripheral part
Conventional MRI Features
and central part. The cortical regions of the lesion were defined as
All 21 patients with recent onset of stroke-like episodes dem-
the peripheral part, and the white matter regions of the lesion were
onstrated hyperintensities on T2WI and DWI. MR findings
defined as the central part. In each patient, three to five round
ROIs (28–32 mm2) were manually and carefully placed in both the
for patients with MELAS are summarized in Table 2. T2
central and the peripheral parts of the lesions on CBF maps using 3D PCASL CHARACTERISTICS. Hyperperfusion was dem-
conventional T2WI and DWI as the reference. The ROIs were onstrated in all 16 acute MELAS lesions, especially in the
drawn by two neuroradiologists with more than 5 years’ experience
peripheral part (Fig. 2), compared with the perfusion defects F2
who were blinded to the results.
or hypoperfusion in all acute ischemic infarctions (Fig. 3). F3
Statistical Analysis The CBF values of 16 lesions in MELAS were 11.20–
All statistical analyses were performed using the statistical package 73.11 ml/100g/min in the central part and 65.33–
SPSS for windows (v. 17.0, Chicago, IL). All data were expressed 169.11 ml/100g/min in the peripheral part. The CBF values
as mean 6 standard deviation (SD). The comparison of CBF values of 14 lesions in AIS were 12.32–19.94 ml/100g/min in the
January 2017 201

TABLE 2. MRI Characteristics of Nine Consecutive Patients With MELAS
Patient Lesion T1WI T2WI DWI ADC MRA Perfusion

no. location (overall)
1 R POT Slight hypo Slight hyper Hyper Hyper Normal Obvious hyper
2 R PO Slight hypo Slight hyper Hyper Hyper/hypo N/A Hyper
3 LPO/BL Slight hypo Hyper/ Obvious Hyper/hypo N/A Slight hyper
T/R Slight hyper hyper
cerebrum
4 BL POT Slight hypo Slight hyper Hyper Hypo Normal Hyper (L PO)
5 L POT Slight hypo Hyper/ Iso/Slight Hyper N/A Regional hyper
Slight hyper Hyper
6 LF/BL POT Slight hypo Slight hyper Iso/Slight Hyper N/A Regional hyper
Hyper
7 LT Slight hypo Slight hyper Hyper Hyper N/A Hyper
8 R POT Slight hypo Slight hyper Hyper Hyper/hypo Normal Hyper
9 R POT Slight hypo Slight hyper Hyper Hyper Normal Hyper
R 5 right, L 5 left, BL 5 bilateral, F 5 frontal lobe, T 5 temporal lobe, P 5 parietal lobe, O 5 occipital lobe, Hyper 5 hyperin-
tensity, Hypo 5 hypointensity, Iso 5 isointensity, N/A 5 not available.
central part and 11.66–18.37 ml/100g/min in the peripheral Discussion

F4 part (Fig. 4). The CBF value of the peripheral part MELAS has unpredictable clinical presentations and fea-
(119.80 6 35.41) in MELAS was significantly higher than tures, which makes it difficult to be timely and accurately
that (14.66 6 2.61) in AIS (P 5 0.0001), demonstrating a diagnosed, and it can be commonly misdiagnosed as viral
significant difference of CBF in the peripheral part of the encephalitis, ischemic infraction, or even brain tumor, espe-
lesions between the two groups. The CBF value of the cen- cially in young patients.16
tral part (37.71 6 19.27) in MELAS was also higher than The conventional MRI of MELAS demonstrates
that (17.15 6 2.69) in AIS (P < 0.05) (Fig. 4). abnormal signal involving both cortex and subcortical white
matter, particularly in the occipital and parietal areas that
MRI FOLLOW-UP. In a 17-year-old male MELAS patient strongly mimics ischemic stroke.17 However, the distribution
(case 1), 3D pCASL revealed hyperperfusion in the right of the stroke-like lesions is incongruent with vascular territo-
F5 occipital lobe at onset (Fig. 5d). Follow-up MR examina- ries and cerebral angiography confirmed to vessels without
tion 1 week later demonstrated decreased perfusion in the luminal stenosis or occlusion.18 Therefore, we can speculate
right occipital lobe but hyperperfusion in the left occipital the underlying mechanism of MELAS is different from that
lobe was noted (Fig. 5h), although no abnormal signal of ischemic stroke caused by energy failure from defective
changes were noted in the left occipital lobe on T2WI or oxidative metabolic pathways of energy production.19 The
DWI. precise pathogenesis of the stroke-like episodes in patients
C
O
L
O
R
FIGURE 2: A 28-year-old female MELAS patient with sudden onset of headache, paroxysmal seizure, and unconsciousness for 1
week. The lesion in the right parietal and occipital lobes presents as a: Slight hyperintensity on T2WI. b: Hyperintensity on DWI. c:
Mixed hyper- and hypointensities on ADC map. d: Hyperperfusion in the peripheral parts on 3D pCASL.

C
O
L
O
R
FIGURE 3: A 28-year-old AIS patient with sudden onset of headache and nausea for 2 days. The lesion in the right occipital lobe
presents as a: Slight hyperintensity on T2WI. b: Hyperintensity on DWI. c: Slight hypointensity on ADC map. d: Hypoperfusion on
3D pCASL.
with MELAS has not been fully elucidated, although three pulsed arterial spin labeling (PASL) or continuous arterial
major hypothesis have been proposed: mitochondrial angi- spin labeling (CASL) according to different labeling strat-
opathy, mitochondrial cytopathy, and nonischemic neurovas- egies.26 3D pCASL, as a newly developed ASL sequence,
cular hyperexcitability.18–20 has potentially combined the advantage of both CASL and
DWI reflects the random microscopic motion of water PASL methods,26 allowing for 3D acquisition and fewer sus-
molecules, and it is sensitive in detecting the pathologic ceptibility artifacts by using a spiral acquisition with a back-
changes and can be used to discriminate cytotoxic edema ground suppressed 3D fast spin echo (FSE) technique.
from vasogenic edema to some extent.21 However, there Table 3 summarizes the ASL perfusion findings in T3
were conflicting reports regarding the diffusion appearance stroke-like lesions of MELAS in previous reports,19,27–29 in
of stroke-like lesions in patients with MELAS.18 Several which a total of six patients were studied using 3D pCASL
investigations reported that stroke-like lesions of MELAS technique.27,28 In this study, we compared 3D pCASL per-
showed normal or increased apparent diffusion coefficient fusion between nine MELAS and 12 AIS patients, and we
(ADC), suggesting the presence of vasogenic edema, and found the newly appearing lesions, especially in the periph-
even put forward that an increased ADC could be regarded eral part, were clearly visualized as hyperperfusion in all
as differential point from the cytotoxic edema of an ische- nine MELAS patients compared with the perfusion defects
mic stroke origin.17,22,23 On the other hand, some stud- or hypoperfusion in AIS.
ies24,25 have described decreased ADC within the acute We speculated that, at the acute stage of stroke-like
stroke-like lesions of MELAS, suggesting the presence of episodes in MELAS, the local accumulation of lactate pro-
cytotoxic edema, although the underlying pathophysiological duction due to deficiency or dysfunction of mitochondrial
mechanisms of MELAS was different from that of an early oxidative enzymes might result in a change of regional vas-
ischemic stroke.19 Therefore, it is sometimes difficult to dif- cular properties, such as increased capillary permeability, or
ferentiate MELAS from AIS when ADC partially overlaps induce the local edema and hyperperfusion status.30 Hyper-
in these two conditions. perfusion could be the direct result of abnormal vascular
Arterial spin labeling (ASL) has been used as a novel autoregulatory mechanisms from mitochondrial angiopathy,
MR approach for quantifying and visualizing CBF without especially in the cortical regions (the peripheral part of the
exogenous contrast agents, mainly distinguished as either lesion) because of abundant blood vessels in the cortices.
FIGURE 4: The scatterplot of cerebral blood flow (CBF) value in MELAS and AIS. a: CBF values in the central part. b: CBF values in
the peripheral part.
January 2017 203

C
O
L
O
R
FIGURE 5: A 17-year-old male patient with headache and seizure for about 1 month. The lesion in the right occipital lobes
presents as a: Slight hyperintensity on T2WI. b: Hyperintensity on DWI. c: Hyperintensity on ADC map. d: Hyperperfusion in the
peripheral parts (arrows) and slight hyperperfusion in the central part (*) on 3D pCASL. Follow-up MR examination 1 week later.
e: Isointensity on T2WI. f: Isointensity on DWI. g: Isointensity on ADC map. h: Decreased perfusion in the right occipital lobe
(arrows) but newly developed hyperperfusion in the left occipital lobe (~) on 3D pCASL.
TABLE 3. ASL Perfusion Features of MELAS in the Literature
Reference Methods No. of patient Objective(O) and Conclusion(C)
(Rodan et al 2015) 28 pCASL and three siblings (O)To study the underlying mechanisms of stroke-like
BOLD-MRI (two females) episodes in MELAS patients.
(C) MELAS disease severity and mutation load were
inversely correlated with interictal cerebrovascular
reactivity and directly correlated with frontal CBF.
(Ikawa et al 2013) 27 pCASL three patients (O)To detect the pre-clinically latent lesions in stroke-
like episodes of patients with MELAS
(C) pCASL had great potential for detecting preclini-
cal lesions and predicting the emergence of stroke-like
episodes
(Tsujikawa et al 2012) 29 CASL one patient (O) To show the crossed cerebellar hyperperfusion
after acute MELAS stroke-like episodes.
(C) CASL was useful in the follow up of a patient
with acute symptoms of MELAS.
(Tsujikawa et al 2010) 19 CASL and two (O) To assess the roles of serial MRS and CASL in
MRS patients MELAS patients.
(C) Active MELAS lesions can be differentiated from
chronic inactive lesions using CASL perfusion. Lactate
peak detected by MRS can be considered as systemic
lactic acidosis in MELAS.

However, persistent cellular injury due to impaired adeno- 7. Krishnamurthi RV, deVeber G, Feigin VL, et al. Stroke Prevalence,
Mortality and Disability-Adjusted Life Years in Children and Youth
sine triphosphate production may finally give rise to the Aged 0-19 Years: Data from the Global and Regional Burden of
excessive loss of neurons, resulting in the decreased CBF in Stroke 2013. Neuroepidemiology 2015;45:177–189.
affected brain regions at the chronic phase of stroke-like 8. Putaala J, Metso AJ, Metso TM, et al. Analysis of 1008 consecutive
patients aged 15 to 49 with first-ever ischemic stroke: the Helsinki
episodes. young stroke registry. Stroke 2009;40:1195–1203.
Follow-up 3D pCASL examination in a MELAS
9. Kuklina EV, Tong X, George MG, et al. Epidemiology and prevention
patient (case 1) demonstrated dynamic perfusion changes of stroke: a worldwide perspective. Expert Rev Neurother 2012;12:
with decreased perfusion in the right occipital lobe and 199–208.
newly detected hyperperfusion in the left occipital lobe, 10. Guo L, Zhang Q, Ding L, et al. Pseudo-continuous arterial spin label-
ing quantifies cerebral blood flow in patients with acute ischemic
although no abnormal signal changes were noted in the left stroke and chronic lacunar stroke. Clin Neurol Neurosurg 2014;125:
occipital lobe on T2WI or DWI. Later on, the patient 229–236.
repeatedly suffered from right homonymous hemianopsia 11. Niibo T, Ohta H, Yonenaga K, et al. Arterial spin-labeled perfusion
imaging to predict mismatch in acute ischemic stroke. Stroke 2013;
(presumably caused by the left occipital lobe lesion), indicat-
44:2601–2603.
ing that the latent lesions in the left occipital lobe could be
12. Chen Y, Wang DJ, Detre JA, Test-retest reliability of arterial spin
revealed at the preclinical stage using pCASL. We suggest labeling with common labeling strategies. J Magn Reson Imaging
that the pCASL perfusion technique may be suitable for JMRI 2011;33:940–949.
follow-up examinations during interictal period for the 13. Jiang J, Zhao L, Zhang Y, et al. Comparative analysis of arterial spin
labeling and dynamic susceptibility contrast perfusion imaging for
detection of the latent disease in MELAS when compared to quantitative perfusion measurements of brain tumors. Int J Clin Exp
conventional MRI methodologies. Ikawa et al27 also Pathol 2014;7:2790–2799.
reported that the preclinical latent lesions demonstrated as 14. Deibler AR, Pollock JM, Kraft RA, et al. Arterial spin-labeling in rou-
regional hyperperfusion on pCASL 3 months before the tine clinical practice, part 1: technique and artifacts. AJNR Am J Neu-
roradiol 2008;29:1228–1234.
clinical onset of stroke-like episodes in patients with
15. Alsop DC, Detra JA, Golay X, et al. Recommended implementation
MELAS, which could not be detected by conventional MRI of arterial spin-labeled perfusion MRI for clinical applications: A con-
and DWI in the preclinical phase. sensus of the ISMRM perfusion study group and the European consor-
tium for ASL in dementia. Magn Reson Med 2015;73:102–116.
One limitation of the present study was its relatively
16. Wang YX, Le WD. Progress in diagnosing mitochondrial myopathy,
small sample size. Another limitation was that it was a retro- encephalopathy, lactic acidosis, and stroke-like episodes. Chin Med J
spective and cross-sectional study, and serial pCASL was per- 2015;128:1820–1825.
formed in only one case. Serial pCASL study would provide 17. Oppenheim C, Galanaud D, Samson Y, et al. Can diffusion
more valuable information for the dynamic CBF changes. weighted magnetic resonance imaging help differentiate stroke from
stroke-like events in MELAS? J Neurol Neurosurg Psychiatry 2000;69:
In conclusion, hyperperfusion can be demonstrated in 248–250.
the involved brain regions at clinical onset or even at the 18. Kim JH, Lim MK, Jeon TY, et al. Diffusion and perfusion characteristics
preclinical stage in patients with MELAS. The whole-brain of MELAS (mitochondrial myopathy, encephalopathy, lactic acidosis,
and stroke-like episode) in thirteen patients. Korean J Radiol 2011;12:
3D pCASL technique is useful in differentiating MELAS 15–24.
from AIS when the clinical manifestations and conventional
19. Tsujikawa T, Yoneda M, Shimizu Y, et al. Pathophysiologic evaluation
MR findings overlap in these two conditions. of MELAS strokes by serially quantified MRS and CASL perfusion
images. Brain Dev 2010;32:143–149.
References 20. Iizuka T, Sakai F. Pathogenesis of stroke-like episodes in MELAS: anal-

ysis of neurovascular cellular mechanisms. Curr Neurovasc Res 2005;2:
1. Barshop BA, Naviaux RK, McGowan KA, et al. Chronic treatment of
29–45.
mitochondrial disease patients with dichloroacetate. Mol Genet
Metab 2004;83:138–149. 21. Schaefer PW, Buonanno FS, Gonzalez RG, et al. Diffusion-weighted
imaging discriminates between cytotoxic and vasogenic edema in a
2. El-Hattab AW, Adesina AM, Jones J, et al. MELAS syndrome: Clinical
patient with eclampsia. Stroke 1997;28:1082–1085.
manifestations, pathogenesis, and treatment options. Mol Genet
Metab 2015;116:4–12. 22. Ohshita T, Oka M, Imon Y, et al. Serial diffusion-weighted imaging in
MELAS. Neuroradiology 2000;42:651–656.
3. Yatsuga S, Povalko N, Nishioka J, et al. MELAS: a nationwide pro-
spective cohort study of 96 patients in Japan. Biochim Biophys Acta 23. Kolb SJ, Costello F, Lee AG, et al. Distinguishing ischemic stroke
2012;1820:619–624. from the stroke-like lesions of MELAS using apparent diffusion coeffi-
cient mapping. J Neurol Sci 2003;216:11–15.
4. Sproule DM, Kaufmann P. Mitochondrial encephalopathy, lactic acido-
sis, and strokelike episodes: basic concepts, clinical phenotype, and 24. Wang XY, Noguchi K, Takashima S, et al. Serial diffusion-weighted
therapeutic management of MELAS syndrome. Ann N Y Acad Sci imaging in a patient with MELAS and presumed cytotoxic oedema.
2008;1142:133–158. Neuroradiology 2003;45:640–643.
5. Huang Y, Yang S, Jia J. Factors related to long-term post-stroke cog- 25. Stoquart-Elsankari S, Lehmann P, Perin B, et al. MRI and diffusion-
nitive impairment in young adult ischemic stroke. Med Sci Monit weighted imaging followup of a stroke-like event in a patient with
2015;21:654–660. MELAS. J Neurol 2008;255:1593–1595.
6. Abend NS, Beslow LA, Smith SE, et al. Seizures as a presenting symp- 26. Ferre JC, Bannier E, Raoult H, et al. Arterial spin labeling (ASL) perfu-
tom of acute arterial ischemic stroke in childhood. J Pediatr 2011;159: sion: techniques and clinical use. Diagn Interv Imaging 2013;94:1211–
479–483. 1223.
January 2017 205

27. Ikawa M, Yoneda M, Muramatsu T, et al. Detection of preclinically 29. Tsujikawa T, Yamamoto T, Ikawa M, et al. Crossed cerebellar
latent hyperperfusion due to stroke-like episodes by arterial spin- hyperperfusion after MELAS attack followed up by whole brain contin-
labeling perfusion MRI in MELAS patients. Mitochondrion 2013;13: uous arterial spin labeling perfusion imaging. Acta Radiol 2012;53:
676–680. 220–222.
28. Rodan LH, Poublanc J, Fisher JA, et al. Cerebral hyperperfusion and 30. Iizuka T, Sakai F, Suzuki N, et al. Neuronal hyperexcitability in
decreased cerebrovascular reactivity correlate with neurologic disease stroke-like episodes of MELAS syndrome. Neurology 2002;59:816–
severity in MELAS. Mitochondrion 2015;22:66–74. 824.

TECHNICAL DEVELOPMENT
Magnetic Susceptibility-Induced Echo-Time

Shifts: Is There a Bias in Age-Related
fMRI Studies?
Giang-Chau Ngo, MS,1,2* Chelsea N. Wong, MS,2,3 Steve Guo, BS,1,2
Thomas Paine, MS,2,4 Arthur F. Kramer, PhD,2,5 and Bradley P. Sutton, PhD1,2
Purpose: To evaluate the potential for bias in functional magnetic resonance imaging (fMRI) aging studies resulting
from age-related differences in magnetic field distributions that can impact echo time and functional contrast.
Materials and Methods: Magnetic field maps were taken on 31 younger adults (age: 22 6 2.9 years) and 46 older adults
(age: 66 6 4.5 years) on a 3T scanner. Using the spatial gradients of the magnetic field map for each participant, an echo
planar imaging (EPI) trajectory was simulated. The effective echo time, time at which the k-space trajectory is the closest to
the center of k-space, was calculated. This was used to examine both within-subject and across-age-group differences in
the effective echo time maps. The blood oxygenation level-dependent (BOLD) percent signal change resulting from those
echo time shifts was also calculated to determine their impact on fMRI aging studies.
Results: For a single subject, the effective echo time varied as much as 65 msec across the brain. An unpaired t-test between
the effective echo time across age groups resulted in significant differences in several regions of the brain (P < 0.01). The dif-
ference in echo time was only 1 msec, however, which is not expected to have an important impact on BOLD fMRI percent
signal change (<4%).
Conclusion: Susceptibility-induced magnetic field gradients induce local echo-time shifts in gradient echo fMRI images,
which can cause variable BOLD sensitivity across the brain. However, the age-related differences in BOLD signal are
expected to be small for an fMRI study at 3T.
G radient echo (GRE) imaging in magnetic resonance

imaging (MRI) is used in an ever-increasing set of
applications, including functional MRI (fMRI) using
data and relate it to physiological quantities relies on having
accurate information on the timing of the signal, in particular
the echo time (TE).
blood oxygenation level-dependent (BOLD) contrast,1 resting Differences in magnetic susceptibility between air and tis-
state functional connectivity,2 mapping T2 relaxation,3 sue result in large-scale spatial variations in the main magnetic
susceptibility-weighted imaging,4 and quantitative susceptibil- fields in the body or brain. Accompanying the spatial variation
ity mapping.5 For these methods, the gradient echo method is in magnetic field are spatial gradients in the magnetic field
used to allow perturbations to the magnetic field to accumu- strength, which will be referred to as susceptibility-induced
late sufficient phase to either cause signal cancellation in the magnetic field gradients (SI-MFG). These magnetic field gra-
magnitude images (eg, BOLD fMRI), or, in susceptibility- dients disrupt both the uniformity of the main magnetic field
weighted imaging applications, to enable a measurement of a and the linear magnetic field gradients used for spatial encoding
phase offset from the resulting disruption. For all of these in MRI. These disruptions can lead to several artifacts in MRI
GRE techniques, the ability to quantify the changes in the images, including image distortion, through-plane signal loss,
Received Dec 2, 2015, Accepted for publication May 31, 2016.
*Address reprint requests to: G.-C.N., 1410 Beckman Institute for Advanced Science and Technology, 405 N Mathews Ave., Urbana, IL 61801.
E-mail: gngo@illinois.edu
From the 1Department of Bioengineering, University of Illinois at Urbana-Champaign, Urbana, Illinois, USA; 2Beckman Institute of Technology for Advanced
Technology, University of Illinois at Urbana-Champaign, Urbana, Illinois, USA; 3Neuroscience Program, University of Illinois at Urbana-Champaign, Urbana,
Illinois, USA; 4Illinois Informatics Institute, University of Illinois at Urbana-Champaign, Urbana, Illinois, USA; and 5Department of Psychology, University of
Illinois at Urbana-Champaign, Urbana, Illinois, USA.

V 207

and k-space trajectory distortions.6,7 The first two artifacts have each imaging voxel will have its own net k-space trajectory and its
been extensively studied in the literature and several methods own effective TE.
exist to either minimize their effects or to correct image distor- It has been previously observed7,16,17 that in-plane SI-MFG
tions.8–11 In this work, we focus on the impact of SI-MFG that alter the image intensity or image contrast due to the induced echo
are in the plane of the imaging scan (referred to as in-plane SI- time shifts and k-space trajectory distortions. Reichenbach et al6
MFG) on the k-space trajectory. The distortions caused by in- noted that the susceptibility-induced k-space trajectory distortions
may result in a total signal loss if the k-space trajectory is shifted
plane SI-MFG on the k-space trajectory can impact BOLD
and skewed far enough such that the center of k-space is not
sensitivity.
sampled. In addition, previous literature has also examined the
We considered studies where the two groups being
impact of TE shifts on the BOLD contrast. Deichmann et al7
compared might contain significant differences in magnetic showed that different areas of the brain experienced different TE
field distributions in the brain. In such a study, there is shifts resulting in a nonuniformity of the BOLD sensitivity across
potential for conclusions to be made about differences in the brain of an individual subject. Mannfolk et al17 drew similar
brain function that should, instead, be attributed to differen- conclusions about the variability of the BOLD sensitivity. From
ces in BOLD sensitivity across the two groups. One type of acquired field maps, they estimated an increase of the BOLD sensi-
study that potentially suffers from group-level differences in tivity in the hippocampus and a drop in the anterior region of the
magnetic field distributions are studies of the aging brain. hippocampus with an echo-planar imaging (EPI) acquisition.
Aging is accompanied by many changes in the brain,12 Therefore, TE shifts in a GRE image can have a drastic impact on
including changes in structural organization of the tissues,13 the image contrast, prohibiting extraction of meaningful quantita-
tissue composition (such as iron deposits),14 and in the aver- tive results of brain activity.
In order to determine the potential impacts on echo time in an
age angulation of the head.15 These group-level differences
fMRI acquisition, we simulate net k-space trajectories for each voxel
could have a significant impact on the magnetic field distri-
based on a susceptibility-gradient-free k-space trajectory and esti-
bution across the brain with age, hence leading to age-related
mated SI-MFG. We refer to the gradient in the field map in the read-
k-space trajectory distortions and TE shifts, and potentially to out or X-direction as GX and the gradient in the field map in the
bias in detecting age-related differences in brain function. phase-encode or Y-direction as GY. The in-plane SI-MFG maps (GX
In this article we explore the impact of in-plane SI- and GY only) were used in the calculation of the net k-space trajectory
MFG on k-space trajectory and echo-time shifts for GRE for each voxel over the entire acquisition. The gradients have been
imaging at the commonly used field strength of 3T, focusing defined as positive from left to right (X-direction) and from anterior
on the potential impact on aging studies. A method based on to posterior (Y-direction), corresponding to the positive gradient axes
acquired field maps is used here to estimate the effective k- for the imaging slice prescription. The net k-space trajectory (kx,net,
space trajectories, taking into account the in-plane SI-MFG. ky,net) was estimated using the following formulas:
The potential bias induced in the GRE measures for studies
kx;net ðtÞ5kx ðtÞ1GX t; ky:net ðtÞ5ky ðtÞ1Gy t (1)
across age groups is evaluated by examining the magnetic
field distributions in younger adults and older adults col- kx(t) and ky(t) are susceptibility-gradient-free trajectories in the read-
lected as part of an aging study. We hypothesized that age- and phase-encoding directions, respectively, t is the timing vector for
associated changes in the magnetic field maps would result in each sample point in the trajectory, and GX and GY are the in-plane
regions in the brain that experience a significant difference in SI-MFG. GX and GY have units of Hz/cm, and k-space has units of
effective TE for the two age groups. We note that the current cm21.
work is focused on the impact on BOLD sensitivity of in- Given the net k-space trajectory for each voxel, the effective
plane SI-MFG, as opposed to through-plane SI-MFG. The TE maps were calculated by determining the point in the k-space
trajectory that passed closest to the (kx,ky) 5 (0,0) origin point of
signal dropouts caused by through-plane SI-MFG have been
k-space. Depending of the in-plane SI-MFG, the effective TE can
previously studied and techniques exist to minimize their
be very different from the planned or nominal TE.
impact on the fMRI results.8,9
Experiment Design
Materials and Methods All experiments were performed on a Siemens (Erlangen, Germany)
Echo Shifting due to In-Plane Susceptibility Allegra 3T MRI scanner. All participants signed an informed consent
Gradients in accordance with the local Institutional Review Board. Field maps
For a proton in an area with spatial gradients in the magnetic field, were obtained using the vendor-supplied GRE field mapping
the in-plane SI-MFG act in the same manner as spatial encoding sequence with echo times of 10 and 12.46 msec, chosen to avoid fat/
gradients that are purposefully applied for imaging, resulting in a water frequency differences in the field map. The field map scans
net imaging or k-space trajectory that is different from the one were acquired with the same slice prescription that was used in the
intended. Importantly, the point at which the trajectory would associated fMRI study, oblique-axial slices aligned with the anterior
have refocused, creating a gradient echo, will change resulting in a commissure-posterior commissure (AC-PC) line, a common choice
shift in TE. Due to the spatial variation in the in-plane SI-MFG, for an fMRI study. In all, 35 slices were imaged with a thickness of

Ngo et al.: Age-Related fMRI Bias
4 mm, enabling full brain coverage. The in-plane spatial resolution of

the field map scan was 3.4 3 3.4 mm2, matching the spatial resolu-
tion of the fMRI scan. Additionally, a T1-weighted anatomical scan
was acquired using a magnetization-prepared rapid acquisition of gra-
dient echo (MPRAGE) sequence with TE/TI/TR 5 2.3/900/1900
msec and a spatial resolution of 0.9 3 0.9 3 0.9 mm3. To examine
the magnetic field distribution across age, field maps were acquired
from 31 healthy younger adults (mean age: 22 years; age range: 18– FIGURE 1: Field map and gradient maps for one slice in the
orbitofrontal cortex for one subject.
29 years) and 46 healthy older adults (mean age: 66 years; age range:
60–77 years). Two younger subjects were removed from the study
the mean effective TE of the younger group was larger than the
due to failure in the image normalization step.
one of the older group.
We characterized the impact of effective echo time differen-
Effective Echo Time Map Estimation
ces on fMRI measurements for the worst-case locations in the
Field maps were obtained with units of Hz. Gradients of the field
brain. For these regions with highly significant differences in echo
map were taken by forward differences in the direction of the
time, we analyzed the expected BOLD percent signal change. This
positive readout- and phase-encode axes associated with the imag-
signal change was placed into the context of typically expected
ing slice prescription. They were calculated in Hz/cm by dividing
BOLD differences in aging studies. The expected BOLD percent
the first-order differences by the voxel dimension. Given that the
signal change was calculated using the following formula:
field maps are fairly smooth, this differencing method is expected
to give accurate estimates of the field map gradients. Field maps 2T TE
2T TE

and gradient maps were normalized to standard space using e 2;activation 2e 2;rest
PSC 5 2T TE
(2)
FNIRT in FSL.18 Note that the SI-MFG were obtained in subject e

2;rest
space prior to normalization, so that the readout- and phase-

encode directions would be relative to the subject-space, fMRI where a baseline T2 of 66 msec was assumed21 and an activation-
slice prescription. induced change of T2 was simulated that yielded activations of 1%
GX and GY were then used to calculate the net k-space tra- and 5% BOLD signal change in the susceptibility-free case.
jectories and the effective echo time map using Equation 1. An As head orientation may be one characteristic that leads to
EPI k-space trajectory was simulated with the phase-encoding magnetic field distribution differences, we also examined the aver-
direction defined as the anterior–posterior direction aligned with age nod rotation for the young and old group. We performed lin-
the AC-PC line. The simulated EPI fMRI acquisition sequence ear alignment of the high-resolution structural scan (MPRAGE)
used the following parameters: a nominal echo time of 30 msec, a with the MNI template and extracted the nod rotation angle to
matrix size of 64 in a 22-cm field of view, and a parallel imaging determine if there is an age-related difference in rotation of the
factor of 2. The EPI trajectory had an echo spacing of 0.4 msec, in head.
accordance with our commonly used acquisition parameters. Voxels
were discarded from analysis if the net trajectory (imaging plus in-
Results
plane SI-MFG) predicted that the center of k-space was not An example magnetic field distribution and SI-MFG map
sampled, ie, that the net k-space trajectory did not pass within 1 k- for a young subject is shown in Fig. 1 for a slice including F1
space sampling distance from the origin of k-space. the orbitofrontal cortex. The magnetic field map has values
that exceed 100 Hz in the slice shown. The SI-MFG of the
Statistical Analysis field map, GX, GY, and GZ reach approximately 6 80 Hz/
Using the EPI acquisition described above and the net k-space tra- cm. In order to assess the impact of the in-plane SI-MFG
jectory simulation, which includes the voxel-by-voxel susceptibility on the k-space trajectory, the sign of the in-plane SI-MFG
gradient values, an effective TE map for each subject was esti- is important relative to the sign of the applied imaging gra-
mated. The mean and standard deviation of the effective TE maps dients and hence the planned traversal direction through k-
were calculated for each age group.
space.
An unpaired t-test on the effective echo time maps translated
The impact of the in-plane SI-MFG in the phase-
into standard space was performed to determine if magnetic field
encode direction of an EPI sequence can be seen in Fig. 2. F2
differences result in echo time differences between the two groups.
Figure 2 shows seven cases of k-space shifts and effective TE
We did not include a correction for multiple comparisons but have
displayed our results with a stringent threshold of P < 0.01. To
resulting from different values of in-plane SI-MFG, GY, in
evaluate the effect size of the differences in effective TE between the phase-encode direction only. The impact of SI-MFG in
the two age groups, Cohen’s d19 was also calculated. Several regions the X-direction, read-encoding direction, is shown in Sup-
of interest were examined where differences in effective TE were plementary Fig. S1. In Fig. 2, a GY of 240 Hz/cm, which is
significant. We used the value of 0.2 to indicate reasonable effects within the range of SI-MFG values in the brain as observed
size for aging fMRI studies.20 A Cohen’s d of less than 0.2 was in Fig. 1, can shift the echo time by 8 msec. On the other
considered negligible, whereas a value larger than 0.2 indicates that hand, a GY of 1 40 Hz/cm shifts the k-space trajectory so
January 2017 209

The impact on echo time shifts is not just limited to

slices at the air/tissue interfaces in the brain, such as the
orbitofrontal cortex. Figure 3 shows an example of an esti- F3
mated effective TE map for the EPI trajectory for one older
subject across several axial locations in the brain. All subjects
show similar distribution of effective TE across the brain.
The slice-value labels correspond to the slice location in
mm in the standard MNI space as viewed in FSLview, part
of FMRIB’s Software Library (www.fmrib.ox.ac.uk/fsl).
Regions that resulted in net k-space trajectories that do not
sample the center of k-space are denoted by a lack of color
overlay. The effective TE is not uniform across the brain in
C
the presence of the in-plane SI-MFG. The deviation from
O the nominal or planned TE can be significant (65 msec).
L
O
Additionally, the center of k-space is not sampled in regions
R such as the inferior temporal lobes, known to be regions
FIGURE 2: Impact of in-plane SI-MFG in the phase-encoding with significant magnetic field inhomogeneity.
direction on the k-space trajectory of an EPI sequence. EPI trajec-
Figure 4 shows the mean effective TE and standard F4
tories were simulated with an echo time of 30 msec with typical
timing parameters and a positive phase-encoding direction of deviation maps for three slices for each age group. As
anterior to posterior (up to down). Seven values of SI-MFG in the observed previously, the effective TE varies more than 65
phase-encoding direction were simulated.
msec across the brain for both older and younger adults.
The variation across space is larger than the standard devia-
tion across all subjects, indicating a good general consistency
much that the center of k-space is not sampled anymore. in the field map and TE values across subjects.22
We note that the impact of in-plane SI-MFG includes a An unpaired t-test was performed between the TE
shift in the starting point of the k-space trajectory due to maps of the older and younger groups. The thresholded
in-plane SI-MFG effects during the time from the RF exci- uncorrected P-value map (P 0.01) is shown in Fig. 5A, F5
tation to the start of the readout. In addition, the in-plane overlaid on the mean TE image for the same slices as shown
SI-MFG will skew the trajectory by increasing or decreasing in Fig. 4. Figure 5B,C shows, respectively, the differences
the spacing between the k-space lines, depending on and the absolute mean differences in the TE maps between
whether it is pushing with or against the imaging trajectory. the older and younger groups. Although significant statisti-
All these effects change the sampling time of the center of cal differences between groups were observed around the
k-space (kx,ky) 5 (0,0), and therefore shift the echo time. In occipital and temporal lobes, the difference in effective echo
the case of in-plane SI-MFG in the readout direction, simi- time between the two groups was only on the order of 1
lar observations can be made, as shown in Fig. S1 and in msec. Table S1 in the Supplementary Material describes
Refs. 11,16. For the same value of in-plane SI-MFG, the more precisely the brain regions where significant differences
effect on the k-space trajectory in the readout direction is were found. For these regions, Cohen’s d was calculated and
less impactful than in the phase-encoding direction. varies from 0.43 in the superior temporal gyrus to 0.74 in
C
O
L
O
R
FIGURE 3: Effective TE map for one older subject for an EPI sequence with phase-encode axis in the anterior to posterior direc-
tion with a planned TE of 30 msec. Portions without a color overlay indicate regions in which the k-space distortions are so large
that the center of k-space was not sampled. Z-slice locations are in MNI coordinates.

C
O
L
O
R
FIGURE 4: Mean effective TE in msec (A,B) and standard deviation (C,D) maps for the older age group (A,C) and the younger age
group (B,D). Three representative slices are shown at MNI coordinates of 26, 10, and 26 mm.
the posterior cingulate. For the middle temporal gyrus, the To examine the potential impact of this echo time
Cohen’s d is negative (20.41), indicating that the mean shift on the fMRI signal, the BOLD percent signal change
effective TE in that region for the older group is bigger was estimated for the worst-case shift of 1 msec from the
than the younger group. nominal echo time (resulting in an effective echo time of
C
O
L
O
R
FIGURE 5: Comparison of echo time shifts across older and younger subjects. A: Thresholded uncorrected P-value maps (P £
0.01) overlaid on the mean effective TE map demonstrates several regions with significant difference. B: The differences between
the mean effective TE of the older group and the younger group. C: The absolute differences between the mean effective TE
between the two age group indicating small differences between old and young subjects.
January 2017 211

31 msec) using Equation 2. The percent signal change the magnetic field distribution with age will have a strong
becomes 1.03% compared to the susceptibility-free case of impact on fMRI signal in aging studies at 3T.
an initial percent signal change of 1%. In the case of an ini- Statistically significant differences do exist in the effec-
tial percent signal change of 5%, an increase of 1 msec in tive echo time across the two age groups. The Cohen’s d
echo time due to in-plane SI-MFG will lead to a percent coefficients for the regions of interest describe an important
signal change to 5.17%. effect (>0.2), indicating that there are systematic age-related
Finally, head rotations were examined between the two differences in the magnetic field distribution. However, the
age groups as a potential source of SI-MFG differences. The effective echo time differences are small, 1 msec in the
average head rotation (nod direction) for older subjects worst case, resulting in a difference of 4% or less in the
aligning to the MNI template was 214.38 6 5.88, while it BOLD percent signal change between the two groups.
was 28.88 6 5.28 for younger subjects. This corresponds to These variations are only due to SI-MFG and do not repre-
a significant rotation of the head between the two groups sent any underlying differences in neural activity. However,
due to age (P-value of 6 3 1025). there are two critical factors to consider when assessing the
impact of this level of change on an aging fMRI study. First,
Discussion the ROIs in the current analysis are defined based on maxi-
SI-MFG exists in the brain due to the large number of air– mal changes in echo time. These ROIs are small (20–80
tissue interfaces in and around the head. These magnetic voxels) and would be encompassed by larger ROIs defined
field gradients introduce local shifts in the k-space trajectory by anatomical variations instead of SI-MFG. Second, a vari-
that lead to differences in effective echo time across the ation of 4% in the BOLD percent signal change is small
brain. The impact of these in-plane SI-MFG may need to compared to the difference commonly reported in aging
be taken into account when using gradient echo imaging, as studies, which is around 10–20%24 and is smaller than
the effective echo time can differ from the intended nominal reported BOLD percent signal change standard deviation.25
echo time. As the planned center of the k-space trajectory It is not expected that in-plane SI-MFG will lead to bias in
does not coincide with the center of k-space (kx,ky) 5 (0,0), BOLD aging studies. The method we used to examine this
potential for bias could be applied in future studies to
the signal intensity is altered, resulting in a different image
examine specific age-related fMRI comparisons. By collect-
contrast. If the k-space trajectory is shifted out of the acqui-
ing field maps and simulating k-space trajectories, future
sition window, the majority of the energy of the signal is
studies could window the maximum age-related differences
lost. The impact of in-plane SI-MFG on the BOLD sensi-
in echo time and sensitivity and compare them to the mag-
tivity can exist even without noticeable changes in image
nitude of their age-related fMRI differences.
intensity. Deichmann et al7 showed that even if the image
The method presented here to calculate the net k-space
intensity is acceptable, the BOLD sensitivity can drop con-
trajectory and the effective echo time is based on accurate
siderably. The amount of variation and the impact on quan-
field maps and gradients maps. It is important that the field
titative gradient echo measures will depend on the specific
map slices be aligned with the functional imaging slices and
pulse sequence and its timing in the coverage of k-space. In
that the two datasets are acquired with the same readout
our observations, there was significant spatial variations (of and phase-encoding directions. In order to determine the
around 5 msec) in the effective echo time associated with a gradients in the magnetic field map in each direction, we
BOLD fMRI experiment within the head of each partici- used a first-order difference operation in the positive direc-
pant. However, across subjects, the magnetic field distribution of the gradient, as this provides the most direct rela-
tion was highly conserved, resulting in a variation of the tionship to image encoding gradients and k-space
effective echo time for each location that was much smaller, trajectories. We expect that the field maps are smooth and
on the order of 1 msec. that this operator results in fairly accurate measures of the
Our original hypothesis addressed possible bias in linear SI-MFG within an imaging voxel. However, higher-
BOLD fMRI signal due to systematic differences in the field resolution field map acquisitions could be used to get more
map across age. These systematic differences could result localized estimates of the SI-MFG, with spatial derivatives
from different factors such as selective volumetric declines, accessible within an imaging voxel. This higher spatial reso-
changes in the angulation of the head and spine leading to lution field map may lead to better estimates of the gra-
changes in head orientation, and accumulation of iron dients in regions where curvature of the field map is high.
deposits in particular brain regions. The differences in head In-plane SI-MFG are often highly correlated with
rotation that we observed can be explained by a change of regions with high through-plane SI-MFG. The additional
the curvature of the spine with age and could lead to a effects from these magnetic field distributions can be very
change in the magnetic field, as previously identified.23 significant and dominate or mask the effects due to the in-
However, our results do not indicate that the difference in plane gradients. Through-plane SI-MFG have been shown

to lead to important signal losses due to intravoxel dephas- on gradient echo k-space trajectories. In this work, the
ing. Many techniques including z-shimming or reducing the effects of in-plane SI-MFG on a typical fMRI experiment
slice thickness8,9 have been adopted to minimize the signal were examined. A variation in the effective echo time of 65
dropouts in the slice direction. msec was observed across the brain in each individual on a
Some methods have been previously developed to alle- 3T scanner. When examining older and younger partici-
viate the impacts of in-plane SI-MFG on the BOLD con- pants, significant age-related differences in the in-plane
trast. Deichmann et al7 suggested to use magnetic gradients SI-MFG-induced effective echo time can be seen. However,
to compensate for the SI-MFG and mitigate the BOLD the magnitude of change in TE is small (1 msec) and no
sensitivity loss. Weiskopf et al16 emphasized the importance measurable age-related bias in BOLD fMRI signal due to
of correcting SI-MFG in the readout direction and reduced in-plane SI-MFG is expected. In-plane SI-MFG and
signal loss by decreasing echo time and increasing spatial echo time shifts will become stronger when using higher
resolution in the readout direction. In Ref. 26, spiral-in tra- field scanners and GRE techniques at high field should be
jectories were compared to spiral-out trajectories to further examined for increased sensitivity to SI-MFG.
explore and validate the BOLD sensitivity changes expected
for different acquisition strategies. These techniques offer
the possibility to adjust the imaging prescription at run-
time for a specific subject and targeted ROI. However, they
Acknowledgments
are not able to completely eliminate BOLD sensitivity varia- The project described was supported by Award Number
tions across the entire scanning volume and potential bias R21EB010095 from the National Institute Of Biomedical
across the brain can remain in BOLD studies. Imaging And Bioengineering. The content is solely the
In addition to the impact on BOLD fMRI, SI-MFG responsibility of the authors and does not necessarily repre-
can have an effect on other techniques that rely on GRE sent the official views of the National Institute Of Biomedi-
images, such as the estimation of the relaxation parameter cal Imaging And Bioengineering or the National Institutes
T2 in quantitative imaging techniques. It has been shown of Health.
that due to SI-MFG the T2 decay has a more complex
behavior than a monoexponential decay.27,28 Therefore, a
model taking into account SI-MFG is crucial for the accu-
rate estimation of T2 . References
1. Ogawa S, Tank DW, Menon R, et al. Intrinsic signal changes accompany-
Our study showed that no substantial bias due to in- ing sensory stimulation: functional brain mapping with magnetic reso-
plane SI-MFG is expected in BOLD aging studies at 3T. nance imaging. Proc Natl Acad Sci U S A 1992;89:5951–5955.
However, several limitations of the current study should be 2. Biswal B, Yetkin FZ, Haughton VM, Hyde JS. Functional connectivity in
addressed. First, the study is limited by the small number of the motor cortex of resting human brain using echo-planar MRI. Magn
Reson Med 1995;34:537–541.
subjects involved. Despite the small number of subjects, we
3. Speck O, Hennig J. Functional imaging by I0- and T2*-parameter
did find significantly different regions across age in the field mapping using multi-image EPI. Magn Reson Med 1998;40:243–248.
map. However, more sensitivity to these differences would
4. Haacke EM, Xu Y, Cheng YCN, Reichenbach JR. Susceptibility
come from a larger study population. Also, all subjects were weighted imaging (SWI). Magn Reson Med 2004;52:612–618.
scanned on the same scanner and with the same field map 5. Langkammer C, Schweser F, Krebs N, et al. Quantitative susceptibility
sequence. Some differences in shimming capabilities would mapping (QSM) as a means to measure brain iron?. A post mortem
validation study. Neuroimage 2013;62:1593–1599.
be expected across scanners and this should be investigated
in future work. The conclusions drawn in this work are 6. Reichenbach JR, Venkatesan R, Yablonskiy DA, Thompson MR, Lai S,
Haacke EM. Theory and application of static field inhomogeneity effects
applicable to 3T scanners, but may be even more important in gradient-echo imaging. J Magn Reson Imaging 1997;7:266–279.
at higher field where larger differences in BOLD signal 7. Deichmann R, Josephs O, Hutton C, Corfield DR, Turner R. Compen-
would be expected if a GRE sequence was used. In addition, sation of susceptibility-induced BOLD sensitivity losses in echo-planar
fMRI imaging. NeuroImage 2002;15:120–135.
through-plane SI-MFG were not the focus of this work and
could lead to additional and more significant differences in 8. Hsu JJ, Glover GH. Mitigation of susceptibility-induced signal loss in
neuroimaging using localized shim coils. Magn Reson Med 2005:53:
image contrast across age groups due to signal losses. Finally, 243–248.
as described above, the estimates of the effective echo time 9. Merboldt KD, Finsterbusch J, Frahm J. Reducing inhomogeneity
rely on accurate field maps and SI-MFG maps. By the use artifacts in functional MRI of human brain activation-thin sections vs
gradient compensation. J Magn Reson 2000:145:184–191.
of the first-order difference, accurate SI-MFG are difficult
to obtain on the edges of the brain using the current 10. Jezzard P, Balaban RS. Correction for geometric distortion in echo pla-
nar images from B0 field variations. Magn Reson Med 1995:34:65–73.
approach.
11. Holland D, Kuperman JM, Dale AM. Efficient correction of inhomoge-
In conclusion, magnetic field inhomogeneities due to neous static magnetic field-induced distortion in echo planar imaging.
air/tissue interfaces around the brain have a strong impact Neuroimage 2010:50:175–183.
January 2017 213

12. Creasey H, Rapoport SI. The aging human brain. Ann Neurol 1985:17: 21. Peters AM, Brookes MJ, Hoogenraad FG, et al. T2* measurements
2–10. in human brain at 1.5, 3 and 7 T. Magn Reson Imaging 2007:25:748–753.
13. Good CD, Johnsrude IS, Ashburner J, Henson RN, Friston KJ, 22. Rick J, Speck O, Maier S, et al. Optimized EPI for fMRI using a slice-
Frackowiak RS. A voxel-based morphometric study of ageing in 465 dependent template-based gradient compensation method to recover
normal adult human brains. Neuroimage 2001:14:21–36. local susceptibility-induced signal loss. Magn Reson Mater Phys 2010;
23:165–176.
14. Pfefferbaum A, Adalsteinsson E, Rohlfing T, Sullivan EV. MRI estimates of
brain iron concentration in normal aging: Comparison of field-dependent 23. Truong TK, Clymer BD, Chakeres DW, Schmalbrock P. Three-dimen-
(FDRI) and phase (SWI) methods. NeuroImage 2009;47:493–500. sional numerical simulations of susceptibility-induced magnetic field
inhomogeneities in the human head. Magn Reson Imaging 2002;20:
15. Katzman WB, Wanek L, Shepherd JA, Sellmeyer DE. Age-related 759–770.
hyperkyphosis: its causes, consequences, and management. J Orthop
Sports Phys Ther 2010;40:352–360. 24. Handwerker DA, Gazzaley A, Inglis BA, D’Esposito M. Reducing vas-
cular variability of fMRI data across aging populations using a breath-
16. Weiskopf N, Hutton C, Josephs O, Turner R, Deichmann R. Optimized holding task. Hum Brain Mapp 2007;28: 846–859.
EPI for fMRI studies of the orbitofrontal cortex: compensation of
susceptibility-induced gradients in the readout direction. MAGMA 25. Cohen MS, DuBois RM. Stability, repeatability, and the expression of
2007;20:39–49. signal magnitude in functional magnetic resonance imaging. J Magn
17. Mannfolk P, Wirestam R, Nilsson M, van Westen D, Ståhlberg F, Olsrud
J. Assessment of spatial BOLD sensitivity variations in fMRI using 26. Li TQ, Takahashi A, Wang Y, Mathews V, Glover GH. Dual-echo spiral
gradient-echo field maps. Magn Reson Imaging 2010:28:947–956. in/in acquisition method for reducing magnetic susceptibility artifacts
in blood-oxygen-level-dependent functional magnetic resonance
18. Smith SM, Jenkinson M, Woolrich MW, et al. Advances in functional imaging. Magn Reson Med 2006;55:325–334.
and structural MR image analysis and implementation as FSL. Neuro-
image 2004;23:(Suppl 1)S208–219. 27. Yablonskiy DA, Sukstanskii AL, Luo J, Wang X. Voxel spread function
method for correction of magnetic field inhomogeneity effects in quanti-
19. Cohen J. Statistical power analysis for the behavioral sciences, 2nd tative gradient-echo-based MRI. Magn Reson Med 2013;70:1283–1292.
ed. Hillsdale, NJ: Lawrence Erlbaum Associates; 1988.
28. Ngo GC, Sutton BP. R2* mapping for robust brain function detection
20. Li HJ, Hou XH, Liu HH, Yue CL, Lu GM., Zuo XN. Putting age-related in the presence of magnetic field inhomogeneity. In: 2014 36th Annual
task activation into large-scale brain networks: A meta-analysis of 114 International Conference of the IEEE Engineering in Medicine and Biol-
fMRI studies on healthy aging. Neurosci Biobehav Rev 2015;57:156–174 ogy Society (EMBC). August 26-30 2014, Chicago, p 1537–1540.

ORIGINAL RESEARCH
Quantification of Common Carotid Artery

and Descending Aorta Vessel Wall
Thickness From MR Vessel Wall Imaging
Using a Fully Automated Processing
Pipeline
Shan Gao, M.Phil,1 Ronald van ’t Klooster, PhD,1 Anne Brandts, MD, PhD,2
Stijntje D. Roes, MD, PhD,2 Reza Alizadeh Dehnavi, MD, PhD,3
Albert de Roos, MD, PhD,2 Jos J.M. Westenberg, PhD,1 and
Rob J. van der Geest, PhD1*
Purpose: To develop and evaluate a method that can fully automatically identify the vessel wall boundaries and quantify the
wall thickness for both common carotid artery (CCA) and descending aorta (DAO) from axial magnetic resonance (MR) images.
Materials and Methods: 3T MRI data acquired with T1-weighted gradient-echo black-blood imaging sequence from carotid
(39 subjects) and aorta (39 subjects) were used to develop and test the algorithm. The vessel wall segmentation was achieved
by respectively fitting a 3D cylindrical B-spline surface to the boundaries of lumen and outer wall. The tube-fitting was based
on the edge detection performed on the signal intensity (SI) profile along the surface normal. To achieve a fully automated pro-
cess, Hough Transform (HT) was developed to estimate the lumen centerline and radii for the target vessel. Using the outputs
of HT, a tube model for lumen segmentation was initialized and deformed to fit the image data. Finally, lumen segmentation
was dilated to initiate the adaptation procedure of outer wall tube. The algorithm was validated by determining: 1) its perform-
ance against manual tracing; 2) its interscan reproducibility in quantifying vessel wall thickness (VWT); 3) its capability of detect-
ing VWT difference in hypertensive patients compared with healthy controls. Statistical analysis including Bland–Altman
analysis, t-test, and sample size calculation were performed for the purpose of algorithm evaluation.
Results: The mean distance between the manual and automatically detected lumen/outer wall contours was 0.00 6
0.23/0.09 6 0.21 mm for CCA and 0.12 6 0.24/0.14 6 0.35 mm for DAO. No significant difference was observed
between the interscan VWT assessment using automated segmentation for both CCA (P 5 0.19) and DAO (P 5 0.94).
Both manual and automated segmentation detected significantly higher carotid (P 5 0.016 and P 5 0.005) and aortic
(P < 0.001 and P 5 0.021) wall thickness in the hypertensive patients.
Conclusion: A reliable and reproducible pipeline for fully automatic vessel wall quantification was developed and vali-
dated on healthy volunteers as well as patients with increased vessel wall thickness. This method holds promise for help-
ing in efficient image interpretation for large-scale cohort studies.
H igh-resolution black-blood magnetic resonance imaging

(MRI) has emerged as a tool capable of visualizing not
only the vessel lumen but also the vessel wall, allowing for
noninvasive imaging modality, MRI is well suited for serial
in vivo imaging enabling tracking disease progression and
regression. In clinical trials, vessel wall MRI has been
the detection of early stage vessel wall abnormalities.1,2 As a applied for characterizing atherosclerotic plaque in the aorta
Received Jan 27, 2016, Accepted for publication May 20, 2016.
*Address reprint requests to: R.J.G., Division of Image Processing, Department of Radiology, Leiden University Medical Center, P.O. Box 9600, 2300 RC
Leiden, The Netherlands. E-mail: R.J.van_der_Geest@lumc.nl
From the 1Division of Image Processing, Department of Radiology, Leiden University Medical Center, Leiden, Netherlands; 2Department of Radiology,
Leiden University Medical Center, Leiden, Netherlands; and 3Department of Cardiology, Leiden University Medical Center, Leiden, Netherlands

V 215

and carotid arteries for assessing the patient risk of future subjects were used, of which, 39 subjects were healthy volunteers
cardiovascular and cerebrovascular events.3–5 By quantifying without history of cardiovascular disease and 10 were patients with
the wall thickness, MRI could potentially monitor patients’ essential hypertension.
response to therapy.6–11 In addition, MRI-derived informa- The image data were further divided into subsets, denoted
tion of plaque morphology and composition can be used to A, B, and C, based on the parameter optimization, evaluation pur-
pose, and the availability of manual reference. In the remainder of
identify plaque vulnerability,12,13 and select best-suited indi-
this article, subscript “c”, “a,” and “ca” are used to denote carotid,
vidual treatment for the patient.14
aorta, carotid & aorta, respectively. Set A (referring to the subject
The quantitative assessment of vessel wall morphology
group Aca in Fig. 1) contained carotid and aortic vessel wall images F1
and plaque burden requires vessel segmentation, which is from 9 healthy volunteers (3 men and 6 women, mean age 49 6
generally performed by outlining the boundaries of lumen 14 years) and manual vessel wall delineation from one observer.
and outer vessel wall manually. However, manual contour This set was used for the development of the segmentation algo-
tracing is tedious and subject to observer variability. Accord- rithm. Set B (referring to the subject groups Bc and Ba in Fig. 1)
ingly, significant efforts have been placed on the develop- contained images from 20 healthy volunteers (set Bc: 7 men and 3
ment of efficient computer-aided methods for vessel wall women, mean age 57 6 22 years; set Ba: 7 men and 3 women,
extraction. In previous work, approaches for automatic seg- mean age 23 6 4 years) and intra-/interobserver delineations. This
mentation of carotid and aorta employ classical deformable set was used for the evaluation of the segmentation performance,
models (eg, active contour model,15 active shape model,16 the reproducibility and the intra-/interobserver variability. Set C
discrete dynamic contour model,17 ellipse model,18 cylinder (referring to the subject group Cca in Fig. 1) contained image data
from 10 patients (3 men and 7 women, mean age 49 6 14 years)
model;19,20 deformable model implemented using level-
with essential hypertension (systolic blood pressure 5 149 6 19 mm
set;21–23 graph-cuts;24,25 or intensity probability density
Hg, diastolic blood pressure 5 88 6 13 mm Hg), and image data from
function matching.26 In these approaches, some reported 10 age-matched and sex-matched healthy controls (systolic blood pres-
methods19,22,23 demonstrated only the segmentation of sure 5 124 6 12 mm Hg, diastolic blood pressure 5 80 6 8 mm Hg).
inner border, while other methods15–18,20,21,24–26 are able to All patients were on the antihypertensive treatment at the time of MRI.
segment both inner and outer vessel walls. This set contained vessel wall contours traced by one observer and was
All the above approaches require some amount of user used for assessing the ability of the automated algorithm to differentiate
interaction as an input to the automated process. In general, patients from healthy subjects based on automated quantification of
the requirements range from one interaction for every VWT. The included image data have been used in previously published
slice15–19,24 to one single interaction to start the complete seg- studies, in which, Aca and Cca were used in a study that evaluated the
mentation.20–23,25,26 Additionally, the methods presented are association between arterial wall thickness and stiffness;27 Bc in a study
optimized to a specific type of vessel (eg, carotid15–20,23,25,26 or that assessed the reproducibility of carotid MRI;28 Ba in a study that
investigated the feasibility of aortic vessel wall imaging using a 3D free-
aorta21,22,24) and no generally applicable method has been
breathing MRI technique.29 This study received a new Institutional
described. While semiautomated segmentation methods are
Review Board (IRB) approval to analyze the image data collected in the
generally less time-consuming and more reproducible compared
former research projects. The overview of the data is shown in Fig. 1.
to fully manual delineation, they still remain labor-intensive
MRI was performed on a 3T scanner (Achieva; Philips Medi-
and therefore are not optimally suited for cohort studies where cal Systems, Best, The Netherlands). For carotid MRI examination,
large volumes of data need to be processed. Therefore, a fully cross-sectional vessel wall images of the left carotid were acquired
automated analysis which is applicable to various anatomical with a 1.6-cm thick acquisition stack positioned perpendicular to
regions is highly desirable for application in clinical trials in the course of the common carotid artery starting at the level of the
which large cohorts are scanned at multiple timepoints and the flow divider covering the CCA (6 to 16 mm proximal to the flow
detection of temporal changes is of importance. divider) and the carotid bulb (0 to 6 mm proximal to the flow
The aim of the current study, therefore, was to develop a divider). An ECG-triggered 2D dual IR spoiled segmented k-space
method that can fully automatically segment the vessel wall fast field echo (FFE) sequence was used with TR/TE 5 12/3.6
and calculate the wall thickness for both common carotid msec, flip angle (FA) 5 458, field of view (FOV) 5 140 3 140
mm2, number of slice 5 8, and reconstructed voxel size 5 0.27 3
artery (CCA) and descending aorta (DAO) in MR images.
0.27 3 2 mm3. Details of the employed carotid vessel wall imag-
The current work is based on the previously described tube-
ing protocol have been described previously.28 For aortic MRI
fitting method,20 and the main contribution and novelty of
examination, images were acquired with a 2-cm thick acquisition
our work is: 1) to make the segmentation process fully auto- stack positioned perpendicular to the aorta at the level of the mar-
mated, and 2) to apply the method to aortic segmentation. gin between the eighth thoracic vertebra and the intervertebral disc
as described previously.29 Briefly, an ECG-triggered 3D dual IR
Materials and Methods segmented k-space turbo field echo (TFE) imaging sequence was
Image Data used, and the imaging parameters were TR/TE 5 4.9/2.5 msec,
For the development and evaluation of the automated segmenta- FA 5 208, FOV 5 270 3 203 3 4 mm3, number of slices 5 10,
tion algorithm, carotid and aortic MR vessel wall images from 49 and reconstructed voxel size 5 0.53 3 0.53 3 2 mm3. The

Gao et al.: Quantification of CCA and DAO
C
O
L
O
R
FIGURE 1: Data composition in this study. In this figure, subscript “c” denotes carotid, “a” denotes aorta, “ca” denotes carotid &
aorta.
carotid/aortic MRI scans of the healthy volunteers in set B were proposed model-based segmentation algorithm includes three
repeated to test the reproducibility of the imaging and subsequent steps which are all performed automatically without requiring any
quantitative analysis. After the first scan, the subjects were removed user-interaction: 1) vessel model initialization using modified
from the scanner and the coil was removed; then the subjects were Hough Transform (HT); 2) lumen boundary detection and
repositioned and the second scan was performed. Standardized refinement using tube-fitting and dynamic programming; 3) outer
planning protocols were used to ensure that the same segment of wall segmentation using a similar approach as for lumen. The
carotid/aorta were imaged during the scan–rescan MRI above steps of the automated analysis pipeline are described in
examinations. more detail below.
Manual Delineation
Manual tracing of the contours of lumen and outer wall for the STEP1: DEFORMABLE MODEL INITIALIZATION. When using a
CCA and DAO was performed using VesselMASS software (Leiden deformable model, its shape, location, orientation and size need to
University Medical Center, Leiden, The Netherlands).18 For set A be initialized to be similar to that of the object to be segmented
and set C, image analysis was performed by one observer (A.B.). from the image. Our initialization of the vessel model is based on
For set B, images of first scan and second scan were analyzed by the following assumptions.
observer 1 (R.A.D. analyzed set Bc and S.R. analyzed set Ba); to
- According to the anatomical property of CCA and DAO, we
assess the interobserver variability, first scan images were also ana-
can assume that their centerlines are straight. Additionally, we
lyzed by observer 2 (S.G. analyzed set Bc and J.W. analyzed set
can assume that there is little deviation in cross-sectional size of
Ba); to assess the intraobserver variability, observer 1 reanalyzed the
the vessel in adjacent axial slices since no stenosis has occurred
first scan images (second read) 2 months after the initial review
at the CCA and DAO in our study population. Therefore, the
(first read). All manual analysis were performed under the supervi-
surface of tubular structure CCA and DAO can be initialized
sion of experts J.W. (with 20 years of experience in cardiovascular
MRI) and R.J.vdG. (with 20 years of experience in cardiovascular using a uniform cylinder model.
MR image analysis). - As the transverse slices are planned perpendicular to the axis of
the target vessel during image acquisition, we can assume that
Automatic Segmentation the angle (a) between the vessel axis and the Z-axis of the MR
Based on the imaging protocol, the fully automated processing scanner is relatively small, for instance a < 108. Under this
pipeline was able to select the slices that covered the vessel seg- assumption, the appearance of CCA/DAO in the cross-sectional
ment of CCA (slice 1st5th of the carotid data) or DAO (slice 1st images approximates a circular shape, and the position of
10th of the aortic data) as the region of interest, in which the CCA/DAO changes little in subsequent slices. Therefore, the
subsequent automated segmentation will be performed. In prac- Circular Hough Transform (CHT) can be first applied to detect
tice, the most caudal slice (slice 1) of aortic data was excluded in circular objects in each 2D image. Subsequently, the target ves-
the automated segmentation because of poor image quality. The sel can be located by taking the median of CHT results from
January 2017 217

C
O
L
O
R
FIGURE 2: Illustration of the 3D accumulator and the voting procedure of the Circular Hough Transform. In a MR image, each pixel
(x,y) votes for two circles of radius rk, and the vote accumulation will be the cell value of the 3D accumulator at the k th floor.
all axial slices of one artery, as the peak value in the median c 5x2r cosðhÞ
x1 k
CHT corresponds to the center of CCA/DAO. (2)
cy1 5y2rk sinðhÞ
- In black blood vessel wall MR images, the boundaries of the
inner and outer wall have opposite gradient direction. There- c 5x2r cosðp1hÞ
x2 k
fore, we assume that the influence of outer wall edges can be (3)
excluded in the lumen “circle” detection by incorporating the cy2 5y2rk sinðp1hÞ
image gradient direction into the HT voting procedure. Such
an oriented-edge based CHT (OECHT) was developed in the Where h is the gradient direction at position ðx; yÞ, and ðcx1 ; cy1 Þ,
current study for the lumen centerline and radii estimation in ðcx2 ; cy2 Þ are two potential circle centers calculated with radius rk ,
order to initialize the lumen surface. k51; 2; 3 . . . K denotes the k th radius candidate.
In the experiment, the CHT calculation has two steps: radius
The implementation details of initialization are presented in discretization and center accumulation. To determine the appropri-
the following. ate radius search range for the CHT to detect the desired anatomi-
1) Model location detection cal structure, the mean and standard deviation (SD) of the lumen
For the fully automatic process, localization of the cylinder radius (RLumen) and outer wall radius (Router) were derived from
model is the first step of the segmentation chain. To this end, the training set. To obtain a more robust vessel localization, both
CHT was applied on the cross-sectional images to find the position the contributions of lumen and outer wall “circles” were taken into
of the target vessel. account when estimating the center for the target vessel, which
A circle is characterized by three parameters according to the resulted in a radius range of: [mean of RLumen – 3SD of RLumen,
definition in the Cartesian coordinate system: mean of Router 1 3SD of Router ]. A step size of 1 pixel was chosen
for radius discretization. For a radius candidate rk , every pixel ðx; yÞ
in the MR image was scanned and one vote was added to the
ðx 2 cx Þ2 1ðy 2 cy Þ2 5 r2 (1)
Hough space at coordinates ðcx1 ; cy1 ; rk Þ and ðcx2 ; cy2 ; rk Þ, with
ðcx1 ; cy1 Þ and ðcx2 ; cy2 Þ located along and opposite to the gradi-
Therefore, the parameter space of the CHT is three-dimensional, ent direction of pixel ðx; yÞ at a distance rk (Fig. 2). After all pix-
and each coordinate ðcx ; cy ; rÞ in the 3D Hough space represents a els in the MR image of one slice were scanned, vote
circle candidate (Circleðcx; cy; rÞ ) with center ðcx ; cy Þ and radius r. For accumulation was performed at each coordinate of the Hough
circle detection, a 3D accumulator ACHT n with K floors is defined space to compute the cell value ACHT ðcx ; cy ; rk Þ at the kth floor of
n
F2 over the CHT parameter space (Fig. 2), where K is the number of the 3D accumulator (Fig. 2). The procedure of scanning and
radius candidates in the search range. The value of each accumula- voting was repeated for all radius candidates within the defined
tor cell ACHT
n ðcx ; cy ; rk Þ is computed by summing the votes from search range. To enable the CHT to detect imperfect circles, the
those image pixels which support Circleðcx; cy; rk Þ to be present in the final CHT response map for one image slice is calculated by:
cross-sectional MR image. According to Eq. (1), image pixel ðx; yÞ
supports two circles as it can be potentially on the boundary of
X
K
circle 1 or circle 2 (Fig. 2), and their centers C1 and C2 can be cal- MnCHT ðcx ; cy Þ5Gr ACHT
n ðcx ; cy ; rk Þ (4)
culated by Eqs. (2) and (3). k51

C
O
L
O
R
FIGURE 3: (a,d) vessel wall image; (b,e) CHT response map of a single MR image; (c,f) median CHT response map calculated from
all CCA/DAO slices of one subject.
where n denotes the nth slice of an artery segment, “” denotes convo- implementation of the OECHT can refer to that of the regular CHT;
lution, Gr is a 2D Gaussian smoothing kernel of width r, and a value however, there were two main differences: 1) only the candidate edge
of r53 pixels was used in this study. The value in the CHT response pixels were scanned and participated in the voting; and 2) vote was
F3 map (Fig. 3b,e) indicates the likelihood of the current location to be the added to only one potential circle center which is located opposite to
center of a circular structure. To further remove those CHT peaks pro- the gradient direction of pixel ðx; yÞ when pixel ðx; yÞ was scanned. In
duced by nontarget objects, we calculated the median of CHT response the black blood vessel wall image, the gradient direction at the lumen
maps from all available slices from the current artery segment (Fig. 3c,f). edge is outward, while the gradient direction at the outer wall edge is
Finally, a number of center candidates were obtained by exploring the inward (Fig. 4c). Therefore, only the lumen edge pixels contribute to
highest local maxima in the median CHT response map. These candi- the estimation of the lumen center and radius, while the influence of
dates correspond to the centers of the target vessel and other circular ana- the outer wall edge pixels can be ignored (Fig. 4d). The location with
tomical structures. According to the MRI protocol used in this study, higher value in the OECHT response map MnOECHT would have a
the candidate closest to the center of the field of view was selected as the higher probability to be the lumen center for the target vessel at the
center of the vessel of interest, which defined the approximate location current slice (Fig. 4e).
of the cylinder model. Under the assumption that the axis of the target vessel is
2) Model orientation and size estimation straight and the imaging plane is nearly perpendicular to the vessel
After localizing the vessel of interest, the next step is to esti- axis, we modeled the lumen centerline as a 3D straight line and the
mate the orientation and size of the cylinder model, which will be centerline searching was performed in a volume of interest (VOI).
used for later initialization of the vessel lumen surface within an This VOI is defined based on the orientation of the target vessel rel-
imaging volume. For this purpose, OECHT was calculated in each ative to the image slices. More specifically, the VOI allows the angle
axial slice, followed by the lumen centerline and radii searching (a) between the candidate centerline and the image normal ranging
through the image stack of an artery. from –108 to 1108. If there are N axial images for a vessel of inter-
OECHTwas applied to a circular region of interest (CROI) cen- est, a set of N points (P1 ; P2 ; . . . ; PN ) on the candidate centerline
F4 tered at the target vessel (Fig. 4a). The radius of the CROI was set to 8 will be the detected lumen center of the target vessel in a series of
mm for the CCA and 26 mm for the DAO. The accumulator of the 2D images, where each point Pn is represented by x and y coordi-
OECHT AOECHT n was calculated from the oriented edge map. To nate. The centerline detection problem is therefore reformulated as
extract the edges and compute the oriented edge map, the Sobel edge finding the optimal straight line by maximizing a score function:
detector was used to calculate the gradient magnitude G and direction
h for each pixel within the CROI, and the pixels with G greater than a X
N
threshold (80 percentile of the histogram of gradient magnitudes S5 MnOECHT ðPn Þ (5)
within the CROI) were selected as candidate edge pixels (Fig. 4b). The n51
January 2017 219

C
O
L
O
R
FIGURE 4: (a) CROI (yellow); (b) candidate edge pixels (white); (c) gradient direction of lumen edge (red arrow) and outer vessel
wall edge (green arrow); (d) voting scheme of OECHT; (e) OECHT response map of one cross-sectional image.
Such lumen centerline was used to define the central axis of the control points to update the surface in each iteration, so that
the cylinder model. the surface was moving toward the lumen boundary. The parame-
For an axial slice, the radii rLumen of the target lumen can be ters of tube-fitting were selected in the training phase. Finally, con-
estimated once the OECHT accumulator AOECHT n is calculated and tour refinement was performed on the original image within each
the lumen center Pn is located. The best lumen radius at the nth slice using a minimum cost approach, which was based on
slice is given by: dynamic programming.31
rLumen 5arg max AOECHT

n ðPn ; rk Þ (6) STEP 3: OUTER WALL CONTOUR DETECTION. In each slice,
rk
the lumen contour was dilated by 0.5 to 2.6 mm in steps of 0.05
The radius of the cylinder model was obtained by taking the mm to find the initial outer wall boundary. For each step, the aver-
median of lumen radii from all axial slices of a target vessel. age gradient strength under the dilated contour was calculated, tak-
ing into account the edge direction from bright to dark. The
STEP 2: LUMEN CONTOUR DETECTION. Having the lumen dilated contour with the highest gradient strength was selected as
cylinder defined, lumen contours can be detected by automatically the initial outer wall boundary for the current slice. The outer ves-
fitting a deformable 3D cylindrical NURBS surface 30 to the mul- sel wall was segmented using a tube-fitting method, which was
tislice vessel wall images. The tube-fitting algorithm has been similar to what was used for lumen segmentation, with the excep-
described previously by van ’t Klooster et al.20. In brief, the proce- tion that the tube surface was fitted to the bright-inside and dark-
dure of lumen segmentation can be summarized in the following: outside edges.
1) Initializing the NURBS surface: for each slice, a ring with con-
trol points was created based on the estimated lumen center and Global and Regional Validation Metrics
radius. 2) Calculating the image force in each iteration. The image To evaluate the effectiveness of the HT-based initialization method,
force, which will pull the surface towards the lumen boundary, was the difference between the HT and manual detected lumen radii
computed at each surface point. A signal intensity (SI) profile Dr, and the distance between the lumen center points DC detected
along the surface normal was used for lumen boundary searching. by both methods were assessed for each slice from the first scan of
For black-blood vessel wall image, the position with the strongest the 10 healthy volunteers (subject group Bc and Ba). The center of
dark to bright transition from inside to outside in the SI profile gravity of the manually delineated lumen contour was used as the
corresponds to the expected lumen boundary. 3) Calculating the manually detected lumen center, and the manually detected radius
movement of the control points based on the image force and the was taken as the average distance between the lumen center and all
influence of the surface points on the control points. 4) Moving the contour points. A positive Dr indicates that the lumen radii

C
O
L
O
R
FIGURE 5: (a, c) Four equiangular sectors of CCA/DAO wall for regional validation. (b, d) Bar-plot showing the mean and standard
deviation of the distance between the manual and automatically detected contours.
estimated by HT is larger than that obtained by manual Statistical Analysis

delineation. To evaluate the performance and reproducibility of the automated
To directly compare the observer traced contour and the segmentation algorithm in the assessment of vessel wall thickness, for
automatically generated contour, the mean contour distance each slice, 1) Bland–Altman analysis was performed to investigate the
(MCD) was calculated by taking the average of the distances differences between paired measurements of VWT; 2) a two-tailed
between corresponding contour points. An MCD with a positive paired t-test was used to determine the statistical significance of the
value indicates that automatically detected contour is larger than observed differences, P < 0.05 was considered statistically significant.
manual traced contour. To assess the agreement between manual To compare the artery wall thickness between the healthy
and automated segmentations, between observers, between first volunteers and the patients with hypertension, an unpaired t-test
read and second read of one observer, the degree of similarity was performed.
(DoS) between two contours was calculated as the percentage of Finally, a power analysis was performed to calculate the sam-
locations where the absolute distance between the corresponding ple size required for the automated segmentation method to detect
points on both contours does not exceed a pixel.32
certain VWT changes in the longitudinal studies based on its inter-
As vessel wall thickness (VWT) is more relevant in clinical scan reproducibility. The number of subjects required to detect an
practice, VWT derived from the detected vessel wall contours was effect depends on the power level, level of significance, and the
also used to compare the segmentations between manual and auto-
effect size. For this the following formula was used:
matic approaches and between scan sessions. For each slice, the
VWT was calculated by taking the median length of 100 chords mean difference over time period
connecting the lumen and outer wall boundaries.20 These 100 effect size5
SDpd
chords were sampled equidistantly and perpendicular to the center-
line between the boundaries of inner and outer wall. where, SDpd is the standard deviation of the paired differences between
Furthermore, the vessel wall was divided into four sectors, the interscan VWT. In the current study, power analysis was performed
each spanning 90 , to provide a regional evaluation of boundary using the software G*Power (version 3.0)33 based on a paired two-tailed
detection. Based on the anatomy, for the CCA, the sectors 1, 2, 3, t-test with power level of 80% and significance level of 0.05.
and 4 are respectively located in the anterolateral, anteromedial,
F5 posteromedial, and posterolateral region (Fig. 5a). For the DAO,
the sectors 1, 2, 3, and 4 are defined as anterolateral, posterolat-
Results
eral, posteromedial, and anteromedial, respectively (Fig. 5c). Subse- Tube Initialization Accuracy
quently, the MCD, DoS, and VWT were also calculated for each In the initialization, the radius search range for the HT was
sector. accordingly set for the CCA (2.35.3 mm) and the DAO
January 2017 221

TABLE 1. Distance Between the Manual and Automatically Detected Contours
Artery Region of analysis Lumen contour Outer wall contour
CCA Sector 1 0.04 6 0.12 20.05 6 0.15

Sector 2 0.01 6 0.22 20.07 6 0.21
Sector 3 20.05 6 0.32 20.09 6 0.28
Sector 4 20.02 6 0.20 20.18 6 0.14
Overall 0.00 6 0.23 20.09 6 0.21
DAO Sector 1 0.05 6 0.27 20.05 6 0.41
Sector 2 0.13 6 0.23 20.28 6 0.17
Sector 3 0.16 6 0.23 20.15 6 0.29
Sector 4 0.08 6 0.24 20.07 6 0.43
Overall 0.12 6 0.24 20.14 6 0.35
Mean 6 Standard Deviation in mm, 10 healthy volunteers. Pixel resolution: 0.27 mm/pixel for CCA images, 0.53 mm/pixel for
DAO image.
(5.515.4 mm). For carotid scans having a pixel size of manual-automated agreements were observed at the poste-
0.27mm/pixel, in 95% of cases, the DC ranged from 0 to rior part of the CCA (sectors 3 and 4, DoS 5 80.3% and
1.1 pixel (00.3 mm) and the Dr ranged from –0.9 to 71.3%), and at the anterior part of the aorta (sectors 1 and
10.4 pixel (–0.20.1 mm). For aorta scans having a pixel 4, DoS 5 86% and 84.7%).
size of 0.53 mm/pixel, in 95% of cases, the DC ranged Bland–Altman analysis for comparing the assessment
from 0 to 1.4 pixel (00.7 mm) and the Dr ranged from – of vessel wall thickness between manual and automated seg-
0.4 to 10.8 pixel (-0.20.4 mm). mentations, and between inter- and intraobserver are pre-
sented in Table 2. For CCA, the VWT was 1.05 6 0.23 T2
Comparisons Between Manual and Automated
mm by manual delineation and was 0.93 6 0.21 mm by
Segmentations, Between Inter- and Intraobserver
Mean distances in mm between the manual and automati- automated segmentation; by comparison, the VWT was
cally detected vessel wall contours for the CCA and DAO slightly, but significantly, underestimated by the automated
T1 are summarized in Table 1 and further illustrated as bar- segmentation; however, the difference between observer 1
plots in Fig. 5. The mean differences between the manually and observer 2 in the VWT assessment was also statistically
and automatically detected contours were below 1 pixel, significant. In addition, the interobserver bias was signifi-
with the mean difference being smaller for the lumen con- cantly higher than the manual-automated bias (0.20 mm vs.
tours compared to the outer wall contours. Compared to 0.12 mm, P < 0.001). For DAO, the VWT was 1.44 6
manual delineation, automated detection slightly underesti- 0.12 mm by manual and 1.14 6 0.15 mm by automatic
mated the outer wall contours for both CCA (0.09 mm) approach. The interobserver variation was smaller than the
and DAO (0.14 mm), and slightly overestimated the lumen manual-automated variation as both the bias and the stand-
contours for DAO (0.12 mm). ard deviation (SD) of the bias between the interobserver
Results of similarity between contours obtained with were smaller than those between the manual-automated
different segmentation approaches, observers, and reads are (–0.05 6 0.13 mm vs. 0.30 6 0.16 mm), although both
F6 presented in Fig. 6. For the lumen contour detection in the biases were statistically significant. Compared to other sec-
CCA and DAO, both the sector-wise and slice-wise average tors, sector 4 of CCA and sector 2 of DAO showed the
DoS between the manual and automatic approach were highest manual-automated segmentation variability in the
greater than 95%. In addition, the agreement between the regional vessel wall quantification.
manual and automated segmentation was similar to the
inter- and intraobserver agreement. For the outer wall con- Comparison Between Scan and Rescan
tour detection, the comparison of automatic against manual Table 2 also provides the results of Bland–Altman analysis
approach resulted in the slice-wise average DoS of 82.8% for the interscan vessel wall thickness assessment. No signifi-
for CCA and 89.2% for DAO. The sector-wise average cant difference of overall VWT was observed between scan
DoS ranged from 71.3% to 91.3% for CCA and 84.7% to sessions using automated segmentation for both CCA (P 5
94.7% for DAO. Compared to other sectors, the lower 0.19) and DAO (P 5 0.94). In contrast, the manual

C
O
L
O
R
FIGURE 6: Illustration of four vessel wall sectors and sector-/slice-wise degree of similarity results of the comparisons between
manual and automatic segmentations, between inter- and intraobserver for the CCA and the DAO. (a, b, c, d) Manual and auto-
mated detected lumen (red, yellow) and outer wall (green, blue) contours. SCM: sternocleidomastoid muscle. LCM: longus capitis
muscle. IJV: internal jugular vein. LA: left atrium. CT: connective tissue. VB: vertebral body. (e, f, g, h) Mean 6 95%CI of DoS for
lumen and outer wall contour.
delineation showed a significant variation in the interscan variation (bias 5 0.08 mm, P < 0.001 and bias 5 –0.05
overall VWT quantification for DAO (P 5 0.02). The mm, P 5 0.01) for the automated VWT assessment.
sector-wise measurement revealed that sector 2 of CCA and Scan–rescan reproducibility influences the number of
sector 3 of DAO showed a small but significant interscan subjects needed in longitudinal studies. Table 3 lists the T3
January 2017 223

224
TABLE 2. Bland-Altman Analysis of Vessel Wall Thickness Assessment in 10 Healthy Volunteers
Artery Region
of analysis
Bland-Altman analysis of VWT assessment (Bias 6 Standard Deviation in mm)
Between intersegmentation methods, inter- and intraobserver Between scan and rescan sessions
Manual vs. Observer 1 vs. Observer 1a vs. Automated Manual
automatic observer 2 observer 1b segmentation segmentation
CCA Sector 1 0.09 6 0.16 0.20 6 0.16 20.00 6 0.13 0.00 6 0.19 20.02 6 0.14
(P 5 0.001) (P < 0.001) (P 5 0.99) (P 5 0.46) (P 5 0.43)
Sector 2 0.12 6 0.18 0.18 6 0.14 0.02 6 0.12 0.08 6 0.13 0.00 6 0.12
(P < 0.001) (P < 0.001) (P 5 0.24) (P < 0.001) (P 5 0.90)
Sector 3 0.03 6 0.21 0.17 6 0.13 20.00 6 0.12 0.06 6 0.18 0.01 6 0.11
(P 5 0.19) (P < 0.001) (P 5 0.88) (P 5 0.07) (P 5 0.23)
Sector 4 0.19 6 0.19 0.26 6 0.12 20.00 6 0.14 0.01 6 0.17 0.00 6 0.11
(P < 0.001) (P < 0.001) (P 5 0.97) (P 5 0.31) (P 5 0.86)
Overall 0.12 6 0.15 0.20 6 0.11 0.01 6 0.10 0.03 6 0.13 20.00 6 0.09
(P < 0.001) (P < 0.001) (P 5 0.43) (P 5 0.19) (P 5 0.91)
DAO Sector 1 0.08 6 0.40 20.11 6 0.16 20.06 6 0.18 0.04 6 0.37 20.05 6 0.21
(P 5 0.02) (P < 0.001) (P 5 0.005) (P 5 0.38) (P 5 0.06)
Sector 2 0.43 6 0.20 20.01 6 0.13 20.05 6 0.12 0.00 6 0.24 20.06 6 0.19
(P < 0.001) (P 5 0.47) (P < 0.001) (P 5 0.14) (P 5 0.01)
Sector 3 0.35 6 0.24 20.02 6 0.18 20.04 6 0.13 20.05 6 0.16 20.01 6 0.16
(P < 0.001) (P 5 0.39) (P 5 0.008) (P 5 0.01) (P 5 0.44)
Sector 4 0.21 6 0.26 20.05 6 0.19 20.07 6 0.11 20.05 6 0.35 20.02 6 0.17
(P < 0.001) (P 5 0.02) (P < 0.001) (P 5 0.66) (P 5 0.34)

Overall 0.30 6 0.16 20.05 6 0.13 20.06 6 0.09 20.02 6 0.16 20.04 6 0.13
(P < 0.001) (P < 0.001) (P < 0.001) (P 5 0.94) (P 5 0.02)
Stage:

Volume 45, No. 1
Page: 224
TABLE 3. Sample Size Calculation Based on Interscan Reproducibility Using Automated Segmentation
Artery of Mean SD of Detected changes (mean VWT difference over time period)
interest VWT (mm) inter-scan
measured VWT (mm) 0.05 mm 0.10 mm 0.30 mm 0.60 mm
at scan 1 difference (CCA:5%, (CCA:11%, (CCA:32%, (CCA:65%,
DAO: 4%) DAO: 9%) DAO:26%) DAO:53%)
Required number of subjects

CCA 0.93 0.13 56 16 4 3
DAO 1.14 0.16 83 23 5 3
required sample sizes calculated for fully automatic segmen- DAO, manually detected VWT difference was more sub-
tation for detecting changes in overall VWT. The excellent stantial (P < 0.001 vs. P 5 0.021).
reproducibility of the pipeline developed in this study ena-
bles detection of 0.1 mm changes in CCA/DAO wall thick- Discussion
ness during follow-up using a small sample size of 16/23 This study proposed a fully automated 3D segmentation
patients with a 80% power at the significant level of 0.05. method allowing one to identify lumen and outer boundaries
for both common carotid artery and descending aorta in vessel
wall MRI. The main advantage of the proposed technique is
Comparison Between Patients and Healthy that no user interaction is required and therefore does not
Controls limit the amount of data to be analyzed. To evaluate the feasi-
Comparative results of overall VWT between hypertensive bility of the method to be used as an alternative to manual
patients and age- and sex-matched healthy control subjects contouring for vessel wall morphology assessment in the future
F7 are shown in Fig. 7. Both the manual and automated seg- cohort studies, the reliability of the proposed algorithm has
mentation revealed significantly higher CCA and DAO wall been validated comprehensively, and our result demonstrates:
thickness in the hypertensive patients. Compared to manual 1) good similarity between the automatically detected and
contour delineation, the CCA wall thickness difference manual delineated vessel wall contours, although slight under-
between these two groups was more profound (P 5 0.005 estimation of VWT provided by automated segmentation; 2)
vs. P 5 0.016) when using the automatic approach. For excellent scan–rescan reproducibility for VWT measurement
C
O
L
O
R
FIGURE 7: CCA and DAO wall thickness in patients and in healthy control subjects. The horizontal black line that pass through
each cluster indicates the mean. * indicate statistical significant difference with P < 0.05, ** with P < 0.01, and *** with P < 0.001.
January 2017 225

using automated segmentation; 3) robust capability of auto- the DoS of sectors 1 and 2. For DAO, the lowest DoS was
mated VWT estimation to detect differences in CCA and found in sectors 1 and 4, corresponding to the anterior part
DAO wall morphology in patient with hypertension compared of the aorta. In this region the outer wall boundary cannot
to healthy controls. always be sufficiently differentiated from the adjacent ana-
In this study, 3D tube fitting for the lumen and outer tomical structures with the same intensity, such as connec-
contours detection was initialized by using the lumen cen- tive tissue, left atrium, and left pulmonary vein. At the
terline and radii estimated by HT, making the whole vessel anterior region, the observer traces the outer boundary based
segmentation process fully automated. This means no man- on the prior knowledge of the healthy vessel wall thickness;
ual corrections were applied to the automatically detected while the automated algorithm keeps the shape of the outer
centers, radii, and contours. Therefore, good automatic ini- tube as the initialization status because there is no force to
tialization of the lumen tube is very important for the suc- guide the segmentation. Therefore, sectors 1 and 4 demon-
cess of the final segmentation. The results of this study strated good interscan reproducibility despite the slightly
demonstrated that our modified HT provided reliable center lower manual-automated agreement.
localization and radii estimation for the lumen of target ves- As revealed by the regional analysis, in certain regions
sel. In addition, our 3D tube-fitting approach was able to the outer wall contour could not be reliably detected no
achieve good vessel wall boundaries delineation. matter whether using manual or automated segmentation
Manual outlining of the vessel contours requires expe- methods because the vessel wall is not possible to be differ-
rience, and is tedious and prone to observer variability. entiated from adjacent tissues in these locations, which will
Instead, a fully automated segmentation technique, which is lead to inaccurate quantification of vessel wall morphology.
fully independent of the human observer, is more practical Therefore, regional wall thickness/volume, measured only in
in large-scale and longitudinal studies. As interscan reprodu- those sectors with reliable image information tends to be
cibility is key to the clinical applications where repeated more correct and needs to be considered in future vessel
examinations are needed, the excellent reproducibility pre- wall studies.
sented in this study enables the proposed algorithm to be Several studies have reported automated vessel wall
applied in future clinical trials where serial examinations are segmentation approaches for the CCA and DAO, and our
required to evaluate the effect of the medical treatment on results are comparable to previous publications. Using the
the changes in the vessel wall thickness. For instance, it DoS to compare the manual and automated contouring
requires only a small sample size of 16 subjects in each results, van ‘t Klooster et al 20 reported that, for CCA,
group to detect 15% regression in carotid VWT, which can 96.2% and 75.3% of the lumen and outer contours were
be expected after 18 months lipid-lowering therapy,34 with successfully segmented. In our study, 95.4% of CCA lumen
80% power. contours were successfully segmented, while this was 82.7%
Global DoS and VWT measurement has usually been for the outer contours. In addition, the results also show
used to compare the manual and automated approaches for that the interscan reproducibility for VWT measurements
vessel wall segmentation;20,35 however, local segmentation obtained from our method was similar to that of the
errors might be masked when using these averaged parame- method developed by van ‘t Klooster et al (interscan VWT
ters. Therefore, both global and regional analyses were difference: 0.03 6 0.13 mm vs. –0.02 6 0.06 mm). Taken
applied to evaluate the segmentation performance of the together, these findings prove that increasing segmentation
proposed algorithm against the observer contouring. For performance while preserving segmentation reproducibility
CCA, sector 4 showed the lowest DoS due to the property can be achieved by the initialization without user input or
of the algorithm to fit the tube surface to the bright-inside the use of an additional MRA image series. For the DAO,
dark-outside edge with maximum image gradient while the compared with the study by Adame et al,35 in which,
observer traces the boundary at the carotid wall to longus Bland–Altman analysis was used to compare the automated
capitis muscle interface based on visual perception. The true calculated VWT with the manual reference, our method
vessel wall boundary is therefore not necessarily at the loca- showed larger bias but smaller standard deviation (manual-
tion with strongest edge. It is worth noting that the visual automated VWT difference: 0.30 6 0.16 mm vs. 0.11 6
determination for the separation/interface between the 0.40 mm). This finding indicates that the developed auto-
carotid wall and the longus capitis muscle could be difficult mated method has lower variability and therefore may be
in this sector, as the two structures can be very close to each better suited for detection of temporal changes in vessel wall
other, in addition to their similar signal intensities, resulting thickness.
in the lowest interobserver agreement for sector 4. In sector There are several potential limitations to our study.
3 of CCA, the outer wall edge is often weak and thus the First, slice-alignment between the scan and rescan MRI data
outer tube surface can be drawn to the lumen edge of the is dependent on manual planning of the image stack by the
internal jugular vein, explaining the lower DoS compared to scanner operator, which may influence the reproducibility of

the fully automated vessel wall segmentation. Nevertheless,

the reproducibility obtained by our automated contour Acknowledgments
detection was as good as manual delineation. Adding inter- Contract grant sponsor: Dutch Technology Foundation
scan image registration into the automatic analysis pipeline STW, which is the applied science division of NWO as part
could potentially result in improved reproducibility. Second, of project CARISMA 11631 “Automated carotid plaque
we applied the algorithm in the segmentation of straight analysis from quantitative MRI.”
vessel segments alone, and it might be argued that our
method is not suitable for tortuous vessels. In fact, our seg-
mentation pipeline can be adapted to vessels with moderate
curvature by fitting the vessel centerline with a polynomial References
1. Yuan C, Kerwin WS, Yarnykh VL, et al. MRI of atherosclerosis in clinical
curve and initializing the 3D lumen surface as a curved
trials. NMR Biomed 2006;19:636–654.
tube. However, for highly tortuous vessels, accurate VWT
2. van der Kolk AG, Zwanenburg JJ, Brundel M, et al. Intracranial vessel
assessment would require 3D imaging with isotropic spatial wall imaging at 7.0-T MRI. Stroke 2011;42:2478–2484.
resolution since the vessel orientation changes rapidly and 3. Takaya N, Yuan C, Chu B, et al. Association between carotid plaque
partial volume effect is likely to appear when using a 2D characteristics and subsequent ischemic cerebrovascular events: a pro-
spective assessment with MRI—initial results. Stroke 2006;37:818–823.
acquisition. Third, data from patients with severe stenosis or
eccentric vessel wall thickening were not available in this 4. Maroules CD, Rosero E, Ayers C, Peshock RM, Khera A. Abdominal
aortic atherosclerosis at MRI is associated with cardiovascular events:
study. Nevertheless, as we can learn from the tube-fitting the Dallas heart study. Radiology 2013;269:84–91.
process, the deformable model would be able to drive the 5. Kwee RM, van Oostenbrugge RJ, Mess WH, et al. MRI of carotid ath-
initial surface to the desired boundaries if these boundaries erosclerosis to identify TIA and stroke patients who are at risk of a
recurrence. J Magn Reson Imaging 2013;37:1189–1194.
are within the edge detection range. Moreover, the balance
6. Yonemura A, Momiyama Y, Fayad ZA, et al. Effect of lipid-lowering
between the constraint and the flexibility of the tube model
therapy with atorvastatin on atherosclerotic aortic plaques detected
is optimized in the training phase, which allows the algo- by noninvasive magnetic resonance imaging. J Am Coll Cardiol 2005;
rithm to outline any regular and irregular vessel wall boun- 45:733–742.
daries caused by stenosis or eccentric wall thickening. It is 7. Zhao XQ, Yuan C, Hatsukami TS, et al. Effects of prolonged intensive
lipid-lowering therapy on the characteristics of carotid atherosclerotic
also worth noting that a previous study36 has demonstrated plaques in vivo by MRI: a case-control study. Arterioscler Thromb
that the segmentation of stenotic carotid arteries in MRA Vasc Biol 2001;21:1623–1629.
images using a similar tube-fitting method was in good 8. Yogo M, Sasaki M, Ayaori M, et al. Intensive lipid lowering therapy
agreement with the expert results in terms of lumen diame- with titrated rosuvastatin yields greater atherosclerotic aortic plaque
regression: Serial magnetic resonance imaging observations from
ter, cross-sectional area, and stenosis grading. Future studies RAPID study. Atherosclerosis 2014;232:31–39.
need to be carried out to firmly establish the ability of the 9. Sibley CT, Vavere AL, Gottlieb I, et al. MRI-measured regression of
proposed algorithm to perform reliable segmentation of the carotid atherosclerosis induced by statins with and without niacin in a
vessel wall in patients with atherosclerotic plaque. Fourth, randomised controlled trial: the NIA plaque study. Heart 2013;99:
1675–1680.
testing the segmentation on images acquired with one MR
10. Saam T, Yuan C, Chu B, et al. Predictors of carotid atherosclerotic
sequence only may also raise questions about the generaliz- plaque progression as measured by noninvasive magnetic resonance
ability of our algorithm. Nevertheless, although subtle imaging. Atherosclerosis 2007;194:e34–42.
changes in the image contrast may occur due to the use of 11. Lima JA, Desai MY, Steen H, Warren WP, Gautam S, Lai S. Statin-
induced cholesterol lowering and plaque regression after 6 months of
other imaging pulse sequence, it is reasonable to believe that
magnetic resonance imaging-monitored therapy. Circulation 2004;
the presented segmentation algorithm could also work well 110:2336–2341.
for other MR sequences as long as the vessel wall boundaries 12. Phinikaridou A, Ruberg FL, Hallock KJ, et al. In vivo detection of vul-
are reliable. Finally, this is a retrospective study, but the nerable atherosclerotic plaque by MRI in a rabbit model. Circ Cardio-
vasc Imaging 2010;3:323–332.
evaluation results reported here suggest that the newly-
13. Saam T, Hatsukami TS, Takaya N, et al. The vulnerable, or high-risk,
developed pipeline can be used in future prospective studies
atherosclerotic plaque: noninvasive MRI for characterization and
for assessing the effect of pharmacological treatment of assessment. Radiology 2007;244:64–77.
atherosclerosis. 14. Orion KC, Ruppert J, Call D, Malas M, Piazza K, Perler BA. The role
In conclusion, we presented a fully automated method of advanced diagnostic technology in the selection of a patient with
symptomatic but hemodynamically insignificant disease for carotid
to detect vessel wall boundaries and quantify VWT in both endarterectomy. J Vasc Surg Cases 2015;1:90–93.
CCA and DAO. The results demonstrated that our pre-
15. Yuan C, Lin E, Millard J, Hwang JN. Closed contour edge detection
sented method has good agreement with manual segmenta- of blood vessel lumen and outer wall boundaries in black-blood MR
tion, excellent scan–rescan reproducibility, and robust images. Magn Reson Imaging 1999;17:257–266.
capability in detecting changes in VWT, and therefore, is 16. Underhill HR, Kerwin WS, Hatsukami TS, Yuan C. Automated mea-
surement of mean wall thickness in the common carotid artery by
especially suitable to be applied in large-scale longitudinal MRI: a comparison to intima-media thickness by B-mode ultrasound.
studies to evaluate the treatment effect. J Magn Reson Imaging 2006;24:379–387.
January 2017 227

17. Ladak HM, Thomas JB, Mitchell JR, Rutt BK, Steinman DA. A semi- MRI study of vessel wall thickening and stiffening in hypertension.
automatic technique for measurement of arterial wall from black Invest Radiol 2013;48:86–91.
blood MRI. Med Phys 2001;28:1098–1107.
28. Alizadeh Dehnavi R, Doornbos J, Tamsma JT, et al. Assessment of
18. Adame IM, van der Geest RJ, Wasserman BA, Mohamed MA, Reiber the carotid artery by MRI at 3T: a study on reproducibility. J Magn
JH, Lelieveldt BP. Automatic segmentation and plaque characteriza- Reson Imaging 2007;25:1035–1043.
tion in atherosclerotic carotid artery MR images. MAGMA 2004;16:
227–234. 29. Roes SD, Westenberg JJ, Doornbos J, et al. Aortic vessel wall mag-
netic resonance imaging at 3.0 Tesla: a reproducibility study of respi-
19. Jin Y, Ladak HM. Software for interactive segmentation of the carotid ratory navigator gated free-breathing 3D black blood magnetic
artery from 3D black blood magnetic resonance images. Comput resonance imaging. Magn Reson Med 2009;61:35–44.
Methods Programs Biomed 2004;75:31–43.
30. Makowski P, de Koning PJH, Angelie E, Westenberg JJM, van der
20. van ’t Klooster R, de Koning PJ, Dehnavi RA, et al. Automatic lumen Geest RJ, Reiber JHC. 3D cylindrical B-spline segmentation of carotid
and outer wall segmentation of the carotid artery using deformable arteries from MRI images. Lect Notes Comput Sc 2006;4072:188–196.
three-dimensional models in MR angiography and vessel wall images.
J Magn Reson Imaging 2012;35:156–165. 31. vanderGeest RJ, Buller VGM, Jansen E, et al. Comparison between
manual and semiautomated analysis of left ventricular volume param-
21. Subasic M, Loncaric S, Sorantin E. Model-based quantitative AAA eters from short-axis MR images. J Comput Assist Tomo 1997;21:
image analysis using a priori knowledge. Comput Methods Programs 756–765.
Biomed 2005;80:103–114.
32. Angelie E, Oost ER, Hendriksen D, Lelieveldt BP, Van der Geest RJ,
22. Volonghi P, Tresoldi D, Cadioli M, et al. Automatic extraction of Reiber JH. Automated contour detection in cardiac MRI using active
three-dimensional thoracic aorta geometric model from phase con- appearance models: the effect of the composition of the training set.
trast MRI for morphometric and hemodynamic characterization. Magn Invest Radiol 2007;42:697–703.
Reson Med 2016;75:873–882.
33. Faul F, Erdfelder E, Lang AG, Buchner A. G*Power 3: A flexible statis-
23. Tang H, van Walsum T, van Onkelen RS, et al. Semiautomatic carotid tical power analysis program for the social, behavioral, and biomedi-
lumen segmentation for quantification of lumen geometry in multi- cal sciences. Behav Res Methods 2007;39:175–191.
spectral MRI. Med Image Anal 2012;16:1202–1215.
34. Corti R, Fuster V, Fayad ZA, et al. Lipid lowering by simvastatin indu-
24. Duquette AA, Jodoin PM, Bouchot O, Lalande A. 3D segmentation of ces regression of human atherosclerotic lesions: two years’ follow-up
abdominal aorta from CT-scan and MR images. Comput Med Imaging
by high-resolution noninvasive magnetic resonance imaging. Circula-
Graph 2012;36:294–303.
tion 2002;106:2884–2887.
25. Arias A, Petersen J, van Engelen A, et al. Carotid artery wall segmen-
35. Adame IM, van der Geest RJ, Bluemke DA, Lima JA, Reiber JH,
tation in multispectral MRI by coupled optimal surface graph cuts.
Lelieveldt BP. Automatic vessel wall contour detection and quantifica-
IEEE Trans Med Imaging 2015 [Epub ahead of print].
tion of wall thickness in in-vivo MR images of the human aorta.
26. Ukwatta E, Yuan J, Rajchl M, Qiu W, Tessier D, Fenster A. 3-D carotid J Magn Reson Imaging 2006;24:595–602.
multi-region MRI segmentation by globally optimal evolution of
36. Suinesiaputra A, de Koning PJ, Zudilova-Seinstra E, Reiber JH, van
coupled surfaces. IEEE Trans Med Imaging 2013;32:770–785.
der Geest RJ. Automated quantification of carotid artery stenosis on
27. Brandts A, Westenberg JJ, van Elderen SG, et al. Site-specific cou- contrast-enhanced MRA data using a deformable vascular tube
pling between vascular wall thickness and function: an observational model. Int J Cardiovasc Imaging 2012;28:1513–1524.

ORIGINAL RESEARCH
Thrombus-Mimicking Artifacts in
Two-Point Dixon MRI: Prevalence,
Appearance, and Severity
Tilman Schubert, MD,1,2* Peter Bannas, MD,1,3 Sonja Kinner, MD,1,4
Samir Sharma, PhD,5 James H. Holmes, PhD,1,5 Mahdi Salmani Rahimi, PhD,5
Frank R. Korosec, PhD,1,5 and Scott B. Reeder, MD, PhD1,5,6,7
Purpose: To evaluate the incidence and severity of potentially thrombus mimicking, flow-induced misallocation artifacts
in a clinical setting. Two-point “Dixon” fat–water separation methods, with bipolar readout gradients, may suffer from
flow-induced fat–water misallocation artifacts. If these artifacts occur within blood vessels, they may mimic thrombus.
Materials and Methods: Two-point Dixon coronal and axial images acquired in 102 consecutive patients were retro-
spectively evaluated for the presence of flow-induced artifacts in arteries and veins. Artifacts were graded on a 3-point
scale (none, mild, severe) by two independent readers. Interreader agreement was evaluated with kappa statistics.
Results: Reader 1 reported 63 artifacts in 46 (45%) of the cases (severe in 19 cases, 18.6%). Reader 2 reported 51 artifacts in
43 (42.2%) of the cases (severe in 18 cases, 17.6%). Misallocation of fat and water was apparent in all datasets with severe arti-
facts, whereas variable signal intensity changes in water and fat images were observed in mild artifacts. Interreader agree-
ment was good for artifacts appearing in coronal images (j 5 0.7) and fair for artifact appearance in axial images (j 5 0.24).
Conclusion: Our study shows a high incidence of flow-induced mild and severe artifacts in a two-point Dixon method
with bipolar readout gradients. This artifact should not be misinterpreted as intravascular thrombus.
C hemical shift-based two-point Dixon fat–water separation

magnetic resonance imaging (MRI) techniques provide
robust fat saturation and have recently been applied to MR
partial k-space acquisitions and parallel imaging are applied
to reduce the scan time.5,6 The use of bipolar readout gra-
dients acquired in the same relaxation time (TR) is another
angiography (MRA).1,2 The advantage of the Dixon fat–water common method to reduce scan time7 and is routinely
separation in MRA is its ability to provide T1-weighted water- implemented in commercially available pulse sequences.
only images with arterial contrast enhancement. This improves However, with a bipolar readout acquisition, the first
the quality of maximum-intensity projection (MIP) images by moments of the two echoes are different, leading to phase
suppressing background fat-signal without the need for subtrac- differences between the two echoes described by the follow-
tion from precontrast datasets. Apart from application in MRA,
ing equation:
two-point fat–water separation methods are increasingly used
D/5cDM1 m (1)
for body MRI, as they not only provide robust fat saturation,
but also include information about the fat content of tissue.3,4 where c is the gyromagnetic ratio of 1H, DM1 is the differ-
However, the acquisition of two separate echoes leads ence in the first moment of the readout gradient between
to an increase in scan time. Therefore, methods such as echoes, and v is the velocity of a moving spin along the
Received Nov 28, 2015, Accepted for publication Jun 6, 2016.
This article was published online on 5 July 2016. An error was subsequently identified. This notice is included in the online and print versions to indicate
that both have been corrected 20 July 2016.
*Address reprint requests to: T.S., Wisconsin Institutes for Medical Research (WIMR), 1111 Highland Ave., WI 53705-2275. E-mail: tschubert2@wisc.edu
From the 1Department of Radiology, University of Wisconsin-Madison, Madison, Wisconsin, USA; 2Clinic for Radiology and Nuclear Medicine, Basel
University Hospital, Basel, Switzerland; 3Department of Radiology, University Medical Center Hamburg-Eppendorf, Hamburg, Germany; 4Department of
Diagnostic and Interventional Radiology and Neuroradiology, University Hospital Essen, Essen, Germany; 5Department of Medical Physics, University of
Wisconsin-Madison, Madison, Wisconsin, USA; 6Department of Biomedical Engineering, University of Wisconsin-Madison, Madison, Wisconsin, USA; and
7
Department of Medicine, University of Wisconsin-Madison, Madison, Wisconsin, USA

V 229

readout direction. Spins that are moving during the readout Imaging parameters at 1.5T included: coronal slab excitation,
(eg, due to vascular flow) will accrue additional phase due field of view (FOV) 5 26 cm (S/I) 3 23.4 cm (R/L) 3 16.8 cm
to the bipolar acquisition. Because two-point fat–water (A/P), with 288 3 192 3 84 matrix size for an acquired spatial
imaging relies on the phase information, these additional resolution of 0.9 mm (R/L) 3 1.2 mm (S/I) 3 2.0 mm (A/P)
flow-induced phase shifts confound the separation of the fat interpolated to 0.5 3 0.5 3 2.0 mm through zero-filling. Other
parameters included TR/TE1/TE2 5 6.6/2.1/4.2 msec, flip
and water signals, and can lead to misallocation of the phase
angle 5 158, bandwidth (BW) 5 683.3 kHz. Parameters for the
information of spins flowing in the readout direction. If the
axial excitation included: FOV 5 26 cm (R/L) 3 19.2 cm (S/I) 3
phase accrual between the two echoes equals 1808, complete
21 cm (A/P), with 288 3 192 3 64 matrix size for an acquired
misallocation occurs.7 In addition, velocity-induced phase spatial resolution of 0.9 mm (R/L) 3 1.1 mm (A/P) 3 3.0 mm (S/
dispersion within a voxel containing spins of varying veloci- I) interpolated to 0.4 3 0.5 3 1.5 mm through zero-filling. Other
ty will result in magnitude signal dropout. parameters included TR/TE1/TE2 5 6.6/2.1/4.2 msec, flip
A recent study discovered the artifact and demonstrat- angle 5 158, BW 5 690.9 kHz.
ed that the different first moments of the readout gradient Imaging parameters at 3.0T system included: coronal slab
at the two echo times were the main cause of flow-induced excitation, FOV 5 48 cm (S/I) 3 38 cm (R/L) 3 32 cm (A/P),
misallocation of fat and water signals.7 Specifically, differen- with 320 3 192 3 160 matrix size for an acquired spatial resolu-
ces in first moments at the two echo times lead to potential- tion of 1.5 mm (S/I) 3 2.0 mm (R/L) 3 2.0 mm (A/P), interpolat-
ly thrombus-mimicking artifacts in vessels aligned in the ed to 0.8 3 0.9 3 1.5 mm through zero-filling. Other parameters
included TR/TE1/TE2 5 4.1/1.2/2.4 msec, flip angle 5 128,
readout direction.7
BW 5 6142.86 kHz. The parameters for the axial excitation
However, the prevalence, severity, and potential clinical
included: FOV: 34 cm (R/L) 3 30 cm (A/P) 3 30 cm (S/I), with
impact remains to be elucidated. It is critical that radiolog- 320 3 224 3 100 matrix size for an acquired spatial resolution of
ists and MR scientists are aware of flow-induced artifacts in 1.0 mm (R/L) 3 1.3 mm (A/P) 3 3.0 mm (S/I) interpolated to
two-point Dixon techniques, including their frequency and 0.7 3 0.6 3 1.5 mm through zero-filling. Other parameters
appearance, as intravascular artifacts have the potential to be included TR/TE1/TE2 5 4.1/1.2/2.4 msec, flip angle 5 128,
misdiagnosed as thrombus.8,9 The purpose of this work was BW 5 6142.86 kHz.
to characterize the incidence, severity, and location of these MRI was performed after injection of gadobenate dimeglu-
artifacts in body MRI. We further investigated the causes mine (MultiHance, Bracco Diagnostics, Princeton, NJ) at a dose of
and probable solutions that may help to avoid misallocation 0.1 mmol/kg with an injection rate of 2 ml/s. Two-point datasets
artifacts in two-point Dixon MRI. were acquired 2–5 minutes after injection of contrast in the
steady-state equilibrium phase. All two-point Dixon fat–water
Materials and Methods image datasets were acquired in a single breath-hold at end-
expiration.
This retrospective study was approved by the local Institutional
Fat–water separated images were reconstructed online using
Review Board (IRB) and was Health Insurance Portability and
the product implementation of the two-point Dixon fat–water sep-
Accountability Act (HIPAA) compliant. Given the retrospective
aration technique.
nature of the study, the IRB granted a waiver of informed consent
T1-weighted, single echo spoiled gradient recall (SGRE)
to perform this retrospective review.
imaging postcontrast were reviewed in order to verify absence of
In all, 102 consecutive patients (age range 18–83 years,
thrombus. Sequence parameters included: TR/TE 5 4.54/1.40
mean age 41 years, 65 female) undergoing a body MRI scan msec, BW 5 6 127 kHz, FOV 5 400 3 360 3 320 mm3, 224 3
including coronal and axial two-point Dixon MRI at our institu- 160 3 160 matrix, for true spatial resolution interpolated to 0.8
tion were evaluated. MR examinations comprised 72 chest exams, 3 0.7 3 1.1 mm3 through zero-filling.
11 exams of both the chest and abdomen, and 17 abdominal
exams. Indications included: suspected pulmonary embolism Qualitative Image Analysis
(n 5 49), vasculitis (n 5 10 cases), follow-up of aortic aneurysm Two radiologists with 10 years (S.K.) and 7 years (T.S.) of body
(n 5 9 cases), aortic dissection (n 5 9 cases), post-liver or renal MR experience independently evaluated the contrast-enhanced
transplant assessment (n 5 5 cases), uncharacterized abdominal coronal and axial water and fat images. Both radiologists were
lesion (n 5 6 cases), aortic valve assessment (n 5 3 cases), gonadal blinded to the other’s assessment. Based on the appearance in the
vein insufficiency, pulmonary artery aneurysm, aortic coarctation water and/or fat images, the severity of intravascular artifacts was
repair follow-up, pretransplant assessment (n 5 2 each), and jaun- recorded and graded on a 3-point scale (no visible artifact 5 no
dice (n 5 1). visually detectable intravascular signal changes in water and/or fat
image; mild 5 visually detectable intravascular mild-moderate
MRI Protocol decrease in water and/or signal increase in fat image; severe 5 clear
MR data were acquired using 1.5T clinical scanners (Optima intravascular signal dropoff in water image with corresponding fat
450w and Signa HDxt, GE Healthcare, Waukesha, WI) in 73 cases signal at same location on the fat-only image). Single-echo conven-
and 3.0T clinical scanners (Discovery 750 and Discovery 750w, tional postcontrast T1 sequences acquired either in the arterial and
GE Healthcare) in 29 cases. venous phase or only in venous phase served as the standard of

Schubert et al.: Incidence of Flow-Induced Artifacts
0.21–0.40 fair agreement; 0.41–0.60 moderate agreement; 0.61–

0.80 good agreement; and 0.81–1.00 excellent agreement.10
To test for a relationship between artifact appearance and
IVC cross-sectional area change, a Spearman-Rank correlation anal-
ysis was performed for artifact severity in the IVC in the coronal
plane and IVC cross-sectional area change.
To test for a relationship between artifact appearance and
magnetic field strength, a chi-square test with artifact severity
(none/mild/severe) and magnetic field strength (1.5T/3.0T) was
performed. This analysis was performed for both readers.
P-values less than 0.05 were considered statistically
significant.
All statistical analyses were performed using commercially
available software (SPSS, v. 23.0; SPSS, Chicago, IL)
Results
Figure 2 shows examples of two-point Dixon images with F2
absent, mild, and severe water–fat signal misallocation arti-
FIGURE 1: Choice of imaging slices for IVC surface area calcula-
facts in the IVC. Separated water and fat images demon-
tion. White lines indicating the locations of IVC surface area strate how misallocation of intravascular signal from the
measurements in the axial plane. water-only into the fat-only image mimics thrombus on the
water-only image. T1-weighted spoiled gradient recall
reference and were reviewed to exclude thrombus in cases when
(SGRE) imaging (not shown) verified the absence of IVC
intravascular signal misallocation artifacts were present.
thrombus in all cases.
The following thoracic vessels were evaluated: 1) arteries:
Figure 3 shows a similar example of water–fat misallo- F3
ascending and descending aorta, brachiocephalic artery, proximal left
cation artifact mimicking thrombus in the renal arteries.
and right subclavian arteries, and proximal left and right common
carotid arteries, 2) veins: superior vena cava, subclavian veins, jugular Increased signal on the fat-only image is visualized, and con-
veins, and azygos vein. The following abdominal vessels were evaluat- ventional T1-weighted SGRE imaging verifies the absence of
ed: 1) arteries: descending aorta, common and external iliac arteries, intra-arterial clot.
renal arteries, splenic artery, hepatic artery, superior mesenteric artery, Figures 4 and 5 show cases where misallocation arti- F4 F5
2) veins: inferior vena cava (IVC), left and right renal veins, splenic facts were present in coronal and axial images at the same
vein, right, middle and left hepatic veins, and main portal vein. location. In both cases, T1-weighted SGRE images proved
absence of thrombus.
Quantitative Analysis Figure 6 shows a patient with a large tumor thrombus F6
The IVC luminal change between the suprarenal segment to the extending from the right renal vein into the inferior vena
smallest lumen of the hepatic segment of the IVC was evaluated in cava. Although the thrombus partly obstructs the IVC, no
order to determine whether a potential correlation between flow flow-induced artifact was present.
induced misallocation and flow acceleration due to diameter Reader 1 reported 63 (35 mild, 28 severe) artifacts in
F1 change was present (Fig. 1). In order to quantify the change of the
46 (45%) of the cases (severe in 19 cases, 18.6%). Reader 2
inner lumen of the IVC, the luminal area of the IVC was mea-
reported 51 artifacts (28 mild, 23 severe) in 43 (42.2%) of
sured using an oval-shaped region of interest (ROI) at both levels.
the cases (severe in 18 cases, 17.6%).
The percentage of cross-sectional area reduction of the IVC from
the suprarenal to hepatic segment was calculated as follows:
In two patients, intravascular thrombus was detected
(inferior vena cava, left hepatic vein). However, in none of
ðIVC supraren2IVC hepÞ these two cases were flow-induced artifacts present.
3100 (2)
IVC supraren
where IVC supraren refers to the cross-sectional area of the suprare- Artifacts in Coronal Imaging
nal segment of the IVC and IVC hep refers to the cross-sectional Mild intravascular fat–water misallocation artifacts were
area of the hepatic IVC. reported in 25 cases (24.5%) by Reader 1 and in 23 cases
by Reader 2 (22.5%). Severe intravascular fat–water misallo-
Statistical Analysis cation artifacts were reported in 18 cases (17.6%) by Reader
Interreader agreement of artifact rating was assessed using Cohen’s 1 and in 17 cases by Reader 2 (16.7%). Interreader agree-
kappa statistics. A j-value of 0.10–0.20 indicated slight agreement; ment was good in the coronal plane (j 5 0.7).
January 2017 231

FIGURE 2: None (A, scanned at 1.5T, 13.6% IVC cross-section area reduction), mild (B, 1.5T, 75% IVC cross-section area reduc-
tion), and severe (C, 1.5T, 64% IVC cross-section area reduction) misallocation artifact in the IVC (arrowheads) with the corre-
sponding fat-only image (D,E) below each water image.
Artifact in Axial Imaging

Mild intravascular fat–water misallocation artifacts were
reported in 10 cases (10.2%) by Reader 1 and in five cases
by Reader 2 (4.9%). Severe intravascular fat–water misallo-
cation artifacts were reported in seven cases (6.7%) by Read-
er 1 and in three cases by Reader 2 (2.9%). Interreader
agreement was fair in the axial plane (j 5 0.24). Location
T1 and severity of artifacts are summarized in Table 1.
IVC cross-sectional area change from the suprarenal to the

hepatic segment showed a mean decrease of 32% (635% SD),
ranging from 91% decrease to an increase of 29%. Spearman-
Rank correlation analysis revealed a significant correlation
between IVC luminal area change and IVC artifact frequency in
the coronal plane (correlation coefficient 5 0.254 / P 5 0.021,
F7 Fig. 7), indicating a higher incidence of flow-induced misalloca-
tion artifacts at a greater reduction in IVC cross-sectional area.
Reader 1 reported artifacts in 37/73 patients (20 with
mild, 17 with severe artifacts) scanned at a magnetic field
strength of 1.5T (49% of patients scanned at 1.5T; 27%
mild, 23% severe) and in nine (seven mild with mild, two
with severe artifacts) out of 29 (31% of patients scanned at
3.0T; 24% mild, 7% severe) patients scanned at 3.0T. The FIGURE 3: Severe flow induced fat and water misallocation in
bilateral renal arteries (arrows): A: water-only image, B: fat-only
chi-square test revealed no significant difference in artifact image. The corresponding MRA shows patent bilateral renal
incidence at 1.5T and 3.0T (P 5 0.06). arteries (C).

FIGURE 4: Severe misallocation artifact in the aorta and right common iliac artery in a patient with Takayashu’s disease and multi-
ple arterial stenoses. Artifacts are seen in coronal and axial fat- (B,E, arrows) and water-only (A,D, arrows) images in the same
location. MRA (C) revealed patent but stenotic vessels.
Reader 2 reported artifacts in 35/73 patients (22 with axial plane. Due to the smaller diameter of blood vessels
mild, 13 with severe artifacts) scanned at a magnetic field coursing left–right in the axial plane (eg. renal arteries and
strength of 1.5T and in eight (three with mild, five with veins) compared to large vessels coursing in the craniocau-
severe artifacts) out of 29 patients scanned at 3.0T. The chi- dad direction (eg. aorta, IVC), artifacts in axial images tend
square test revealed no significant difference in artifact inci- to be much smaller than those seen in the coronal plane.
dence at 1.5T and 3.0T (P 5 0.1). Matching our experience, the signal misallocation arti-
fact was described in the intrahepatic inferior vena cava
Discussion (IVC) in a small number of clinical cases in a previous
In this study we investigated the prevalence, severity, and study.7 Therefore, we performed a quantitative analysis of
location of potentially thrombus-mimicking, flow-related the cross-sectional area change of the IVC in order to evalu-
fat–water misallocations in two-point Dixon images ate a potential effect of cross-sectional area reduction on
acquired with bipolar readout gradients. Our study revealed artifact incidence and severity. A larger cross-sectional area
that misallocation artifacts appear in almost half of patients reduction results in a greater degree of physiologic stenosis
with two-point Dixon MR exams. The artifacts were of the IVC with a greater flow acceleration. However, this
observed particularly in the coronal slice orientation, and in quantitative analysis revealed only a weak correlation
vessels with flow in the same direction as the readout gradi- between artifact appearance and apparent decrease in cross-
ent. Misallocation artifacts appeared more frequently in the sectional area. Therefore, based on our results, a physiologi-
venous system with a predilection to occur in the intrahe- cal IVC-stenosis might not necessarily lead to a greater like-
patic IVC. lihood of flow-induced artifact appearance.
The qualitative analysis of the presence of artifacts The reason for the difference in artifact incidence
revealed a higher distinctiveness of misallocation artifacts in between arteries and veins cannot yet be fully explained.
the coronal plane. This finding is reflected by a good inter- Arteries and veins (eg, hepatic veins) may have pulsatile
reader agreement versus a fair interreader agreement in the flow. The flow in veins, unlike flow in arteries, may also
January 2017 233

FIGURE 5: Severe misallocation artifact in the proper and common hepatic artery (arrows) in axial and coronal planes in the same
locations. A,C: Water-only images; B,D: Fat-only images.
vary with respiration. It remains unclear how these differen- cases in our study. Nevertheless, we experienced two cases
ces impact the incidence of the observed artifact. Further with severe flow-induced artifacts in the same location in
studies in phantoms may be helpful to elucidate the under- both the axial and coronal planes. This may be due to vorti-
lying mechanisms. cal blood flow patterns that contained velocity components
Flow-induced artifacts were previously reported to the- in both the craniocaudad and left–right directions. In this
oretically appear more frequently at 1.5T due to longer study, contrast-enhanced non-Dixon T1-weighted spoiled
echo spacing and a greater first moment (DM1, Eq. 1).7 gradient echo (SGRE) MRI acquired in either the arterial
Indeed, artifacts appeared more frequently at 1.5T com- and venous phase or in the venous phase only. This is wide-
pared to 3.0T in our study, albeit without reaching a statis- ly regarded as the reference standard for IVC clot.12,13
tically significant difference. Therefore, we recommend that in addition to water and fat
Our study has clinical implications, as two-point Dix- images, single-echo postcontrast T1 sequences should be
on MRI is increasingly used in clinical imaging due to the reviewed when a potential thrombus-mimicking artifact is
robust fat suppression and high image quality.11 Generally, detected. Other solutions include swapping the readout and
imaging artifacts in MRI are known to have the potential to phase-encoding directions as well as the application of dual
mimic thrombus.8 However, severe flow-induced misalloca- pass, unipolar readout methods.7,14 However, unipolar read-
tion artifacts in two-point Dixon MRI can be correctly diag- out methods lead to increased scan time, which may be
nosed as artifacts. Due to the fact that the artifact is caused challenging for breath-hold examinations in body imaging.
by signal misallocation of fat and water due to flow, artifacts A further way to reduce these artifacts are optimized readout
always appear dark in the water-only and bright in the fat- gradients in order to minimize DM1 at the opposed-phase
only image. Therefore, the artifact can be identified with and in-phase echo times7 as well as readout gradients that
certainty if the fat images are reviewed together with the are designed to be first-moment nulled.
water images. Artifacts that were graded as mild in our Future prospective studies performing direct compari-
study did not consistently show definite signal change in the son of Dixon MRI to conventional fat-suppressed T1-
fat-only image. However, these are unlikely to be misinter- weighted imaging may be helpful to determine the true clin-
preted as thrombus due to the mild intravascular signal ical impact of the thrombus-mimicking artifact.
drop in the water-only image. The artifact appeared only in Lastly, flow-induced misallocation artifacts should not
either the coronal or axial plane in the large majority of be confounded with the well-known fat–water swaps caused

FIGURE 6: Case of a large, right-sided renal tumor (asterisk), scanned at 3T. A thrombus extends from the right renal vein into the
IVC (arrows). No fat–water misallocation was detected in the water only images (coronal: A, axial: D) and the corresponding fat only
images (coronal: B, axial: E). A single-echo, T1-weighted sequence postcontrast confirmed the presence of thrombus (C, arrow).
by magnetic field inhomogeneities and the natural ambigui- generally appear as nonanatomic geographic fat–water misal-
ty that can occur when voxels contain primarily all water or locations. Fat–water misallocations due to magnetic field
all fat.15–17 These swaps represent a separate artifact that inhomogeneities do not lead to the thrombus-mimicking
artifact as described in this study. Further, magnetic field
TABLE 1. Total Numbers of Artifacts Within the inhomogeneities do not exacerbate velocity-induced phase
Cohort of 102 Patients With Regard to Localization shifts, and therefore we do not expect such inhomogeneities
and Severity of Flow-Induced Artifacts Reported by to affect the thrombus-mimicking artifact in any way.
Reader 1 and Reader 2 (Italics).
Mild artifact Severe artifact

Coronal Axial Coronal Axial
IVC 17; 15 3; 2 17; 16 —
Aorta 7; 5 — — —
Liver veins — 4; 3 — 2; 1
Azygos vein — 3; 0 — 2; 0
Superior vena cava 1; 3 — — —
Hepatic artery — — 1; 1 1; 1
Splenic artery — — — 2; 1
Common iliac artery — — 1; 1 1; 1
FIGURE 7: Artifact severity (x-axis) plotted vs. % IVC cross-
Renal artery — — — 1; 1 sectional area reduction (y-axis). A reduction in cross-sectional
area correlates with the incidence of flow-induced artifacts.
January 2017 235

Limitations of our study include different sizes of the 4. Qayyum A, Goh JS, Kakar S, Yeh BM, Merriman RB, Coakley FV.
Accuracy of liver fat quantification at MR imaging: comparison of out-
patient groups imaged at 1.5T and 3.0T and patient groups of-phase gradient-echo and fat-saturated fast spin-echo techniques—
with thoracic and abdominal exams. Furthermore, in- and initial experience. Radiology 2005;237:507–511.
opposed-phase source images were not available for most 5. Pruessmann KP, Weiger M, Scheidegger MB, Boesiger P. SENSE: sen-
sitivity encoding for fast MRI. Magn Reson Med 1999;42:952–962.
datasets, as these are not typically stored.
In conclusion, this study has shown that flow-related 6. Sodickson DK, Manning WJ. Simultaneous acquisition of spatial har-
monics (SMASH): fast imaging with radiofrequency coil arrays. Magn
fat–water misallocations occur in almost half of two-point Reson Med 1997;38:591–603.
Dixon exams acquired with bipolar readout gradients, and 7. Rahimi MS, Holmes JH, Wang K, Reeder SB, Korosec FR. Flow-
that these have the potential to be misinterpreted as throm- induced signal misallocation artifacts in two-point fat-water chemical
shift MRI. Magn Reson Med 2015;73:1926–1931.
bus. As Dixon algorithms continue to evolve, this may
8. Bannas P, Schiebler ML, Motosugi U, Francois CJ, Reeder SB, Nagle
reduce the incidence rate of this artifact. However, it SK. Pulmonary MRA: differentiation of pulmonary embolism from trun-
remains important for radiologists to be aware of the poten- cation artefact. Eur Radiol 2014;24:1942–1949.
tial for these artifacts during the evaluation of 2-point Dix- 9. Nishibori H, Kanematsu M, Kondo H, Matsuo M, Hoshi H. Pseudo-
on fat–water separation methods. thrombosis in the portal venous system: a potential pitfall with
gadolinium-enhanced dynamic gradient-recalled echo imaging of the
liver. J Magn Reson Imaging 2000;12:763–768.
Acknowledgments 10. Landis JR, Koch GG. The measurement of observer agreement for
categorical data. Biometrics 1977;33:159–174.
Contract grant sponsor: National Institutes of Health
11. Dogan BE, Ma J, Hwang K, Liu P, Yang WT. T1-weighted 3D dynamic
(NIH); Contract grant sponsor: (Helmut-Hartweg-Fonds) contrast-enhanced MRI of the breast using a dual-echo Dixon tech-
from the Swiss Academy of Medical Sciences (fellowship nique at 3 T. J Magn Reson Imaging 2011;34:842–851.
grant to T.S.) 12. Kanne JP, Lalani TA. Role of computed tomography and magnetic
resonance imaging for deep venous thrombosis and pulmonary
embolism. Circulation 2004;109:115–121.
13. Katz DS, Hon M. Current DVT imaging. Tech Vasc Interv Radiol 2004;
7:55–62.
References
14. Eggers H, Bornert P. Chemical shift encoding-based water-fat separa-
1. Leiner T, Habets J, Versluis B, et al. Subtractionless first-pass single
tion methods. J Magn Reson Imaging 2014;40:251–268.
contrast medium dose peripheral MR angiography using two-point
Dixon fat suppression. Eur Radiol 2013;23:2228–2235. 15. Bley TA, Wieben O, Francois CJ, Brittain JH, Reeder SB. Fat and water
magnetic resonance imaging. J Magn Reson Imaging 2010;31:4–18.
2. Michaely HJ, Attenberger UI, Dietrich O, et al. Feasibility of
gadofosveset-enhanced steady-state magnetic resonance angiogra- 16. Hernando D, Kellman P, Haldar JP, Liang ZP. Robust water/fat separa-
phy of the peripheral vessels at 3 Tesla with Dixon fat saturation. tion in the presence of large field inhomogeneities using a graph cut
Invest Radiol 2008;43:635–641. algorithm. Magn Reson Med 2010;63:79–90.
3. Ma J. Breath-hold water and fat imaging using a dual-echo two-point 17. Sharma SD, Artz NS, Hernando D, Horng DE, Reeder SB. Improving chemi-
Dixon technique with an efficient and robust phase-correction algo- cal shift encoded water-fat separation using object-based information of
rithm. Magn Reson Med 2004;52:415–419. the magnetic field inhomogeneity. Magn Reson Med 2015;73:597–604.

ORIGINAL RESEARCH
Optimization of Two-Compartment-
Exchange-Model Analysis for Dynamic
Contrast-Enhanced MRI Incorporating
Bolus Arrival Time
Guy Nadav, MSc,1,2 Gilad Liberman, PhD,1,3 Moran Artzi, PhD,1
Nahum Kiryati, PhD,2 and Dafna Ben Bashat, PhD1,4,5*
Purpose: To optimize the analysis of dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) under the
two-compartment-exchange-model (2CXM) and to incorporate voxelwise bolus-arrival-time (BAT).
Materials and Methods: The accuracy of the pharmacokinetic (PK) parameters, extracted from 3T DCE-MRI using 2CXM,
was tested under several conditions: eight algorithms for data estimation; correction for BAT; using model selection; differ-
ent temporal resolution and scan duration. Comparisons were performed on simulated data. The best algorithm was
applied to seven patients with brain tumors or following stroke. The extracted perfusion parameters were compared to
those of dynamic susceptibility contrast MRI (DSC-MRI).
Results: ACoPeD (AIF-corrected-perfusion-DCE-MRI), an analysis using a 2nd derivative regularized-spline and incorpo-
rating BAT, achieved the most accurate estimation in simulated data, mean-relative-error: Fp, F, vp, ve: 24.8%, 41.7%,
26.4%, 27.2% vs. 76.5%, 190.8%, 78.8%, 82.39% of the direct four parameters estimation (one-sided two-sample t-test,
P < 0.001). Correction for BAT increased the estimation accuracy of the PK parameters by more than 30% and provided
a supertemporal resolution estimation of the BAT (higher than the acquired resolution, mean-absolute-error 0.2 sec).
High temporal resolution (2 sec) is required to avoid biased estimation of PK parameters, and long scan duration
(20 min) is important for reliable permeability but not for perfusion estimations, mean-error-reduction: E: 12%, ve:
6%. Using ACoPeD, PK values from normal-appearing white matter, gray matter, and lesion were extracted from
patients. Preliminary results showed significant voxelwise correlations to DSC-MRI, between flow values in a patient
following stroke (r 5 0.49, P < 0.001), and blood volume in a patient with a brain tumor (r 5 0.62, P < 0.001).
Conclusion: This study proposes an optimized analysis method, ACoPeD, for tissue perfusion and permeability estimation
using DCE-MRI, to be used in clinical settings.
Q uantification of perfusion and vessel permeability in

the human body and brain is of some clinical impor-
tance. Within the brain, dynamic contrast-enhanced mag-
models used in DCE-MRI analysis neglect perfusion-related
parameters such as the cerebral blood flow (CBF). For per-
fusion extraction, the four-parameter two-compartment
netic resonance imaging (DCE-MRI) is commonly used to exchange model (2CXM) is commonly used,2–4 and has
estimate permeability-related pharmacokinetic (PK) parame- shown promising results for estimating CBF, cerebral blood
ters. A widely used model for DCE-MRI analysis is the volume (CBV), and permeability from DCE-MRI,3,4
three-parameter Extended Toft’s Model (ETM),1 which although alternative models are available.1 However, the
includes the volume transfer constant (Ktrans), plasma vol- estimation process, techniques, and corrections for inaccura-
ume (vp), and extravascular extracellular space (ve). Most cies in the 2CXM model vary among research groups.2–8
Received Oct 17, 2015, Accepted for publication Jun 10, 2016.
*Address reprint requests to: D.B.B., Wohl Institute for Advanced Imaging, Tel Aviv Sourasky Medical Center, 6 Weizman St. Tel-Aviv 64239 Israel.
E-mail: dafnab@tlvmc.gov.il
From the 1Functional Brain Center, Wohl Institute for Advanced Imaging, Tel Aviv Sourasky Medical Center, Tel Aviv, Israel; 2School of Electrical
Engineering, Tel Aviv University, Tel Aviv, Israel; 3Department of Chemical Physics, Weizmann Institute, Rehovot, Israel; 4Sackler Faculty of Medicine, Tel
Aviv University, Tel Aviv, Israel; and 5Sagol School of Neuroscience, Tel Aviv University, Tel Aviv, Israel.
Additional Supporting Information may be found in the online version of this article.

V 237

Methods for DCE-MRI analysis varied between studies. proposed in this study: improvement in accurate approxima-
In a thorough discussion, Sourbron et al4 present different tion to the AIF convolution matrix using a piecewise linear
monoexponential regimes for exchange models, and suggested discretization; a simple transformation of the generalized sin-
estimating the parameters of the full four-parameter model gular value decomposition to a standard ridge regression for-
using a nonlinear curve fitting on the concentration–time curve mulation; selection of the regularization term; and using the
(CTC) with a variation of the Levenberg–Marquardt algo- Akaike information criterion with correction for finite sample
rithm.9 The analysis was done using the assumption that the sizes for model selection.
signal is proportional to the relaxation rate (1/T1), which is
only valid for low-contrast agent concentrations. Larson et al2 Theory
proposed a computationally exhaustive generalized singular ACoPeD (AIF-Corrected Perfusion DCE-MRI): The
value decomposition approach for the deconvolution problem. Suggested Algorithm for 2CXM Parameter
Bolus arrival time (BAT) has been shown to have an Estimation
impact on the estimated parameters when the analyzed area The model chosen in this method is an updated version of the
is distant from the location where the arterial-input-function 2CXM described by Larsson et al3 and based on the model
(AIF) was measured.10–12 Previous works on dynamic sus- standardization presented by Sourbron and Buckly,1 consist-
ceptibility contrast MRI (DSC-MRI)13–15 describe the probing of vp, ve, plasma flow (Fp), and permeability–surface area
lem of estimating perfusion under bolus delay (referred to product (PS). To estimate the 2CXM’s parameters we followed
here as BAT) and dispersion. A circular deconvolution the model-free deconvolution as described by Larsson et al3
method was previously proposed,15 which constructs a which treats the CTC as a convolution of the AIF and an
block-circulant deconvolution matrix that is time-invariant impulse response function (IRF).
and reduces the sensitivity of the estimated perfusion param- Figure 1 shows the general scheme of the proposed F1
eters to BAT. However, BAT estimation in DCE-MRI analy- method. The CTC was deconvolved with the AIF to estimate
sis has received limited attention, probably due to the low the IRF. The deconvolution was performed using the minimi-
temporal resolution (6 sec) commonly used, which is zation of a cost function given in a matrix form. The full deri-
much longer than the intertissue difference.13 Accounting vation of the minimization problem for the model-free IRF
for BAT was recently shown to improve the accuracy of the estimation is given in Supporting Appendix 1. The final mini-
estimated ETM parameters in DCE-MRI.16 Kershaw and mization problem is described in Eq. 1:
Buckley11 suggested estimation of BAT in DCE-MRI using
the adiabatic approximation to the tissue homogeneity kD0 V 0 2C k2 1k2 kV 0 2V 0o k2 (1)
(AATH) model, based on the linear-quadratic model
Where jD0 ¢ABL21 ; V 0 ¢LV ; V 00 ¢LV 0 .
(LQ-model) originally proposed by Cheong et al.17 This
A is the AIF convolution matrix, B is the basis functions
analysis assumes that contrast agent exchange only takes
matrix, V is the basis coefficients vector, Vo is the initial esti-
place at the venous end of the capillary bed, and although it
mation for the basis coefficients vector and L is a matrix opera-
often forms a good approximation, this assumption cannot
tor (eg, derivative estimation). After solving for V0, one can
be justified physiologically, especially in the brain.1 In this
extract the IRF by IRF 5 BL-1V0 (see Supporting Appendix 2),
prior study we incorporated BAT into the 2CXM, which is
which shows how this regularized minimization problem can
the most commonly used four-parameter model applied to
be transformed into a standard form. This enabled the use of a
DCE-MRI in the brain.1
simple ridge regression instead of the computationally expen-
Using the correct model is very important for accurate
sive generalized singular value decomposition.3 To enforce pri-
estimation and interpretation of the PK maps. Ewing et al18
or assumptions (eg, smoothness) on the IRF, it was modeled as
thoroughly demonstrated the use of model selection in
a linear combination of basis functions, ie:
DCE-MRI, using a nested three-parameter model with
F-test criterion on clinical data. Buckley and Sourbron7 IRF 5Basis Matrix Coefficients Vector
addressed model selection for four-parameter models and
suggested using the Akaike information criterion, while using a cubic-spline basis with a reduced number of eigen-
Ingrisch et al used model selection with Akaike information vectors to reduce the number of free parameters in the min-
criterion and Akaike weights to estimate perfusion and imization problem, as previously suggested2,3 (with N/5,
permeability parameters in patients with multiple sclerosis.8 where N is the number of data samples).
The aim of this study was to optimize the analysis meth- Piecewise linear discretization of the AIF,13 instead of
od for DCE-MRI under 2CXM while incorporating voxelwise the piecewise-constant discretization,3 was used In ACoPeD,
BAT. The starting point of the method proposed in this study to enhance accuracy in the discretization of the convolution
is the work of Larson et al2,3 and advancements in the analysis integral with no computational cost. This is shown in Sup-
model are based on Sourbron et al.1,4,7 Several advances are porting Appendix 1, Eq. [A1.4].

Nadav et al.: Optimization of 2CXM DCE-MRI With BAT
C
O
L
O
R
FIGURE 1: ACoPeD (AIF corrected perfusion DCE-MRI) flow chart. Given a concentration–time curve (CTC) and an arterial-input-
function (AIF), a set of shifted versions of the AIF is created (A). The CTC is deconvolved with every shifted version (B) and the
resulting impulse response functions (IRFs) are then fitted to the 2CXM biexponential model (C). The fitted IRFs are then
convolved back with shifted AIFs (D) and the shifted AIF corresponding to the best fit (marked in blue) is chosen (E).
After estimating the IRF, Fp was set to be the maximal This method forced the IRF into a biexponential
value of the IRF. A nonlinear biexponent fit to the IRF was shape, while providing a good fit to the data. The computa-
then carried out to estimate the other PK parameters, while tion time increases linearly with the number of AIF shifts,
keeping Fp fixed. ie, is inverse to the desired temporal resolution of the BAT
estimation, and to the BAT range.
BAT Correction Using ACoPeD
Model Selection in ACoPeD
To estimate the voxelwise BAT and to improve the accuracy
Model selection was done using the corrected Akaike informa-
of the PK parameters estimation, the following procedure
tion criterion, to compensate for finite sample size. In total,
was used (see Fig. 1):
five models were calculated based on model boundary
A: The AIF was upsampled, (ie, more data points were regimes, as discussed by Sourbron et al1,4: four-parameter
estimated) using piecewise cubic interpolation, and 2CXM (Fp, E, vp, ve), three-parameter compartment-uptake
then shifted in superresolution time steps (eg, 0.1 sec- model (Fp, E, vp), two-parameter highly vascularized / no con-
ond shift when the data are at 2-second resolution). trast agent exchange (Fp, vp), one-parameter, highly perfused
B: The CTC was deconvolved with varying predeter- with no contrast agent exchange (vp), and no contrast agent in
mined shifted versions of the measured AIF to get pos- the tissue (the null hypothesis). BAT was considered an
sible IRFs. additional parameter in the first four models, resulting in nine
C: Fp was fixed to be the maximal IRF value. The IRFs possible models.
were then fitted to a biexponent model, which reflects
the four-parameter 2CXM. Materials and Methods
D: Each of the fitted IRFs was convolved with the corre- Overview
sponding shifted AIF to produce CTC estimation. To test the accuracy of the proposed ACoPeD method, we com-
E: The shift that minimized the root mean square (RMS) pared its results to six different deconvolution techniques and to
error between the estimated CTC and the measured an algorithm that was previously suggested4 (a total of eight com-
one was used as the BAT. parisons) without adding BAT to the simulations. To test the effect
January 2017 239

of incorporating BAT on the accuracy of parameter estimation,
Direct 4-params
BAT was added in simulation and ACoPeD’s estimation accuracy
190.85 6 34.48
76.58 6 10.99
78.88 6 13.74
82.39 6 15.23
was compared with three algorithms that were previously suggested
to deal with BAT (a total of four algorithms). To investigate the
contribution of using model selection, the parameters were estimat-
ed assuming a full five-parameter model and using model selection.
ACoPeD was additionally tested for different temporal resolutions
and scan durations in simulations. We then applied the optimized
163.2 6 11.7
39.74 6 0.85
44.11 6 6.23
49.49 6 5.28
method to data from patients and compared flow and blood vol-
ume parameters extracted from DCE-MRI to those extracted from
Wiener
DSC-MRI. ACoPeD was coded in MatLab (MathWorks, Natick,

MA), and a publicly available implementation of the method is
offered.
Simulated Data
221.01 6 13.47
32.28 6 0.65
36.29 6 1.95
54.21 6 1.07
CTC GENERATION. An AIF was simulated using a population

PCA 2nd
averaged AIF19 and convolved through 1000 IRFs with varying

parameter values, generated from uniform distributions over physi-
ologically feasible ranges: Fp: 10–150 [mL/100mL/Min], extraction
Mean and standard error of the mean (SEM) [in %] of the 2CXM parameters, estimated using different analysis techniques.
fraction (E): 0–0.5 [arbitrary units (a.u.)], vp: 3–20 [mL/100mL],

ve: 3–20 [mL/100mL].3 According to Larsson et al,3 estimation is
226.11 6 13.07
Mean relative error 6 SEM [%]
acceptable as long as permeability is not too large (E < 0.5), which

40.92 6 0.66
33.62 6 1.27
56.62 6 1.05
is often the case with real data.3 The IRFs were built using the
PCA 1st
Analysis technique
2CXM and the convolved results were set as the CTCs. These
values were used as ground truth for validation. Gaussian noise
was added to the AIF and the CTCs in order to obtain a realistic
contrast-to-noise ratio (CNR) of 15.3 CNR was defined as the
mean concentration value divided by the CTC noise standard devi-
237.50 6 12.61
47.42 6 0.64
33.32 6 1.50
59.50 6 1.25
ation. All simulated CTCs were 6 minutes long with 2 seconds

temporal resolution, unless otherwise specified.
COMPARING DIFFERENT METHODS FOR PARAMETER ESTIMA-

PCA
TION. A total of eight different methods for estimation of the

PK parameters using 2CXM were tested and compared: seven
methods with different deconvolution techniques and a direct four-
41.72 6 3.20
24.82 6 0.67
26.41 6 1.54
27.29 6 0.94
parameter nonlinear fit using the Levenberg–Marquardt algorithm4

TABLE 1. 2CXM Parameter Estimation: Analysis Techniques
Spline 2nd
(Table 1). Two methods were used to construct a linear combina- T1

tion of basis functions for the IRF: a cubic spline basis (as used in
ACoPeD) and principal component analysis (PCA), which was
built from the eigenvectors of typical values 2CXM IRFs, to mini-
mize the reconstruction error.20 PCA basis vectors were built by
150.58 6 11.11
generating 100,000 2CXM IRFs with parameters uniformly

28.76 6 0.48
38.00 6 2.38
33.51 6 1.13
distributed in the mentioned typical range, from which the first N/

Spline 1st
5 components were taken. The cubic-spline and PCA methods

were tested with an additional 1st and 2nd derivative operator on
the regularization term4 to ensure smooth solutions. Higher deriva-
tives were tested but did not improve accuracy (data not pre-
sented). Constraining for a smooth IRF was also performed using
189.78 6 12.30
a Wiener filter21 for the deconvolution step.

28.90 6 0.48
35.88 6 2.52
49.73 6 1.10
BAT CORRECTION AND PARAMETER ESTIMATION. Simulated

Spline
data were created by adding varying time shifts to the AIF before
convolving, from a uniform distribution in the range of 0–20 sec-
onds with 0.1 seconds temporal resolution. This was done to simu-
late local BAT in the brain,5 with respect to the measured AIF.22
VP
FP
The accuracy of BAT estimation and the effect of incorporating

Ve
E
BAT in the 2CXM model on the accuracy of the estimated PK

C
O
L
O
R
FIGURE 2: Estimation of PK parameters using different analysis techniques. Estimation values and SEM: green 5 Spline with 2nd
derivative, blue 5 PCA with 2nd derivative, cyan 5 Wiener Filter, dashed black line 5 ground truth value.
parameters were tested using four methods: ACoPeD (Fig. 1); following the event. Inclusion criteria for all subjects were normal
LQ-model11; circular deconvolution15 with the IRF’s maximum glomerular filtration rate, and no contraindication to MRI scan.
value time (MVT) as the BAT estimation; and ACoPeD with no This study was approved by the hospital Review Board, and writ-
BAT correction. The circular deconvolution is the common proce- ten informed consent was obtained from all patients.
dure for handling BAT in DSC-MRI15 while the LQ-model
MRI PROTOCOL. MRI scans were performed on a 3.0T MRI
showed good results when dealing with two-compartment models
scanner (GE Signa Excite, Milwaukee, WI) using an eight-channel
in DCE-MRI.11
head coil, or Siemens system (Magnetom Prisma, Erlangen, Ger-
CHOOSING THE CORRECT MODEL. Simulated AIFs were con- many) using a 20-channel head coil. For the DCE-MRI data, eight
volved with 1000 IRFs for eight models: 2CXM, compartment- slices, with 5 mm thickness and no gap, were acquired using multi-
uptake, highly vascularized, and highly perfused models (using the phase 3D SPGR/FLASH T1-weighted imaging, during injection of
range for the parameters as described before), with and without 0.2 cc/kg (single dose) gadolinium dotarem (gadoterate meglu-
BAT. All parameters were estimated twice, using ACoPeD (which mine), using a power injector (Medrad, Pittsburgh, PA, Spectris
assumes a full model and results in five parameters) and using the Solaris EP), at a constant rate of 5 cc/sec, and with a delay of 30
correct ground truth model. seconds (field of view [FOV] / matrix 5 250mm/2563256,
EFFECT OF TEMPORAL RESOLUTION AND SCAN
TR/TE 5 4.11/2.1 msec, flip angle [FA] 5 208, 180 repetitions in
DURATION. Simulated AIF and CTCs were generated with a temporal resolution of 2 sec and a total scan time of 6 min). For
temporal resolution of 1 msec, and then subsampled to a temporal the T1 maps, variable FA spoiled SPGR (VFA-SPGR) data was
resolution of 2, 4, and 6 seconds before applying deconvolution acquired with nominal FAs 5 58, 108, 158, 208, 308. Slices were
estimation. The highest temporal resolution of 2 seconds was cho- located at the center of the tumor / ischemic stroke area (as
sen based on a previous study suggesting that this temporal resolu- defined based on the anatomical images).
tion is an acceptable resolution for perfusion estimation.3 In two patients, one with GB and one following stroke,
For the effect of scan duration, PK parameters were estimat- DSC-MRI data were additionally acquired 15 minutes after
ed and compared to the ground truth values, using ACoPeD, for DCE-MRI using a 2D gradient echo, echo planar imaging (GRE-
time durations of 6 and 20 minutes and using a temporal resolu- EPI) sequence (19 slices with 5 mm thickness and no gap, with
tion of 2 seconds. FOV/matrix 5 220–240mm/1283128, TR/TE 5 1300/30 msec,
and 92 repetitions in temporal resolution of 1.3 sec and a total
Data From Patients scan time of 2 min), during the injection of 0.4 cc/kg (double
Seven patients were included in this study, four patients with high dose) of gadolinium dotarem at a constant rate of 5 cc/sec, after a
grade brain tumors (biopsy proven glioblastoma, GB), and three delay of 15 seconds. A double dose was used as previously
patients following acute ischemic stroke, scanned 4–5 days suggested when combined with DCE-MRI.23
January 2017 241

C
O
L
O
R
FIGURE 3: Estimating BAT and affected parameters. The different techniques: ACoPeD (green), circular deconvolution (blue), LQ-
model (cyan), and no correction (red) was used for BAT correction and estimation. The top graph compares the different BAT esti-
mation methods while the rest of the graphs show the effect of estimation on other PK parameters. Dashed black line 5 ground
truth value.
DATA ANALYSIS. Preprocessing included brain extraction per- defined within the sagittal sinus, relative signal enhancement was cal-
formed using the brain extraction module of the FSL toolbox (http:// culated, and deconvolved with the AIF. The partial-volume correc-
www.fmrib.ox.ac.uk/fsl),24 and coregistration of all DCE volumes by tion factor is determined from the time-integral of the IRF.
rigid transformation using SPM8b realign module.25 The DCE-MRI Conversion from blood volume and flow (vb and Fb) to plasma (vp
data was coregistered and converted to CTC in a standard fashion.21 and Fp) was done using an assumed hematocrit value of 0.45.4 The
T1 values were calculated with correction for inaccuracy of the flip displayed flow map was normalized by setting the mean Fp value in
angle as suggested by Liberman et al.26 DCE-MRI data from patients the normal-appearing white matter (NAWM) to 30 (mL/100g/
was analyzed using ACoPeD with model selection. The AIF was min).3,4 BAT estimation was performed with a temporal resolution of
defined to be the mean CTC from several voxels extracted manually 0.1 seconds, and the range was set to be up to 5 seconds, a realistic
from arteries in each patient. The same procedure was used to extract value for high-grade glioma (indeed, no ceiling effect was evident in
an average venous concentration–time curve in order to apply a par- the results). Maps were filtered for signal intensity by removing voxels
tial volume effect correction method, as used by Sourbron et al.4 In for which the median absolute CTC was less than 0.15% of the maxi-
order to correct for partial volume effect, a mask was manually mal median absolute signal in that scan.

TABLE 2. BAT Correction and Parameter Estimation

ACoPeD LQ Cyclic deconv. No correction
FP 24.32 6 0.72 40.47 6 1.01 48.81 6 0.93 53.12 6 0.74

E 55.14 6 4.74 288.7 6 12.8 277.6 6 11.4 291.2 6 13.1
vp 33.97 6 3.16 62.56 6 3.32 46.64 6 1.98 60.1 6 2.42
ve 31.91 6 1.22 87.84 6 4.85 74.26 6 3.85 91.85 6 4.88
BAT 22.13 6 1.91 26.11 6 2.15 59.90 6 2.70 -
Mean relative error and standard error of the mean (SEM) [in %] of the bolus arrival time (BAT) and pharmacokinetic (PK) parame-
ters, estimated using different methods for BAT correction.
DSC-MRI data analysis was performed using the PErfusioN bins. In each bin, the mean relative error and the standard error of
Graphical User INterface (PENGUIN) software,15 from which the mean (SEM) were calculated and compared. In addition, the
CBF and CBV maps were extracted (referred to as F-DSC). The mean relative error and the standard deviation (of the results com-
estimation method in PENGUIN is according to Wu et al,15 in pared to the simulated ground truth values), were calculated for
which for every voxel the time signal curve is converted to CTC Fp, vp, ve, E and BAT, extracted for each analysis method.
using the known relationship between T2 signal intensity and trac-
DATA FROM PATIENTS. Correlation between DCE-MRI and
er tissue concentration. Maps of relative CBV are estimated by DSC-MRI for the flow and blood volume parameters was assessed
integrating the area under the CTC. Maps of relative CBF are esti- using Pearson correlation coefficient. Significant correlations in this
mated by measuring the height of each voxel’s IRF. To obtain the study were considered as r 0.25 and P < 0.05.
IRF, an AIF, usually derived directly from imaging data, is decon-
volved from the measured CTC using a block-circulant deconvolu- Results
tion matrix.
The optimized analysis method, ACoPeD, includes a 2nd
PK PARAMETER VALUES IN PATIENTS. PK values were derivative spline-based regularization term, BAT correction
extracted for the contralesional NAWM, normal-appearing gray and model selection. The open-source MatLab implementa-
matter (NAGM), and lesion in each patient. Semiautomatic thresh- tion is available at: https://github.com/guyov1/DCE_perfu-
old based segmentation of the lesion areas was performed using sion. ACoPeD’s total run time on 6 minutes of whole brain
AnalyzeDirect software v. 11.0, (Mayo Clinic, Rochester, MN). data with temporal resolution of 2 seconds was around 3
Segmentation was performed on the postcontrast T1-weighted
minutes on an Intel Xeon CPU with 16 cores @2.4 GHz
images for the patients with GB and from fluid attenuation inver-
when allowing parallelization on 10 cores. For every time
sion recovery (FLAIR) images for the patients following stroke.
shift, an additional overhead of 75% of the total run time
The NAWM and NAGM masks were extracted from the precon-
was observed. For the given BAT search range of up to 20
trast T1-weighted images using FMRIB’s FAST algorithm (FAST,
part of FSL). Mean values of Fp, vp, ve, E and BAT parameters seconds with 0.5 intervals, the total run time was 31(20/
were extracted from each mask. 0.5)*0.75*3 5 93 minutes.
COMPARISON BETWEEN DCE-MRI AND DSC-MRI. Comparison Simulated Data

between DCE-MRI and DSC-MRI was conducted in one patient
with GB and one patient following stroke, who had DSC-MRI COMPARING DIFFERENT METHODS FOR PARAMETER
data. For each subject, DCE-MRI vp and Fp maps were realigned ESTIMATION USING 2CXM. Table 1 summarizes the
and resliced to the DSC-MRI space using SPM8b’s rigid-body mean relative error in estimation of PK parameters over the
transformation module (MatLab statistical parametric mapping entire parameter range, for the eight different estimation
tool). Voxel-based correlations for blood flow and blood volume methods, seven with different deconvolution techniques and
maps were performed where applicable according to the model the direct four-parameter nonlinear fit. Figure 2 shows the F2
selection maps estimation of each PK parameter versus its ground truth val-
ue for the best three deconvolution estimation techniques,
in each category. The 2nd derivative regularized-spline tech-
SIMULATED DATA. For each parameter, the results obtained nique resulted in the lowest mean error (Fp, vp, ve 25%, E
using different analysis methods were divided uniformly into 15 41%) while the direct four-parameter nonlinear fit
January 2017 243

TABLE 3. Parameter Estimation: Choosing the Correct Model
Mean error 6 SEM [%]
Full model Using the correct model

vp FP E ve BAT vp FP E ve BAT
2CXM11BAT2 26.41 61.54 21.72 613.09 19.98 617.12 27.29 60.94 8.70 64.50 26.41 61.54 21.72 613.09 19.98 617.12 27.29 60.94 8.70 64.50
2CXM 39.61 649.19 27.88 610.75 21.91 619.39 29.45 618.69 - 31.18 643.92 19.31 614.54 20.16 616.34 25.27 619.07 -
Uptake1BAT 85.72 695.97 51.75 613.69 54.13 644.04 - 13.03 610.43 34.71 623.25 52.61 610.69 56.82 650.21 - 12.37 68.72
Uptake 129.09 6131.29 56.20 616.83 49.25 619.28 - - 46.09 629.82 48.81 611.59 40.15 649.94 - -
3
High. Vasc. 1 35.03 68.95 20.92 615.24 - - 9.77 65.88 8.65 64.66 18.59 615.48 - - 10.95 65.01
BAT
High. Vasc. 61.36 617.16 24.11 621.78 - - - 4.17 64.02 24.41 613.38 - - -
High. Perf.41 46.67 69.78 - - - 11.30 67.30 1.63 61.21 - - - 12.60 66.85
BAT
High. Perf. 46.23 67.67 - - - - 0.64 60.42 - - - -
1
2XCM – Two Compartment Exchange Model.
2
BAT – Bolus Arrival Time.
3
High. Vasc. – Highly Vascularized.
4
High. Perf. – Highly Perfused.
Mean and standard error of the mean (SEM) [in %] of the 2CXM, compartment-uptake, highly vascularized, and highly perfused (with and without BAT) model parameters, estimated when assum-
ing a full model of 2CXM1BAT and when using the correct ground truth model.
Stage:

Page: 244
C
O
L
O
R
FIGURE 4: Estimation with different temporal resolutions. Using the spline with 2nd derivative method, the estimation process was
repeated for varying temporal resolutions of 2 (green), 4 (blue), and 6 (cyan) second intervals. Degradation is apparent as resolu-
tion gets lower. vp is more robust than the other parameters. Dashed black line 5 ground truth value.
resulted in the highest mean error (Fp, vp, ve 80%, E and when using the correct ground truth model. As seen, esti-
190%). Thus, we used the 2nd derivative regularized- mation accuracy is higher when using the correct ground truth
spline technique as the deconvolution method in ACoPeD, model. As expected, the estimation accuracy improves when
for the rest of the analyses. the ground truth model has fewer parameters.
BAT CORRECTION AND PARAMETER ESTIMATION. Figure EFFECT OF TEMPORAL RESOLUTION AND SCAN
F3 3 shows comparison of BAT and PK parameter estimation DURATION. Figure 4 shows the estimation error of the PK F4
using three different methods for BAT estimation, and with parameters as a function of the acquired temporal resolution.
no correction for BAT. ACoPeD, which includes the BAT Mean relative errors over the entire range for all PK parameters
correction procedure (described in Fig. 1), achieved the are given in Table 4. As can be seen, the error in estimation of T4
highest estimation accuracy for BAT parameters with mean Fp, E, and ve decreased when the temporal resolution was
error of 22.13 6 1.91, compared to 26.11 6 2.15 for the reduced from 6 to 2 seconds, and the differences are more pro-
LQ-model and 59.90 6 2.70 for the cyclic deconvolution nounced at higher values. However, the vp estimation seems to
MVT. All other PK parameters were also estimated with be less sensitive to the decrease in temporal resolution.
higher accuracy using ACoPeD compared to the other A longer scan duration, 20 minutes relative to 6
methods, with mean errors of: Fp: 24%, E: 55%, vp: 34%, minutes, was shown to improve the estimation of all PK
and ve: 32%, over the entire range of PK values. Note that parameters. However, as expected, parameters that relate to
Fig. 3 shows the accuracy of estimation in one parameter, the slow washout of contrast agent from extracellular space,
while the ground truth values of the other parameters are ie, E and ve are more sensitive to this factor (reduction in
uniformly sampled from the range described above (simulat- mean error: E: 12%, ve: 6%).
ed data). Relative errors over the entire range for the other
T2 BAT estimation methods are presented in Table 2. Patient Data
T3 CHOOSING THE CORRECT MODEL. Table 3 summarizes PK PARAMETER VALUES IN PATIENTS. ACoPeD was
the result of parameter estimation when assuming a full model used for estimation of parameters in seven patients: four
January 2017 245

TABLE 4. Effect of Temporal Resolution and Scan Duration
Temporal resolution Scan duration

2 sec 4 sec 6 sec 20 min 6 min
FP 24.82 6 0.67 39.18 6 0.67 47.67 6 0.68 21.84 6 0.43 24.82 6 0.67
E 41.72 6 3.20 51.96 6 3.5 84.66 6 3.5 29.77 6 3.11 41.72 6 3.20
VP 26.41 6 1.54 29.18 6 1.94 34.23 6 2.30 23.53 6 1.39 26.41 6 1.54
Ve 27.29 6 0.94 49.84 6 2.74 56.90 6 2.13 21.63 6 0.95 27.29 6 0.94
1
BAT - - - - -
Mean relative error and standard error of the mean (SEM) [in %] of the PK parameters, estimated using different acquisition parame-
ters and scan duration.
1
BAT – Bolus Arrival Time
patients with GB and three patients following stroke. Figure models without BAT are shown in Fig. 5F. The model selec-
F5 5 shows a representative example of parameter maps tion map segmented the brain into GM (gray), WM (blue),
obtained from a patient with GB (patient #4) using and tumor (red and yellow).
ACoPeD. In this case, only a negligible number of voxels Median values of the 2CXM parameters (after model
resulted in a model that included BAT; therefore, only selection process) for the NAGM, NAWM, and lesion for
C
O
L
O
R
FIGURE 5: Parameters estimation in a high-grade tumor. DCE-MRI PK parameters extracted using ACoPeD, from a patient with
brain tumor (B,C,E,H) on top of a postcontrast SPGR image (A). F: Model selection map. D,G: CBV and CBF maps extracted using
DSC-MRI.

each patient are summarized in Table 5. As expected, both T5
Lesion
Fp and vp are higher (1.5–2-fold) in the NAGM relative to
0.18
0.17
0.09
0.10
NA
NA
NA
the NAWM. The three patients with stroke, had no
enhanced lesions, thus no ve and E values were detected
within the lesion. All three patients were scanned at the sub-
NAWM
E [a.u.]
acute phase (5 days following the event) and showed some
NA
NA
NA
NA
NA
NA
NA
evidence for tissue reperfusion, thus the Fp and vp within
the lesion were similar to that of the NAWM. For the four
NAGM
patients with GB, all lesions were enhanced, and showed

high Fp, vp, ve, and E values.
NA
NA
NA
NA
NA
NA
NA
COMPARISON BETWEEN DSC-MRI AND DCE-MRI. Pre-
liminary results in two patients (who had DSC-MRI data):
Lesion
1.27
6.59
1.88
5.77
one following stroke and the other with GB, demonstrated
NA
NA
NA
correlation between DCE-MRI and DSC-MRI. Figure 5

ve [mL/100g]
shows an example of parameter maps obtained from a

NAWM
patient with GB using ACoPeD, along with CBV and CBF

maps obtained from DSC-MRI. Figure 6 shows parameter F6
NA
NA
NA
NA
NA
NA
NA
maps obtained from a patient following stroke; Fp and BAT

maps estimated using ACoPed and CBF obtained from
NAGM
DSC-MRI, along with DWI image. High BAT values can

Median value
NA
NA
NA
NA
NA
NA
NA
Median of the 2CXM parameters estimated from patients, three following stroke and four with gliobastoma (GB).
be seen within the stroke lesion.

A significant (P < 0.001) voxel-based correlation was
detected for blood flow between DCE-MRI and DSC-MRI
Lesion
in the patient following stroke (r 5 0.49). In the patient

1.20
1.08
1.22
6.45
6.57
3.84
2.72
with GB a weak correlation was detected (P < 0.001,

r 5 0.2). For blood volume, significant (P < 0.001) voxel-
vp [mL/100g]
NAWM
based correlations were detected between methods in the

patient with GB (r 5 0.62) and in the patient following
0.85
0.89
0.80
0.64
0.89
0.65
0.87
stroke (r 5 0.35).
Discussion
NAGM
1.20
1.05
1.50
1.40
1.10
0.82
0.94
This study proposes ACoPeD: an optimized method for tis-

sue perfusion and permeability estimation using DCE-MRI,
NAWM - contralesional hemisphere normal appearing white matter
based on 2CXM with spline-based deconvolution using 2nd

NAGM - contralesional hemisphere normal appearing grey matter
151.63
Lesion
derivative regularization, accounting for BAT, and incorpo-

34.30
31.25
26.00
58.83
65.53
31.51
rating model selection. The method was applied to simulat-

FP [mL/100g/min]
ed data, demonstrating the most accurate estimation of the

PK parameters compared to other methods tested. Direct
2
TABLE 5. Parameter Estimation in Patients
NAWM
four-parameter estimation resulted in poorer estimation

26.48
28.56
24.73
20.10
28.89
21.33
24.06
than the deconvolution based methods, which may stem

from the complexity of the nonlinear four-parameter space.
Incorporating the deconvolution technique results in a fair
1
NAGM
estimation of the plasma flow, and the direct three-

38.42
42.15
44.42
42.81
39.28
28.19
24.81
parameter estimation of the resulting IRF seems to improve

the estimation accuracy overall. High temporal resolution (2
sec) and long scan duration (20 min) were shown to provide
Patient
more accurate results on simulated data. Simulation results

showed that incorporating BAT and using the correct model
1
2
3
4
5
6
7
significantly improved PK parameter estimation.

The choice of the most suitable model was suggested
Stroke
by Ewing et al17 and Buckley and Sourbron7 among others

GB
1
2
to have an important role in the interpretation of the
January 2017 247

preferable in the case of a stroke over DCE-MRI. On the

contrary, in patients with brain tumors the information
regarding permeability is of great clinical importance, high
spatial resolution images are preferable due to the highly
heterogeneous nature of high-grade tumors, and susceptibili-
ty artifacts are more common. Using the 2CMX model ena-
bles acquisition of all vascular parameters in one method,
which requires only a single dose of contrast agent. Thus, in
patients with brain tumors, DCE-MRI using the 2CMX
model might be preferable over DSC-MRI. While this study
focused on cerebral perfusion, the quantification methods
may equally be applicable to other organs.
Several limitations to this study should be addressed
regarding technical and study design: Accurate estimation of
PK parameters was shown to require high temporal resolu-
tion. Acquiring high temporal and spatial resolution data
C currently results in reduced brain coverage. Therefore, our
O
L
results are based on only a few slices. However, with new
O MRI technologies that enable increased brain coverage with
R
high temporal resolution, this method may be routinely
FIGURE 6: Parameter estimation in stroke. DCE-MRI Fp and
BAT maps (B,C) extracted by the ACoPeD tool from a patient used for patient assessment and diagnosis. In the conversion
following stroke on top of a diffusion-weighted image (A). D: of signal to concentration, the precise influence of limited
CBF map extracted using DSC-MRI. water exchange27 was ignored. In addition, the postprocess-
ing time was relatively long. However, since the voxelwise
calculations are independent it can be easily parallelizable
extracted PK maps. Results from this study are in line with
for multicore processors or graphical processing units. With
their findings. In this study, models with and without BAT
regard to study design, this study shows results on a relative-
were selected according to the corrected Akaike information
ly small number of patients and the comparison to DSC-
criterion. BAT was shown to improve the estimation of the
PK parameters in simulated data. The method was applied MRI was performed in only two patients. Additional studies
to data of patients with brain tumors and patients following are needed to confirm and validate our results and evaluate
stroke. The model selection map segmented the brain tissue the contribution of this method in other neurological
in patients with brain tumor into GM, WM, and tumor disorders.
area, reflecting the different vascular properties of the nor- In conclusion, this study proposes the ACoPeD tool—
mal and pathological brain tissue components, and may a robust approach for flow estimation using DCE-MRI data
have a clinical role in patient assessment and follow-up. in clinical setup. An open source implementation is publicly
Incorporating flow in DCE-MRI analysis has been available, aiming to improve standardization of DCE-MRI
previously proposed,2–4,11 yet the applicability of such a and to enable the analysis of routinely acquired clinical data
method in clinical settings is still limited. Currently, DSC- with the 2CXM model.
MRI is the method of choice to detect cerebral blood flow
in patients following stroke and patients with brain tumors. Acknowledgments
However, there are several inherent drawbacks to the DSC-
Contract grant sponsor: Yitzhak and Chaya Weinstein
MRI method including the relatively low spatial resolution
Research Institute for Signal Processing
and high sensitivity to susceptibility artifacts, such as in
We thank Vicky Myers for editorial assistance. This work was
regions containing blood, calcification, metallic surgical
performed in partial fulfillment of the requirements for an
materials, and in areas adjacent to the temporal lobes or
MSc degree of Guy Nadav, School of Electrical Engineering,
skull base. In stroke patients, DSC-MRI provides the
Tel Aviv University, Israel.
required clinical information regarding the location and
amount of reduced perfusion with high sensitivity, and has
higher CNR in areas of WM relative to DCE-MRI. In
addition, susceptibility artifacts are less common in patients References
following stroke, and during the acute phase the lesion per- 1. Sourbron SP, Buckley DL. Classic models for dynamic contrast-
meability is intact. Therefore, DSC-MRI seems to be enhanced MRI. NMR Biomed 2013;26:1004–1027.

2. Larsson HB, Hansen AE, Berg HK, Rostrup E, Haraldseth O. Dynamic 15. Wu O, Østergaard L, Weisskoff RM, Benner T, Rosen BR, Sorensen
contrast-enhanced quantitative perfusion measurement of the brain AG. Tracer arrival timing-insensitive technique for estimating flow in
using T1-weighted MRI at 3T. J Magn Reson Imaging 2008;27:754–762. MR perfusion-weighted imaging using singular value decomposition
with a block-circulant deconvolution matrix. Magn Reson Med 2003;
3. Larsson HB, Courivaud F, Rostrup E, Hansen AE. Measurement of
50:164–174.
brain perfusion, blood volume, and blood-brain barrier permeability,
using dynamic contrast-enhanced T1-weighted MRI at 3 tesla. Magn 16. Liberman G, Louzoun Y, Artzi M, Nadav G, Ewing JR, Bashat
Reson Med 2009;62:1270–1281. DB. DUSTER: Dynamic contrast enhance up-sampled temporal
resolution analysis method. Magn Reson Imaging 2016;34:
4. Sourbron S, Ingrisch M, Siefert A, Reiser M, Herrmann K. Quantifica-
442–450.
tion of cerebral blood flow, cerebral blood volume, and blood-brain-
barrier leakage with DCE-MRI. Magn Reson Med 2009;62:205–217. 17. Cheong L, Koh T, Hou Z. An automatic approach for estimating bolus
arrival time in dynamic contrast MRI using piecewise continuous
5. Fluckiger JU, Schabel MC, DiBella EV. Toward local arterial input func-
regression models. Phys Med Biol 2003;48:N83–88.
tions in dynamic contrast-enhanced MRI. J Magn Reson Imaging
2010;32:924–934. 18. Bagher-Ebadian H, Jain R, Nejad-Davarani SP, et al. Model selection
for DCE-T1 studies in glioblastoma. Magn Reson Med 2012;68:
6. Liberman G, Louzoun Y, Colliot O, Ben Bashat D. T1 Mapping, AIF and
241–251.
pharmacokinetic parameter extraction from dynamic contrast enhance-
ment MRI data. Multimodal Brain Image Anal 2011;7012:76–83. 19. Parker GJ, Roberts C, Macdonald A, et al. Experimentally-derived
functional form for population-averaged high-temporal-resolution
7. Sourbron SP, Buckley DL. On the scope and interpretation of the
arterial input function for dynamic contrast-enhanced MRI. Magn
Tofts models for DCE-MRI. Magn Reson Med 2011;66:735–745.
Reson Med 2006;56:993–1000.
8. Ingrisch M, Sourbron S, Morhard D, et al. Quantification of perfusion
20. Bishop CM. Pattern recognition and machine learning. New York:
and permeability in multiple sclerosis: dynamic contrast-enhanced
Springer; 2006.
MRI in 3D at 3T. Invest Radiol 2012;47:252–258.
21. Gobbel G, Fike J. A deconvolution method for evaluating indicator-
9. Ahearn T, Staff R, Redpath T, Semple S. The use of the Levenberg-
dilution curves. Phys Med Biol 1994;39:1833–1854.
Marquardt curve-fitting algorithm in pharmacokinetic modelling of
DCE-MRI data. Phys Med Biol 2005;50:N85–92. 22. Liberman G, Nadav G, Louzoun Y, Artzi M, Ben Bashat D. Bolus arriv-
al time extraction using super temporal resolution analysis of DCE. In:
10. Irving B, Tanner L, Enescu M, et al. Personalised estimation of the
Proc 22nd Annual Meeting ISMRM, Milan; 2014.
arterial input function for improved pharmacokinetic modelling of
colorectal cancer using dceMRI. In: Abdominal imaging computation 23. Essig M, Nguyen TB, Shiroishi MS, et al. Perfusion MRI: the five most
and clinical applications. New York: Springer; 2013. p 126–135. frequently asked clinical questions. AJR Am J Roentgenol 2013;201:
W495–510.
11. Kershaw LE, Buckley DL. Precision in measurements of perfusion and
microvascular permeability with T1-weighted dynamic contrast- 24. Smith SM. Fast robust automated brain extraction. Hum Brain Mapp
enhanced MRI. Magn Reson Med 2006;56:986–992. 2002;17:143–155.
12. Calamante F, Willats L, Gadian DG, Connelly A. Bolus delay and dis- 25. Wells WMI, Viola P, Atsumi H, Nakajima S, Kikinis R. Multi-modal vol-
persion in perfusion MRI: implications for tissue predictor models in ume registration by maximization of mutual information. Med Imag
stroke. Magn Reson Med 2006;55:1180–1185. Anal 1996;1:35–51.
13. Østergaard L, Weisskoff RM, Chesler DA, Gyldensted C, Rosen BR. 26. Liberman G, Louzoun Y, Ben Bashat D. T1 Mapping using variable flip
High resolution measurement of cerebral blood flow using intravascu- angle SPGR data with flip angle correction. J Magn Reson Imaging
lar tracer bolus passages. Part I: Mathematical approach and statisti- 2014;40:171–180.
cal analysis. Magn Reson Med 1996;36:715–725.
27. Buckley DL, Kershaw LE, Stanisz GJ. Cellular-interstitial water
14. Calamante F, Gadian DG, Connelly A. Delay and dispersion effects in exchange and its effect on the determination of contrast agent con-
dynamic susceptibility contrast MRI: simulations using singular value centration in vivo: Dynamic contrast-enhanced MRI of human internal
decomposition. Magn Reson Med 2000;44:466–473. obturator muscle. Magn Reson Med 2008;60:1011–1019.
January 2017 249

ORIGINAL RESEARCH
Application of Texture Analysis on

Parametric T1 and T2 Maps for
Detection of Hepatic Fibrosis
HeiShun Yu, MD,1* Anne-Sophie Touret, MS,1 Baojun Li, PhD,1
Michael O’Brien, MD,2 Muhammad M. Qureshi, MBBS,1 Jorge A. Soto, MD,1
Hernan Jara, PhD,1 and Stephan W. Anderson, MD1
Purpose: To assess the utility of texture analysis of T1 and T2 maps for the detection of hepatic fibrosis in a murine
model of hepatic fibrosis.
Materials and Methods: Following Institutional Animal Care and Use Committee approval, a dietary model of hepatic
fibrosis was used and 15 ex vivo murine livers were examined. Images were acquired using a 30 mm bore 11.7T mag-
netic resonance imaging (MRI) scanner with a rapid acquisition with relaxation enhancement sequence. Texture analysis
was then employed, extracting texture features including histogram-based, gray-level co-occurrence matrix-based
(GLCM), gray-level run-length-based features (GLRL), gray-level gradient matrix (GLGM), and Laws’ features. Areas
under the curve (AUCs) were then calculated to determine the ability of texture features to detect hepatic fibrosis.
Results: Texture analysis of T1 maps identified very good to excellent discriminators of hepatic fibrosis within the histo-
gram and GLGM categories. Histogram feature interquartile range (IQR) achieved an AUC value of 0.90 (P < 0.0001)
and GLGM feature variance gradient achieved an AUC of 0.91 (P < 0.0001). Texture analysis of T2 maps identified very
good to excellent discriminators of hepatic fibrosis within the histogram, GLCM, GLRL, and GLGM categories. GLGM
feature kurtosis was the best discriminator of hepatic fibrosis, achieving an AUC value of 0.90 (P < 0.0001).
Conclusion: This study demonstrates the utility of texture analysis for the detection of hepatic fibrosis when applied to
T1 and T2 maps in a murine model of hepatic fibrosis and validates the potential use of this technique for the noninva-
sive, quantitative assessment of hepatic fibrosis.
H epatic fibrosis is a reparative process in which excess

proteins, such as collagen, accumulate in the extracellu-
lar space secondary to chronic insults.1,2 Common etiologies
limitations. The limitations of this technique include sam-
pling error, as fibrosis is a heterogeneous process, as well as
subjectivity in histopathologic analysis resulting in intra-
of hepatic fibrosis include alcohol abuse, viral hepatitis, and and interobserver variability.5,6 Another limitation includes
nonalcoholic steatohepatitis (NASH).2 Left untreated, exces- inherent potential complications associated with the biopsy,
sive accumulation of extracellular matrix proteins results in including patient discomfort, bleeding, and infection.1,3–5,7
distortion of hepatic architecture through the formation of The limitations of liver biopsy have necessitated devel-
fibrous scar tissue and nodular regeneration of hepatocytes, opment of alternative means to assess hepatic fibrosis. Mul-
which may ultimately lead to hepatic cirrhosis.3,4 tiple noninvasive techniques including serological and
Early diagnosis and treatment is essential to prevent urinary markers as well as imaging have been explored as
the progression to endstage liver disease. The current gold possible alternatives to percutaneous biopsy.1,7 Serological
standard for the detection and assessment of fibrosis is non- markers have been shown to be effective in the detection of
targeted, percutaneous core liver biopsy, which has inherent advanced fibrosis but are limited in the detection of
Received Mar 8, 2016, and in revised form May 18, 2016. Accepted for publication May 18, 2016.
*Address reprint requests to: H.S.Y., Boston University Medical Center, Department of Radiology, 820 Harrison Ave., FGH Building, 3rd Fl., Boston, MA
02118. E-mail: heishun.yu@bmc.org
From the 1Department of Radiology, Boston University Medical Center, Boston, Massachusetts, USA; and 2Department of Pathology and Laboratory
Medicine, Boston University Medical Center, Boston, Massachusetts, USA

V

Yu et al.: Texture Analysis on Parametric T1 and T2
intermediate grades of fibrosis. Another limitation is that NEX 5 2 and the IR-DE-TSE (TI 5 400 msec, TR 5 4.4 sec,
many of these markers are nonspecific and can be altered by and NEX 5 2).
other pathologic conditions.7 In the imaging realm, conven-
Image Segmentation
tional ultrasound imaging is generally limited to the identi-
Segmentation of liver specimens was performed using an in-house-
fication of advanced cases of liver disease. Magnetic
developed dual-clustering segmentation algorithm using MathCad
resonance imaging (MRI) and ultrasound (US) elastography (PTC, Needham, MA), which is a semiautomated method that has
for the assessment of liver stiffness has also been studied previously been described in the literature.17–19
recently as a potential method of assessing fibrosis,8–11
although both require specialized hardware and/or software Image Analysis
for operation and may not be routinely available. T1 and T2 maps were computed with model-conforming qMRI
An image postprocessing technique that is currently algorithms developed in-house based on solutions to the Bloch
being widely explored is texture analysis, which employs a equations, as applicable to the combined IR-DE-TSE and DE-TSE
mathematical and statistical analysis of pixel values relative pulse sequences applied sequentially with identical prescan
parameters.19
to neighboring pixels. This technique is useful, as it allows
Texture analysis was performed using an in-house-developed
for quantification of disease, can be applied to many imag-
MatLab-based program (MathWorks, Natick, MA), which
ing modalities, and does not require the acquisition of addi-
extracted 41 texture features from the liver segments. Texture fea-
tional images.12–15 To this end, the purpose of this study tures included 12 histogram features, 5 gray-level co-occurrence
was to explore the utility of texture analysis for the assess- matrix features (GLCM), 11 gray-level run-length features
ment of hepatic fibrosis when applied to T1- and T2- (GLRL), 4 gray-level gradient matrix (GLGM) features, and 9
weighted images in a murine model of hepatic fibrosis. Laws’ features. Histogram-based features included in the analysis
were mean, median, standard deviation, range, geometric mean,
Materials and Methods harmonic mean, 2nd standard deviation, standard deviation (in a 5
Study Design and 9 pixel neighborhood), 4th moment, IQR, and entropy. Unlike
This study was approved by the Institutional Animal Care and Use all other texture features, these features are spatially invariant,
Committee (IACUC) at our institution. The study included 15, 6- meaning that the arrangement of the pixels relative to one another
week-old male C57BL/6 mice (Charles River Laboratories, Wil- does not affect the analysis.12,13,20
mington, MA), which were divided into two groups. The first The GLCM features employed included contrast, correlation,
group was a control group, which consisted of two mice that were energy, entropy, and homogeneity.12,13 The GLCM is a symmetric
fed normal chow and were sacrificed immediately before initiating matrix with rows (i) and columns (j) ranging from 0 to Ng, where
the experimental diet. The experimental group was fed a 0.1% (w/ Ng is equal to the number of gray-levels within the image. GLCM
w) 3,5-diethoxycarbonyl-1,4-dihydrocollidine (DDC)-supple- elements in row i and column j represent the frequency in which a
mented diet (TestDiet, Richmond, IN) in order to induce hepatic given gray level of value i is horizontally adjacent to gray level j.
fibrosis.16 This diet was continued for 14 weeks, sacrificing approx- Such calculations can be performed in any direction and are spa-
imately one mouse each week to produce varying levels of hepatic tially dependent.12,13,20 For the purposes of this work, these calcu-
fibrosis. Immediately after sacrifice, the livers were extracted from lations were performed in vertical, horizontal, 458, and 1358
the mice and fixed in formalin for 18 hours and stored in directions, which were then averaged together to minimize direc-
phosphate-buffered saline (PBS, pH 5 7.4). tional dependence in the samples.
Gray-level run-length features included in the analysis were
Image Acquisition short run emphasis (SRE), long run emphasis (LRE), gray-level
Imaging was performed with an 11.7 T MRI scanner (Bruker Bio- nonuniformity (GLN), run-length nonuniformity (RLN), run per-
spin, Billerica, MA). Scanning of the fixed liver specimens was percentage (RP), low gray-level run emphasis (LGRE), high gray-level
formed using a 20-mm transmit/receive quadrature radiofrequency run emphasis (HGRE), short run low gray-level emphasis
(RF) coil. Specimens were placed inside a 15-mm glass vial and (SRLGE), short run high gray-level emphasis (SRHGE), long run
secured to minimize vibrations. The temperature was controlled low gray-level emphasis (LRLGE), and long run high gray-level
using an MRI-compatible temperature sensor to ensure that the emphasis (LRHGE). In this analysis, the rows (i) represent the
temperature was 22.5 6 18C and that there were no major fluctua- gray-levels while the columns (j) represent the run-length, or the
tions. The imaging protocol included two double-echo turbo spin- consecutive number of pixels with a particular gray-level. Elements
echo that were run in tandem with identical receiver and gain set- within the matrix represent the frequency of pixel line segments
ting; one with (IR-DE-TSE) and one without an inversion pulse with a run-length j and a gray-level i.12,13,20
(DE-TSE). These two sequences were implemented with the same The GLGM features included in the analysis were mean,
geometry parameters (16 transverse contiguous slices, voxel 5 117 variance, kurtosis, and skewness. As previously described in the lit-
3 117 3 700 lm3, 128 3 128 matrix) and identical dual-echo erature, these features provide a histogram of the absolute gradient
TSE readout parameters (ETL 5 4, ESP 5 6.8 msec, TE1eff 5 values in the liver segment.15,20 In this analysis, differences in all
8.855 msec, TE2eff 5 21.615 msec). The other key contrast and pixel values within a liver segment were analyzed using a 3 3 3
signal-to-noise ratio (SNR) parameters were TR 5 4 seconds and neighborhood.
January 2017 251

FIGURE 1: Representative parametric T1 and T2 maps of segmented mouse livers with T1 maps on top and T2 maps on bottom. A:
Fibrosis Grade 0: No fibrosis (DIA with 2.01% area of fibrosis). B: Fibrosis Grade 2: Periportal fibrosis or rare septa (DIA with
19.22% area of fibrosis). C: Fibrosis Grade 4: Widespread fibrosis and cirrhosis (DIA with 33.70% area of fibrosis). D: Fibrosis
Grade 0: No fibrosis (DIA with 2.01% area of fibrosis). E: Fibrosis Grade 2: Periportal fibrosis or rare septa (DIA with 19.22% area
of fibrosis). F: Fibrosis Grade 4: Widespread fibrosis and cirrhosis (DIA with 33.70% area of fibrosis).
Laws’ features L1, L2, L3, L4, L5, L6, L7, L8, and L9 were which stained blue using Masson’s trichrome, were expressed as a
included in the analysis. These are based on multiplication of two percentage of the total area of liver in the slide.
one-dimensional vectors, which were initially described by Kenneth
Laws. Laws described one-dimensional vectors including level, Statistical Analysis
spot, edge, and spot.15,21 Two-dimensional 5 3 5 pixel filter masks Texture features derived from the T1 and T2 maps of the liver
are created by multiplication of these vectors, which are applied specimens were plotted against digital image analysis (DIA) of
across an image. hepatic fibrosis. Areas under the curve (AUC) were calculated to
determine the ability of texture analysis to differentiate between
varying degrees of fibrosis. Two analyses were performed by divid-
Histologic Analysis ing specimens into <10.1% vs. 10.1% and <19.0% vs.
Liver specimens were embedded in paraffin and sliced into 5-lm 19.0%. Thresholds were chosen to approximate the digital image
sections. These sections were stained with hematoxylin and eosin analysis cutoffs for differentiating Ishak fibrosis scores 0–2 from 3–
and Masson’s trichrome stains. A single section through the entire 6 and 0–3 from 4–6.22,23 AUCs were classified as follows: 0.50–
liver was analyzed by a board-certified pathologist (M.O.) with 0.59, poor; 0.60–0.69, fair; 0.70–0.79, good; 0.80–0.89, very
over 20 years of experience in the pathology of chronic liver dis- good; 0.90–1.0, excellent.
ease. Specimens were assessed for the degree of fibrosis using a
scale of 0 to 4 such that 0 represented no fibrosis, 1 represented Results
portal fibrosis, 2 represented periportal fibrosis or rare septa, 3 rep-
resented septal fibrosis and architectural distortion, and 4 repre-
Histopathology
sented widespread fibrosis and cirrhosis. Subsequently, liver
The DDC-supplemented diet resulted in varying degrees of
specimens were digitized using a digital slide scanner (ScanScope hepatic fibrosis (Fig. 1), which were previously reported. In F1
CS, Aperio Technologies, Vista, CA). An automated software control specimens, digital image analysis determined per-
(Image-Pro Plus, Media Cybernetics, Bethesda, MD) was employed centage areas of fibrosis of 3.24% and 2.01%. In experi-
to perform a color-based segmentation of the digitized images, mental specimens, DIA determined percentage areas of
which converted the image into a binary mask. Regions of fibrosis, fibrosis ranging from 6.54% up to 33.70%.

TABLE 1. AUCs of Histogram Features for the Detection of Hepatic Fibrosis on T1 and T2 maps
<10.1 vs. 10.1 <19.0 vs. 19.0

T1 T2 T1 T2
Mean 0.88 (P < 0.0001) 0.83 (P < 0.0001) 0.80 (P < 0.0001) 0.80 (P < 0.0001)
Median 0.85 (P < 0.0001) 0.83 (P < 0.0001) 0.78 (P < 0.0001) 0.80 (P < 0.0001)
STD 0.67 (P < 0.0001) 0.70 (P < 0.0001) 0.69 (P < 0.0001) 0.69 (P < 0.0001)
Range 0.76 (P < 0.0001) 0.76 (P < 0.0001) 0.73 (P < 0.0001) 0.74 (P < 0.0001)
Geometric mean 0.87 (P < 0.0001) 0.83 (P < 0.0001) 0.81 (P < 0.0001) 0.79 (P < 0.0001)
Harmonic mean 0.88 (P < 0.0001) 0.83 (P < 0.0001) 0.81 (P < 0.0001) 0.80 (P < 0.0001)
nd
2 STD 0.80 (P < 0.0001) 0.79 (P < 0.0001) 0.76 (P < 0.0001) 0.76 (P < 0.0001)
STD5 0.72 (P < 0.0001) 0.79 (P < 0.0001) 0.71 (P < 0.0001) 0.76 (P < 0.0001)
STD9 0.51 (P 5 0.855) 0.75 (P < 0.0001) 0.59 (P 5 0.041) 0.74 (P < 0.0001)
th
4 moment 0.51 (P 5 0.785) 0.56 (P 5 0.153) 0.60 (P 5 0.018) 0.56 (P 5 0.189)
IQR 0.90 (P < 0.0001) 0.79 (P < 0.0001) 0.80 (P < 0.0001) 0.69 (P < 0.0001)
Entropy 0.61 (P 5 0.030) 0.81 (P < 0.0001) 0.50 (P 5 0.945) 0.67 (P 5 0.0001)
AUC 5 area under curve, STD 5 standard deviation, STD5 5 5-pixel neighborhood standard deviation, STD9 5 9-pixel neighbor-
hood standard deviation IQR 5 interquartile range.
Texture Analysis: Histogram Features reported as AUCs in Table 4. Analysis of GLCM features T4
The ability of histogram features to detect hepatic fibrosis on T1 maps using a 10.1% area of fibrosis threshold was
T1 on T1 and T2 maps are reported in Table 1. In general, fair to good. Correlation and contrast had the highest AUC
increasing hepatic fibrosis corresponded with decreased values of 0.73 (P < 0.0001) and 0.71 (P < 0.0001) (Table
mean, median, geometric mean, harmonic mean, and IQR 2). Using a threshold of 19.0% area of fibrosis on T1 maps,
on T1 maps. Similar findings were seen on T2 maps. GLCM features were poor to fair with contrast demonstrat-
When using 10.1% area of fibrosis as a threshold, ing the highest AUC of 0.65 (P 5 0.0002).
AUC values of histogram features on T1 maps were very Analysis of T2 maps of GLCM features using a thresh-
good to excellent. Mean, median, geometric mean, har- old of 10.1% area of fibrosis demonstrated good to very
monic mean, and 2D STD demonstrated very good AUC good AUC values. Entropy and contrast demonstrated the
values ranging from 0.80 (P < 0.0001) to 0.88 (P < highest AUC values of 0.82 (P < 0.0001) and 0.80 (P <
0.0001). The IQR had an excellent AUC value of 0.90 (P 0.0001), respectively (Table 3). Using a threshold of 19.0%
T2 < 0.0001). These results are reported in Table 2. Using area of fibrosis, GLCM features were fair to good, with the
19.0% area of fibrosis as a threshold, AUC values of T1 best discriminating feature being contrast, which demon-
maps were very good. Mean, geometric mean, harmonic strated an AUC of 0.73 (P < 0.0001).
mean, and IQR demonstrated AUC values from 0.80 (P <
0.0001) to 0.81 (P < 0.0001). Texture Analysis: GLRL Features
AUC values of histogram features on the T2 maps The AUC values of gray-level run-length features of T1 and
were very good when using 10.1% area of fibrosis as a T2 maps are reported in Table 5. Analysis of T1 maps with T5
threshold. Mean, median, geometric mean, harmonic mean, GLRL features using a threshold of 10.1% area of fibrosis
and entropy had AUC values of 0.81 (P < 0.0001) to 0.83 demonstrated good AUC values. The best discriminators
T3 (P < 0.0001). These results are reported in Table 3. When were SRHGE, LRLGE, and LRHGE, with AUC values of
using 19.0% area of fibrosis as a threshold, histogram fea- 0.73 (P < 0.0001), 0.72 (P < 0.0001), and 0.72 (P <
tures on T2 maps were very good, with mean, median, and 0.0001), respectively (Table 2). Using a threshold of 19.0%
harmonic mean demonstrating AUC values of 0.80 (P < area of fibrosis, GLRL features were poor, with AUC values
0.0001). ranging from 0.50 (P 5 0.954) to 0.54 (P 5 0.423).
When using 10.1% area of fibrosis as a threshold,
Texture Analysis: GLCM Features analysis of T2 maps demonstrated very good AUC values
Results of gray-level co-occurrence matrix features for the ranging from 0.80 (P < 0.0001) to 0.86 (P < 0.0001).
detection of hepatic fibrosis on T1 and T2 maps are High gray-level run emphasis was the best discriminator,
January 2017 253

254
TABLE 2. Texture Analysis Results for Selected Texture Features With the Highest AUC Values on T1 Maps Using 10.1% Threshold
DIA GLCM GLRL Laws’ Feature

(% area Feature Feature
Histogram Features GLGM Features
fibrosis)
Mean Median 2D STD Geometric Harmonic IQR Correlation SRHGE L6 MGR VGR
Mean Mean
2.01 782.64 777.50 53.10 786.88 781.44 112.70 0.65 6439.82 76021.09 217.46 381282.87
3.24 657.14 640.09 31.17 676.66 668.82 86.91 0.61 1008.18 59814.52 190.04 485195.97
6.54 642.43 635.50 41.93 657.14 651.55 60.60 0.38 1071.93 181370.70 82.70 180688.21
9.03 657.16 650.00 47.13 672.14 665.41 74.42 0.33 947.13 279615.48 76.46 179498.35
14.48 635.85 627.82 40.86 652.14 646.63 57.36 0.36 699.70 194855.73 60.40 134594.52
16.56 650.77 647.23 33.00 660.33 656.74 56.42 0.37 564.36 163256.98 47.49 101304.02
19.22 627.27 623.23 28.82 637.69 634.71 43.19 0.41 812.74 144818.19 48.35 97988.72
19.32 635.70 631.50 17.35 642.52 640.37 49.25 0.79 2963.42 11204.95 130.47 201878.87
19.97 650.30 643.33 41.96 667.72 661.17 69.22 0.46 1007.87 169210.54 89.86 210820.14
22.32 647.00 641.30 35.60 659.42 655.04 61.30 0.39 765.86 193126.79 57.02 127406.30
22.42 645.24 639.50 35.59 654.67 650.87 59.75 0.31 1157.57 146236.51 71.03 145294.84
24.03 661.39 652.55 49.75 677.10 670.86 79.70 0.40 1040.79 255311.56 72.50 163879.62
25.07 631.08 627.92 30.13 637.51 634.99 47.46 0.35 1867.30 135687.75 61.04 114582.27
28.82 652.86 643.83 51.04 672.50 664.47 74.17 0.33 899.38 360617.83 77.74 194281.88
33.70 630.84 626.54 30.00 641.23 638.30 46.46 0.45 1049.82 128128.96 51.73 103226.72
AUC 5 area under curve, DIA 5 digital image analysis, GLCM 5 gray-level co-occurrence matrix, GLRL 5 gray-level run-length, GLGM 5 gray-level gradient matrix, 2D STD 5 2-
dimensional standard deviation, IQR 5 interquartile range, SRHGE 5 short run high gray-level emphasis, L 5 Laws’ feature, MGR 5 mean gradient, VGR 5 variance gradient.
Stage:

Volume 45, No. 1
Page: 254
January 2017
TABLE 3. Texture Analysis Results for Selected Texture Features With the Highest AUC Values on T2 Maps Using 10.1% Threshold
DIA (% GLRL Laws’

area fibrosis) Feature Feature
Histogram Features GLCM Feature GLGM Features
Mean Median Entropy Geometric Harmonic Entropy Contrast HGRE L4 MGR VGR Skewness Kurtosis
Mean Mean
2.01 57.25 49.57 6.05 56.08 51.69 1.33 40.35 52.20 103779.92 41.68 8363.76 2.93 13.61
3.24 63.33 61.23 5.38 64.06 62.86 0.86 8.23 62.83 26894.20 18.33 3556.79 5.10 32.57
6.54 27.97 27.50 3.98 28.13 27.82 1.57 20.77 80.70 14908.08 3.69 327.13 6.13 42.47
9.03 30.76 29.67 4.18 31.66 30.40 0.32 18.59 15.03 63087.86 5.76 1527.99 12.98 271.92
14.48 35.86 35.18 4.36 36.80 35.70 0.23 11.79 11.86 42340.39 5.10 1236.32 14.02 316.80
16.56 30.72 30.38 3.78 31.12 30.63 0.29 8.47 13.71 22024.39 2.92 490.88 17.62 556.15
19.22 34.83 33.92 4.49 35.44 34.59 0.55 7.40 15.41 26546.78 3.85 592.98 12.53 296.07
19.32 27.79 26.58 4.69 28.41 27.43 0.62 3.00 41.72 11009.47 10.44 1018.66 6.09 52.11
19.97 38.77 37.89 4.61 39.78 38.45 0.26 14.03 15.39 55267.15 7.35 1831.47 10.63 167.20
22.32 32.54 31.70 4.26 33.04 32.14 0.24 16.30 10.04 40834.00 4.13 752.27 14.43 460.32
22.42 25.80 25.20 4.11 26.11 25.46 0.22 4.18 7.69 14560.93 4.06 523.10 12.20 291.78
24.03 43.98 42.64 5.01 44.82 43.23 0.50 16.04 13.90 73298.40 6.85 1549.45 9.50 140.38
25.07 22.47 21.77 4.00 22.89 22.35 1.24 12.88 32.14 10692.37 3.21 302.18 7.69 74.47
28.82 27.43 26.67 4.01 28.39 27.21 0.15 26.17 18.77 69564.63 5.31 1766.02 15.73 346.51
33.70 16.26 15.92 3.64 16.57 16.21 0.76 7.61 24.71 6343.70 1.82 121.06 8.68 90.80
AUC 5 area under curve, DIA 5 digital image analysis, GLCM 5 gray-level co-occurrence matrix, GLRL 5 gray-level run-length, GLGM 5 gray-level gradient matrix, HGRE 5 high
gray-level run emphasis, L 5 Laws’ feature, MGR 5 mean gradient, VGR 5 variance gradient.
Stage:

255
Page: 255
TABLE 4. AUCs of GLCM Features for the Detection of Hepatic Fibrosis on T1 and T2 Maps
<10.1 vs. 10.1 <19.0 vs. 19.0

T1 T2 T1 T2
Entropy 0.63 (P 5 0.010) 0.82 (P < 0.0001) 0.53 (P 5 0.489) 0.64 (P 5 0.002)
Contrast 0.71 (P < 0.0001) 0.80 (P < 0.0001) 0.65 (P 5 0.0002) 0.73 (P < 0.0001)
Correlation 0.73 (P < 0.0001) 0.61 (P 5 0.008) 0.60 (P 5 0.032) 0.57 (P 5 0.111)
Energy 0.65 (P 5 0.004) 0.79 (P < 0.0001) 0.53 (P 5 0.491) 0.62 (P 5 0.008)
Homogeneity 0.60 (P 5 0.048) 0.72 (P < 0.0001) 0.52 (P 5 0.745) 0.58 (P 5 0.077)
AUC 5 area under curve, GLCM 5 gray-level co-occurrence matrix.
with an AUC value of 0.86 (P < 0.0001) (Table 3). Using as a threshold, L4 was the best discriminator, with an AUC
19.0% area of fibrosis as a threshold, the analysis demon- of 0.77 (P < 0.0001) (Table 3). When using 19.0% area of
strated fair AUC values, with the best discriminating feature fibrosis as a threshold, AUC values were good. L1, L3, L4,
being HGRE, which had an AUC value of 0.69 (P < and L9 were the best discriminators, with an AUC of 0.74
0.0001). (P < 0.0001).
Texture Analysis: Laws’ Features

AUC values of Laws’ features of T1 and T2 are reported in Texture Analysis: GLGM Features
T6 Table 6. The analysis of T1 maps with Laws’ features using The results of GLGM features for the assessment of hepatic
a threshold of 10.1% area of fibrosis demonstrated fair to fibrosis on T1 and T2 maps are reported in Table 7. The T7
good AUC values. L6 was the best discriminator, with an analysis of T1 maps using Laws’ features and a threshold of
AUC of 0.70 (P < 0.0001) (Table 2). When using 19.0% 10.1% area of fibrosis were very good to excellent. Variance
area of fibrosis as a threshold, AUC values were poor to gradient (VGR) was the best discriminator, with an AUC of
fair. AUC values ranged from 0.57 (P < 0.0001) to 0.63 (P 0.91 (P < 0.0001) (Table 2). Mean gradient (MGR) was a
< 0.0001). very good discriminator, with an AUC of 0.89 (P <
The ability of Laws’ features to detect hepatic fibrosis 0.0001) (Table 2). When analyzing T1 maps with a thresh-
on T2 maps was good. When using 10.1% area of fibrosis old of 19.0% area of fibrosis, VGR and MGR were still the
TABLE 5. AUCs of GLRL Features for the Detection of Hepatic Fibrosis on T1 and T2 Maps
<10.1 vs. 10.1 <19.0 vs. 19.0

T1 T2 T1 T2
SRE 0.61 (P 5 0.015) 0.80 (P < 0.0001) 0.53 (P 5 0.464) 0.62 (P 5 0.005)
LRE 0.61 (P 5 0.017) 0.80 (P < 0.0001) 0.53 (P 5 0.481) 0.62 (P 5 0.004)
GLN 0.61 (P 5 0.013) 0.81 (P < 0.0001) 0.53 (P 5 0.448) 0.62 (P 5 0.004)
RLN 0.61 (P 5 0.017) 0.80 (P < 0.0001) 0.53 (P 5 0.484) 0.62 (P 5 0.006)
RP 0.57 (P 5 0.135) 0.85 (P < 0.0001) 0.50 (P 5 0.954) 0.68 (P < 0.0001)
LGRE 0.57 (P 5 0.137) 0.85 (P < 0.0001) 0.50 (P 5 0.923) 0.68 (P < 0.0001)
HGRE 0.57 (P 5 0.136) 0.86 (P < 0.0001) 0.50 (P 5 0.920) 0.69 (P < 0.0001)
SRLGE 0.57 (P 5 0.139) 0.85 (P < 0.0001) 0.50 (P 5 0.927) 0.68 (P < 0.0001)
SRHGE 0.73 (P < 0.0001) 0.85 (P < 0.0001) 0.54 (P 5 0.423) 0.64 (P 5 0.001)
LRLGE 0.72 P < 0.0001) 0.84 (P < 0.0001) 0.53 (P 5 0.500) 0.63 (P 5 0.002)
LRHGE 0.72 (P < 0.0001) 0.84 (P < 0.0001) 0.53 (P 5 0.469) 0.64 (P 5 0.001)
AUC 5 area under curve, GLRL 5 gray-level run-length, SRE 5 short run emphasis, LRE 5 long run emphasis, GLN 5 gray-level
nonuniformity, RLN 5 run-length non-uniformity, RP 5 run percentage, LGRE 5 low gray-level run emphasis, HGRE 5 high
gray-level run emphasis, SRLGE 5 short run low gray-level emphasis, SRHGE 5 short run high gray-level emphasis, LRLGE 5
long run low gray-level emphasis, LRHGE 5 long run high gray-level emphasis.

TABLE 6. AUCs of Laws’ Features for the Detection of Hepatic Fibrosis on T1 and T2 Maps
<10.1 vs. 10.1 <19.0 vs. 19.0

T1 T2 T1 T2
L1 0.52 (P 5 0.629) 0.76 (P < 0.0001) 0.57 (P 5 0.119) 0.74 (P < 0.0001)
L2 0.66 (P 5 0.001) 0.72 (P < 0.0001) 0.56 (P 5 0.145) 0.71 (P < 0.0001)
L3 0.61 (P 5 0.016) 0.75 (P < 0.0001) 0.63 (P 5 0.002) 0.74 (P < 0.0001)
L4 0.60 (P 5 0.018) 0.77 (P < 0.0001) 0.62 (P 5 0.006) 0.74 (P < 0.0001)
L5 0.69 (P < 0.0001) 0.67 (P < 0.0001) 0.58 (P 5 0.080) 0.69 (P < 0.0001)
L6 0.70 (P < 0.0001) 0.68 (P < 0.0001) 0.57 (P 5 0.099) 0.69 (P < 0.0001)
L7 0.59 (P 5 0.040) 0.73 (P < 0.0001) 0.63 (P 5 0.003) 0.72 (P < 0.0001)
L8 0.68 (P 5 0.0002) 0.62 (P 5 0.011) 0.57 (P 5 0.126) 0.64 (P 5 0.001)
L9 0.60 (P 5 0.021) 0.75 (P < 0.0001) 0.63 (P 5 0.002) 0.74 (P < 0.0001)
AUC 5 area under curve, L 5 Laws’ feature.
best discriminators, with good AUC values of 0.73 (P < of hepatic fibrosis among histogram and GLGM features.
0.0001) and 0.70 (P < 0.0001), respectively. Excellent discriminators included histogram feature IQR as
Analysis of T2 maps using Laws’ features and a 10.1% well as GLGM feature VGR. Several very good discrimina-
area of fibrosis threshold demonstrated very good to excel- tors of hepatic fibrosis were identified in the histogram and
lent AUC values. The best discriminator of fibrosis was kur- GLGM categories. Texture analysis of T2 maps identified
tosis, with an AUC of 0.90 (P < 0.0001) (Table 3). very good and excellent discriminators of hepatic fibrosis.
Skewness, MGR, and VGR were very good discriminators Gray-level gradient matrix feature kurtosis was an excellent
of fibrosis, with AUC values of 0.89 (P < 0.0001), 0.85 (P discriminator of fibrosis. Very good discriminators were
< 0.0001), and 0.83 (P < 0.0001) (Table 3). When using identified in the histogram, GLCM, GLRL, and GLGM
a threshold of 19.0% area of fibrosis, VGR and MGR were categories.
the best discriminators of fibrosis and demonstrated good Texture analysis is an image processing technique that
AUC values of 0.74 (P < 0.0001) and 0.70 (P < 0.0001). extracts textural information from images by studying the
interrelationships of adjacent pixels within an image. This
Discussion technique allows for noninvasive quantification of texture
In this study we evaluated the utility of texture analysis of and mitigates subjectivity.24 An additional advantage of this
T1 and T2 maps to differentiate varying degrees of hepatic technique is that it does not increase imaging time, as this
fibrosis in a murine model of liver disease. This study uti- is a method of postprocessing. Such analyses have demon-
lized high-field MRI in a highly controlled environment to strated utility for the detection of subtle differences in signal
minimize confounding factors during the ex vivo imaging. intensities that are otherwise not detectable by the human
The results of this study demonstrated excellent discrimina- eye.15,24 This is seen in the case of hepatic fibrosis, where
tion of varying degrees of hepatic fibrosis. Texture analysis early disease takes place at the microscopic level and is not
of T1 maps identified very good and excellent discriminators readily detected.15 In a similar experiment studying diffuse
TABLE 7. AUCs of GLGM Features for the Detection of Hepatic Fibrosis on T1 and T2 Maps
<10.1 vs. 10.1 <19.0 vs. 19.0

T1 T2 T1 T2
MGR 0.89 (P < 0.0001) 0.85 (P < 0.0001) 0.70 (P < 0.0001) 0.70 (P < 0.0001)
VGR 0.91 (P < 0.0001) 0.83 (P < 0.0001) 0.73 (P < 0.0001) 0.74 (P < 0.0001)
Skewness 0.76 (P < 0.0001) 0.89 (P < 0.0001) 0.58 (P 5 0.176) 0.68 (P 5 0.0001)
Kurtosis 0.73 (P < 0.0001) 0.90 (P < 0.0001) 0.55 (P 5 0.262) 0.69 (P < 0.0001)
AUC 5 area under curve, GLGM 5 gray-level gradient matrix, MGR 5 mean gradient, VGR 5 variance gradient.
January 2017 257

liver disease and cirrhosis, the authors were able to identify sis in a highly controlled experimental setting using an ex
a few texture features, including histogram features mean, vivo murine model of liver disease. Several texture features
standard deviation, and entropy, which were accurate in were identified as excellent discriminators of varying degrees
detecting hepatic fibrosis.14 of hepatic fibrosis, supporting the application of such tech-
The use of texture analysis has been studied in other niques as a noninvasive assessment of hepatic fibrosis in the
organ systems, including head and neck, cartilage, lung, and setting of chronic liver disease.
many others.25–27 In a recent study, texture analysis was
applied to contrast-enhanced computed tomography (CT) References
images of oropharyngeal squamous cell carcinomas. Those 1. Afdhal NH, Nunes D. Evaluation of liver fibrosis: a concise review. Am
J Gastroenterol 2004;99:1160–1174.
authors identified features, including histogram features
2. Bataller R, Brenner DA. Liver fibrosis. J Clin Invest 2005;115:209–218.
median and entropy, as well as GLCM entropy, which were
able to differentiate between human papillomavirus-positive 3. Lurie Y, Webb M, Cytter-Kuint R, Shteingart S, Lederkremer GZ. Non-
invasive diagnosis of liver fibrosis and cirrhosis. World J Gastroenterol
versus human papilloma virus-negative oropharyngeal can- 2015;21:11567–11583.
cers.25 In musculoskeletal imaging, texture analysis was 4. Friedman SL. Liver fibrosis—from bench to bedside. J Hepatol 2003;
applied to T2 maps of cartilage to assess for differences in 38:S38–S53.
control patients versus those with risk factors for osteoarthri- 5. Regev A, Berho M, Jeffers LJ, et al. Sampling error and intraobserver
variation in liver biopsy in patients with chronic HCV infection. Am J
tis. Those authors identified a few texture features including
Gastroenterol 2002;97:2614–2618.
mean T2 values, GLCM contrast, and GLCM variance that
6. Westin J, Lagging L, Wejstal R, Norkrans G, Dhillon A. Interobserver
were elevated in patients with risk factors for osteoarthri- study of liver histopathology using the Ishak score in patients with
tis.26 In another study, the authors used texture analysis on chronic hepatitis C virus infection. Liver 1999;19:183–187.
unenhanced CT images of lung tumors and determined that 7. Fallatah HI. Noninvasive biomarkers of liver fibrosis?: An overview.
Adv Hepatol 2014;1–15.
tumor heterogeneity has the potential to be used as an inde-
8. Venkatesh SK, Ehman RL. Magnetic resonance elastography of abdo-
pendent predictor of survival in patients with nonsmall-cell
men. Abdom Imaging 2014;40:745–759.
lung cancer.27
9. Venkatesh SK, Yin M, Ehman RL. Magnetic resonance elastography of
The main limitation of our study was that imaging liver: technique, analysis, and clinical applications. JMRI 2013;37:544–555.
was performed on formalin-fixed and ex vivo specimens. 10. Rouviere O, Yin M, Dresner M, et al. MR elastography of the liver:
Imaging of ex vivo specimens introduces variables that may preliminary results. Radiology 2006;240:440–448.
confound results. In a prior study, the authors demonstrated 11. Talwalkar JA, Kurtz DM, Schoenleber SJ, West CP, Montori VM. Ultra-
that quantitative MRI parameters may be altered in sacri- sound-based transient elastography for the detection of hepatic fibro-
sis?: Systematic review and meta-analysis. Clin Gastroenterol Hepatol
ficed animals, which may be secondary to ischemia.28 More 2007;5:1214–1220.
specifically, mean signal intensity in pre- and postsacrificed 12. Barry B, Buch K, Soto JA, Jara H, Nakhmani A, Anderson SW. Quanti-
specimens demonstrated differences on T1, T2, and T2* fying liver fibrosis through the application of texture analysis to diffu-
sion weighted imaging. Magn Reson Imaging 2014;32:84–90.
maps. Another confounding variable that is introduced in
13. Anderson SW, Jara H, Ozonoff A, O’Brien M, Hamilton JA, Soto JA.
the ex vivo specimens is caused by formalin fixation. In Effect of disease progression on liver apparent diffusion coefficient
prior studies, formalin fixation caused alterations in T1, T2, and T2 values in a murine model of hepatic fibrosis at 11.7 Tesla MRI.
and proton density maps, while sparing ADC maps. Pro- J Magn Reson Imaging 2012;35:140–146.
posed mechanisms of these changes suggest hemiacetal-like 14. Hshiao M-H, Chen P-C, Jao J-C, et al. Quantifying liver cirrhosis by
extracting significant features from MRI T2 image. Sci World J 2012;
adducts formed when formaldehyde reacts with peptide 2012:1–5.
linkages in the tissue. Such changes are thought to be revers- 15. Yu H, Buch K, Li B, et al. Utility of texture analysis for quantifying
ible, especially in the first 24 hours of fixation, but can hepatic fibrosis on proton density MRI. J Magn Reson Imaging 2015;
42:1259–1265.
become stable over time.29–31 However, while the ex vivo
experiments that were performed were inherently artificial, 16. Alcoholism VC, Portal I, Mouse N, et al. Experimental liver fibrosis
research: Update on animal models, legal issues and translational
the environment in which the experiment was performed aspects. Fibrogenes Tissue Repair 2013;6:1–24.
was highly controlled. All specimens underwent the same 17. Watanabe M, Sakai O, Norbash AM, Jara H. Accurate brain volumetry
fixation process and were imaged using the same protocol in with diffusion-weighted spin-echo single-shot echo-planar-imaging
and dual-clustering segmentation: Comparison with volumetry-
a temperature-controlled environment. Therefore, the effects validated quantitative magnetic resonance imaging. Med Phys 2010;
of disease progression on the T1 and T2 maps and their sub- 37:1183–1190.
sequently derived texture features are hypothesized to pre- 18. Farraher SW, Jara H, Chang KJ, Hou A, Soto JA. Liver and spleen
dominantly represent differences in the degree of fibrosis volumetry with quantitative MR imaging and dual-space clustering
segmentation. Radiology 2005;237:322–328.
within the tissue specimens.
19. Jara H. Theory of quantitative magnetic resonance imaging. Hacken-
In conclusion, our study provides a proof of concept sack, NJ: World Scientific; 2013.
for the use of texture analysis applied to T1 and T2 maps 20. Castellano G, Bonilha L, Li L, Cendes F. Texture analysis of medical
for the noninvasive, quantitative assessment of hepatic fibro- images. Clin Radiol 2004;59:1061–1069.

21. Laws KI. Textured image segmentation. Los Angeles: USC LA; 1980. controls—data from the osteoarthritis initiative. Arthritis Res Ther
p 1–195. 2011;13:R153–R164.
22. Campos C, Paiva D, Perazzo H, et al. An inexpensive and worldwide 27. Ganeshan B, Panayiotou E, Burnand K, Dizdarevic S, Miles K. Tumour
available digital image analysis technique for histological fibrosis heterogeneity in non-small cell lung carcinoma assessed by CT
quantification in chronic hepatitis C. J Viral Hepat 2014;21:216–222. texture analysis?: A potential marker of survival. Eur Radiol 2012;22:
796–802.
23. O’Brien MJ, Keating NM, Elderiny S, et al. An assessment of digital
image analysis to measure fibrosis in liver biopsy specimens of 28. Shuter B, Tofts P, Pope J. The effect of sacrifice on image signal, T1,
patients with chronic hepatitis C. Anat Pathol 2000;712–718. T2, and T*2 in liver, kidney, and brain of the Wistar rat. Magn Reson
Imaging 1995;13:563–574.
24. Jir
ak D, Dezortova M, Taimr P, H
ajek M. Texture analysis of human
liver. J Magn Reson Imaging 2002;15:68–74. 29. Yong-Hing CJ, Obenaus A, Stryker R, Tong K, Sarty GE. Magnetic
resonance imaging and mathematical modeling of progressive forma-
25. Buch K, Fujita A, Li B, Kawashima Y, Qureshi M, Sakai O. Using tex-
lin fixation of the human brain. Magn Reson Med 2005;54:324–332.
ture analysis to determine human papillomavirus status of oropharyn-
geal squamous cell carcinomas on CT. Am J Neuroradiol 2015;36: 30. Shepherd TM, Thelwall PE, Stanisz GJ, Blackband SJ. Aldehyde fixa-
1343–1348. tive solutions alter the water relaxation and diffusion properties of
nervous tissue. Magn Reson Med 2009;62:26–34.
26. Joseph GB, Baum T, Carballido-Gamio J, et al. Texture analysis of
cartilage T2 maps: individuals with risk factors for OA have higher 31. Tovi M, Ericsson A. Measurements of T1 and T2 over time in formalin-
and more heterogeneous knee cartilage MR T2 compared to normal fixed human whole-brain specimens. Acta Radiol 1992;33:400–404.
January 2017 259

ORIGINAL RESEARCH
Intravoxel Incoherent Motion

Diffusion-Weighted Imaging of the
Pancreas: Characterization of Benign
and Malignant Pancreatic Pathologies
Bohyun Kim MD, PhD,1 Seung Soo Lee MD, PhD,2* Yu Sub Sung PhD,2
Hyunhee Cheong BA,2 Jae Ho Byun MD, PhD,2 Hyoung Jung Kim MD, PhD,2 and
Jin Hee Kim MD, PhD2
Purpose: To evaluate the diagnostic value of apparent diffusion coefficient (ADC) and intravoxel incoherent motion
(IVIM) parameters in differentiating patients with either a normal pancreas (NP), pancreatic ductal adenocarcinoma
(PDAC), neuroendocrine tumor (NET), solid pseudopapillary tumor (SPT), acute pancreatitis (AcP), vs. autoimmune pan-
creatitis (AIP).
Materials and Methods: In all, 84 pathologically confirmed pancreatic tumors (60 PDACs, 15 NETs, 9 SPTs), 20 pancre-
atitis (13 AcPs, 7 AIPs), and 30 NP subjects underwent IVIM diffusion-weighted imaging using 10 b-values (0–900 sec/
mm2) at 1.5T. The ADC, pure molecular diffusion coefficient (Dslow), perfusion fraction (f), and perfusion-related diffusion
coefficient (Dfast) were calculated and compared using a Kruskal–Wallis test and post-hoc Dunn procedure. Receiver
operating characteristic (ROC) analysis was performed to assess diagnostic performance.
Results: The f and Dfast of the PDAC were significantly lower than that of the NP (f 5 0.10 vs. 0.24; Dfast 5 42.21 vs. 71.74 3
1023mm2/sec; P < 0.05). In ROC analysis, f showed the best diagnostic performance (area-under-the-curve, 0.919) among all
parameters in differentiating PDAC from NP (P 0.001). The f values of AcP (0.11) and AIP (0.13) and the Dfast values of
SPT (20.48 3 1023mm2/sec) and AcP (24.49 3 1023mm2/sec) were significantly lower compared with NP (f 5 0.24;
Dfast 5 71.74 3 1023mm2/sec; P < 0.05). For NET, the f (0.21) was significantly higher than that of PDAC (0.10, P < 0.01).
Conclusion: Perfusion-related parameters f and Dfast are more helpful in characterizing pancreatic diseases than ADC or
Dslow. The PDCA, SPT, AcP, and AIP were characterized by reduced f and Dfast values compared with normal pancreas.
The f value might help in differentiating between PDAC and NET.
D iffusion-weighted imaging (DWI) has been increasingly

utilized as a magnetic resonance imaging (MRI)
sequence for evaluating the pancreas.1–5 Quantitative evalua-
molecular diffusion and microcirculatory blood perfusion, it
may not fully account for the tissue characteristics.
On the other hand, intravoxel incoherent motion
tion of DWI can provide additional information about the (IVIM) imaging is an imaging method based on DWI with
mobility of water molecules in pancreatic lesions, as indi- multiple b-values. It allows for the separate analysis of pure
cated by apparent diffusion coefficient (ADC). Earlier stud- molecular diffusion and microcirculation (or perfusion),
ies had demonstrated that ADC might help in with the IVIM-derived parameters of pure molecular diffu-
characterizing solid tumors, inflammation, and cystic lesions sion coefficient (Dslow), perfusion fraction (f ), and
in the pancreas.2,6–8 However, since ADC represents the perfusion-related diffusion coefficient (Dfast).9–11 Several
water mobility as a whole in a tissue and is affected by both prior studies have shown potential values of IVIM in
*Address reprint requests to: S.S.L., Department of Radiology and Research Institute of Radiology, University of Ulsan College of Medicine, Asan Medical
Center, 86 Olympic-to, 43-Gil, Songpa-Gu, Seoul 138-736, Korea. E-mail: seungsoolee@amc.seoul.kr
From the 1Department of Radiology, Ajou University Hospital, Suwon, South Korea; and 2Department of Radiology and Research Institute of Radiology,
Asan Medical Center, University of Ulsan College of Medicine, Seoul, South Korea

V

Kim et al.: IVIM DWI of the Pancreas
characterizing pancreatic neoplasms and in differentiating dedicated six-channel, torso-array coil. The maximum gradient
pancreatic cancer from focal pancreatitis.6,12–14 However, specifications were 45 mT/m for amplitude and 200 mT/m/s for
there still have been relatively small numbers of published the slew rate.
studies regarding this issue, mostly focused on the common The routine biliary-pancreas MRI protocol consisted of
pancreatic diseases such as pancreatic cancer, neuroendocrine breath-hold, dual gradient-echo T1-weighted imaging, navigator-
tumor (NET), and chronic pancreatitis. triggered T2-weighted turbo spin-echo (TSE) imaging, breath-hold
2D thick slab and navigator-triggered 3D TSE MR cholangiopan-
Therefore, the purpose of this study was to evaluate the
creaticography imaging, dynamic contrast-enhanced T1-weighted
diagnostic value of ADC and IVIM parameters in differenti-
imaging with a 3D gradient echo sequence (volumetric interpolated
ating normal pancreas (NP), pancreatic ductal adenocarci- breath-hold examination [VIBE]; Siemens Medical Solutions) after
noma (PDAC), NET, solid pseudopapillary tumor (SPT), a bolus injection of 15 mL of gadoterate meglumine (Dotarem,
acute pancreatitis (AcP), and autoimmune pancreatitis (AIP). Guerbet, Roissy, France) at a rate of 2 mL/sec.
Axial IVIM DW images were acquired by using a single-shot
Materials and Methods echo-planar imaging sequence in a free-breathing manner with the
Study Population following parameters: echo time, 60 msec; repetition time, 2000
Our Institutional Review Board approved this retrospective study msec; echo-planar imaging factor, 115; receiver bandwidth,
and waived the requirement of informed consent. From January 1736Hz per pixel; field of view, 367 3 276 mm; matrix size, 144
2014 to September 2014, a total of 654 patients underwent biliary- 3 192; number of averages, 4; section thickness, 6 mm; intersec-
pancreas MR examinations including IVIM DWI at our institution. tion gap, 1.2 mm; number of sections, 20; and acquisition time,
Among them, 312 patients who underwent MRI for biliary or duo- 3.5 minutes. DW gradients (ie, 10 b-values of 0, 10, 25, 50, 75,
denal malignancy, biliary stone disease, and pancreatic cystic lesions 100, 200, 400, 600, and 900 sec/mm2) were applied in three
were excluded. We also excluded 238 patients who had pancreatic orthogonal directions and were subsequently averaged. Fat suppres-
masses without pathologic diagnosis (n 5 79) or masses that were sion was achieved using a chemical shift-selective, fat-suppression
smaller than 2 cm in diameter (n 5 159). Hence, 84 consecutive technique. A k-space-based, parallel imaging technique (generalized
patients with pathologically confirmed solid pancreatic neoplasm autocalibrating partially parallel acquisition, GRAPPA, Siemens
2 cm in diameter, including PDCA (n 5 60), NET (n 5 15), and Medical Solutions) was used with an acceleration factor of two.
SPT (n 5 9), and 20 patients with clinically or pathologically diag-
nosed AcP (n 5 13) and AIP (n 5 7), were enrolled in this study. IVIM DW Image Analysis
For the control group, 30 subjects with normal pancreas were also DWI data were quantitatively analyzed according to monoexpo-
included in this study based on the following criteria: 1) those who nential and biexponential models using in-house software that was
underwent MRI for the evaluation of benign gallbladder or bile developed as a plug-in functions to ImageJ software (NIH,
duct diseases; 2) no documented history of pancreatic disease; 3) Bethesda, MD). For the monoexponential model, the ADC was
normal results of serum amylase and lipase; and 4) no abnormal calculated by least-square, monoexponential fit of the entire data
finding in the pancreas on MRI as determined by an abdominal from all b-values on a pixel-by-pixel basis according to the follow-
radiologist (S.S.L.). The diagnoses of all solid pancreatic neoplasms ing equation:
were pathologically proven following either surgical resection
(n 5 40) or endoscopic ultrasonography (EUS)-guided needle biopsy SI5SI0 expð2b ADCÞ (1)
(n 5 44). The diagnosis of AcP was made by EUS-guided needle
where SI is the signal intensity at a given b-value and SI0 is the sig-
biopsy and clinical features for two patients or by unequivocal imag-
nal intensity for b 5 0 sec/mm.2
ing and clinical features, ie, characteristic imaging findings, abdomi-
In the biexponential model, the DW signal decay as a func-
nal pain, elevated serum amylase and lipase levels three or more
tion of b-values was modeled according to the following equation
times than normal, and resolution of such findings during follow-
based on the IVIM theory:
up,15 for 11 patients. The diagnosis of AIP was made pathologically
by needle biopsy (n 5 5) or clinically (n 5 2) according to the inter- SI 5SI0 ½ð12f Þ expð2b Dslow Þ1f expð2b Dfast Þ (2)
national consensus diagnostic criteria.16 Of 84 patients with solid
pancreatic neoplasms, 12 patients with PDCA had metastases to the In order to avoid mathematical instability of simultaneous fitting of
liver (n 5 9), peritoneum (n 5 2), and lung (n 5 1), while no patient all three parameters, we adopted a segmented analysis procedure as
with NET and SPT had metastasis. used in previous studies.17–20 First, under the assumption of the
The study population included the pancreatic disease group negligible contribution of the perfusion component at the range of
composed of 60 males and 44 females with mean age of 57.2 6 14 high b-values, Dslow was determined from monoexponential data fit-
years (range 17–79 years) and the control normal pancreas group ting of b 200 sec/mm2 data according to the following equation:
consisting of 16 males and 14 females with mean age of 49.7 6 13
years (range 14–80 years). S5Sint e 2bslow (3)
MR Examination where Sint is the b 5 0 intercept of the monoexponential fit of high

All MRI studies were performed on a 1.5T scanner (Magnetom b-value data. Then f can be estimated according to the following
Avanto; Siemens Medical Solutions, Erlangen, Germany) using a equation:
January 2017 261

f 5ðS0 2Sint Þ=S0 (4)
P valuea
< 0.001
< 0.001
0.626
0.014
With the Dslow and f, Dfast being calculated using a partially con-
Data are median values. Numbers in parentheses are ranges. NP, normal pancreas; PDAC, pancreatic ductal adenocarcinoma; NET, neuroendocrine tumor; SPT, solid pseudopapillary
strained, nonlinear regression of all datasets according to Eq. 2. The
IVIM data fitting processes were performed on a pixel-by-pixel basis.
34.91 (12.73-58.60)
One board-certified abdominal radiologist (B.K.) with 7
1.43 (0.95-1.75) years of experience in abdominal MRI interpretation who was
1.11 (0.91-1.41)
0.13 (0.05-0.23)
blinded to the final diagnoses of the pancreatic lesions performed
AIP (n 5 7)
quantitative image analysis on an in-house plug-in package written

in ImageJ that was implemented with the aforementioned monoex-
ponential and biexponential IVIM fitting algorithms. Using nonen-
hanced T1- and T2-weighted images and contrast-enhanced T1-
24.49 (8.55-131.90) weighted images of the pancreas as anatomic references to identify
the extent of the lesion, freehand regions on interest (ROIs) were
1.49 (1.14-2.45)
1.19 (0.90-1.47)
0.11 (0.03-0.33)
drawn on the solid portion of the pancreatic tumors on two con-

AcP (n 5 13)
secutive slices of DW images at a b-value which best revealed the

tumor margin, while avoiding the vessels, pancreatic duct, and
necrotic/cystic components. The software automatically colocalized
these ROIs on the corresponding sections of DW images at all b-
values, performed the pixel-by-pixel data fitting, and generated
parametric maps. Then the average parametric values within the
20.48 (6.55-72.29)
ROIs were obtained. The process was identical for the control
1.49 (1.00-2.50)
1.05 (0.51-1.84)
0.10 (0.03-0.45)
group, except that the ROIs were drawn on the normal pancreatic
SPT (n 5 9)
parenchyma in the head portion to encompass as large area possi-

ble while excluding the pancreatic duct. In order to assess the reli-
TABLE 1. ADC and IVIM Parameters of Normal Pancreas and Various Pancreatic Diseases
ability of the measurement, the entire procedures including ROI

drawing and data fitting were repeated by the same radiologist fol-
P values were obtained from the Kruskal-Wallis test for differences in ADC and IVIM parameters.
lowing a 1-month washout period. The values from the first mea-
30.90 (7.36-93.93)
surement were used as representative parameters and were used for

1.50 (1.15-2.39)
1.08 (0.88-1.65)
0.21 (0.05-0.46)
all statistical analysis aside from the intraobserver variability assess-

NET (n 5 15)
ment, which was based on the first and the second measurements.
Differences in mean age and sex between the patient and control
groups were tested by independent-sample t-test and the chi-square
test, respectively. For the comparison of the ADC, Dslow, f, and Dfast,
42.21 (5.70-316.73)
the values were first tested for equality of standard deviation and
1.56 (1.04-2.23)
1.25 (0.73-1.96)
0.10 (0.01-0.31)
PDAC (n 5 60)
Gaussian distribution with Bartlett’s test and Kolmogorov–Smirnov

test, respectively. The IVIM parameters among NP, PDAC, NET,
tumor; AcP, acute pancreatitis; AIP, autoimmune pancreatitis.
SPT, AcP, and AIP were then compared using the Kruskal–Wallis
test and post-hoc Dunn procedure. We performed receiver operating
characteristics (ROC) analysis to assess the diagnostic performance
of the ADC value and IVIM parameters. The areas under ROC
71.74 (19.28-244.49)
curve (AUC) were obtained in each parameter, and the sensitivity

and the specificity of each parameter were calculated at the optimal
1.14 (0.89-1.52)
0.24 (0.15-0.36)
1.51 (1.18-2.2)
cutoff values which were determined by the point of the largest You-
NP (n 5 30)
den index.21 Intraobserver reliability of the measurements was

assessed by calculating the intraclass correlation coefficient (ICC).
An ICC greater than 0.75 was considered to reflect good agree-
ment.22 P < 0.05 was considered statistically significant. The statis-
tical analysis was performed using GraphPad InStat v. 3.10 for
Windows and MedCalc (MedCalc Software, Mariakerke, Belgium).
(1023mm2/sec)
(1023mm2/sec)
(1023mm2/sec)
Parameters
Results
Patient Characteristics
ADC
Dslow
Dfast
The mean age of the patient group was significantly greater

f
than that of the control group (57.2 vs. 49.7 years, P 5 0.009).

C
O
L
O
R
FIGURE 1: Boxplots of ADC (a), Dslow (b), f (c), and Dfast (d) for normal pancreas (NP), pancreatic ductal adenocarcinoma (PDAC),
neuroendocrine tumor (NET), solid pseudopapillary tumor (SPT), acute pancreatitis (AcP), and autoimmune pancreatitis (AIP), and
the P values obtained by post-hoc Dunn procedure. (a,b) There was no significant difference in ADC (a) and Dslow (b) among nor-
mal pancreas and pancreatic diseases. (c) PDAC showed significantly lower f than NP and NET. The f value in AcP was significantly
lower than in NP. (d) Dfast of PDAC, NET, SPT, and AcP were significantly lower than that of the NP.
There was no significant difference in the sex distribution cantly lower f (P < 0.001) and Dfast (P < 0.05) than NP
between two groups (P 5 0.682). The mean size 6 standard (Fig. 2). Regarding NET, the f value tended to be higher than F2
deviation (SD) of the pancreatic neoplasm was 3.8 6 1.5 cm those of PDAC, SPT, AcP, and AIP, although a statistically sig-
(range 2–9 cm) for PDAC, 3.7 6 1.7 cm (range 2–9.7 cm) for nificant difference was only noted between NET and PDAC
NET, and 5.7 6 1.7 cm (range 2–9.4 cm) for SPT. (P < 0.01). NET had significantly lower Dfast than NP (P <
0.01) (Fig. 3). SPT showed a tendency toward lower Dslow, f, F3
ADC and IVIM Parameters and Dfast values than NP, but the statistically significant differ-
The mean, SD, and range of ROI size were: 13.3 6 ence was noted only for Dfast (P < 0.01). AcP showed signifi-
11.3 cm2 (3.9–84.5 cm2) for PDCA, 14.4 6 13.5 cm2 (2.6– cantly lower f (P < 0.05) and Dfast (P < 0.01) than NP. AIP
40.3 cm2) for NET, 58.1 6 45.4 cm2 (4.3–122 cm2) for also showed significantly lower f than NP (P < 0.001).
SPT, 13.8 6 7.8 cm2 (5.4–27.9 cm2) for AcP, and 14.2 6
5.9 cm2 (6.9–22.3 cm2) for AIP. Diagnostic Performance of ADC and IVIM
T1 ADC and IVIM parameters for NP, PDAC, NET, SPT, Parameters
F1 AcP, and AIP are summarized in Table 1 and Fig. 1. The Table 3 summarizes the diagnostic performance of ADC T3
T2 results of post-hoc pairwise comparisons are shown in Table 2. and IVIM parameters in differentiating among pancreatic
There were statistically significant differences in Dslow pathologies and NP. The ROC analysis revealed that f had
(P 5 0.014), f (P < 0.001), and Dfast (P < 0.001) among NP the highest AUC among ADC and IVIM parameters for
and various pancreatic diseases, while ADC was not signifi- differentiating PDCA (AUC 5 0.919), AcP (AUC 5 0.828),
cantly different among them (P 5 0.626). PDAC had signifi- and AIP (AUC 5 0.933) from NP, with sensitivities of
January 2017 263

85.0% to 86.7% and specificities of 73.3% to 100% at the
Data are P values. NS indicates nonsignificant difference (P > 0.05). NP, normal pancreas; PDAC, pancreatic ductal adenocarcinoma; NET, neuroendocrine tumor; SPT, solid pseudopapil-
< 0.05
vs. NP
optimal cutoff values. In pairwise comparison, the AUC of f
AIP
NS
NS
for differentiating between PDCA and NP was significantly
higher than those of ADC (P < 0.001), Dslow (P < 0.001),
vs. AIP
and Dfast (P 5 0.001). For differentiating AcP from NP, the
AcP AUCs of f (0.815) and Dfast (0.828) were significantly
NS
NS
NS
higher than that of ADC (0.585) (P 5 0.003 for ADC vs. f;
< 0.05
< 0.01
vs. NP
P 5 0.030 for ADC vs. Dfast). The AUC of f for the differ-
AcP
NS entiation between AIP and NP was significantly larger than

that of Dslow (P 5 0.046). For the differentiation of NET
vs. AIP
and SPT from NP, the AUC (0.838 for NET and 0.896 for
SPT
SPT) of Dfast was the highest among all parameters and was
NS
NS
NS
significantly higher than those of ADC (P 5 0.049 for NET
and 0.005 for SPT) and Dslow (P 5 0.015 for NET and
vs. AcP
0.004 for SPT). In the differentiation between PDCA and

SPT
NS
NS
NS
NET, the AUC of f (0.830) was the highest, with statisti-

cally significant differences compared with those of ADC
< 0.01
vs. NP
(P 5 0.03), and Dfast (P 5 0.02).

SPT
NS
NS
Overall, intraobserver agreement was good except for

Dfast. The ICCs were 0.935 (95% confidence interval [CI],
vs. AIP
0.905, 0.955) for ADC, 0.864 (95% CI, 0.806, 0.906) for
NET
NS
NS
NS
Dslow, 0.824 (95% CI, 0.751, 0.877) for f, and 0.313 (95%
CI, 0.129, 0.476) for Dfast.
vs. AcP
NET
Discussion
NS
NS
NS
Our study demonstrated that the perfusion-related parame-

vs. SPT
ters f and Dfast are more helpful in characterizing pancreatic

NET
diseases than ADC or Dslow. The PDCA was characterized

NS
NS
NS
by significantly lower f and Dfast values than NP, while

< 0.01
vs. NP
ADC and Dslow were not helpful in differentiating between

NET
them. The ROC analysis demonstrated the highest diagnos-

NS
NS
tic performance of f in differentiating PDAC from NP. Our

vs. AIP
findings coincide with the results of previous studies which

PDAC
revealed that f was reduced in PDAC than in normal pan-

NS
NS
NS
creatic tissue.6,12,23,24 In addition, our study demonstrated

lary tumor; AcP, acute pancreatitis; AIP, autoimmune pancreatitis.
that SPT, AcP, and AIP also had reduced f values compared
vs. AcP
PDAC
TABLE 2. Results of the Post-Hoc Dunn Procedure
with NP. Unlike other pancreatic diseases, NET had rela-

NS
NS
NS
tively high f value, which was similar to the f value of NP.

The f of the NET was significantly higher than that of the
vs. SPT
PDAC
PDAC and the AUC of f was the highest among all param-
NS
NS
NS
eters for differentiating between NET and PDCA. These

results in our current study correspond to those of previous
vs. NET
studies,12,25 which consistently reported reduced f values of

< 0.01
PDAC
PDCA compared with NET.

NS
NS
According to IVIM theory,10 perfusion fraction f rep-

< 0.001
resents the fractional volume of flowing water within a given

< 0.05
vs. NP
PDAC
image voxel. This theoretic assumption of f has been shown

NS
in previous phantom and animal experiments.26,27 There-

fore, the difference of f among the pancreatic diseases and
Parameters
the normal pancreas in our study may have been attributed

to the difference in tissue perfusion. The f values of the vari-
Dslow
Dfast
ous pancreatic diseases in our study are consistent with their

f
known tissue characteristics and enhancement patterns on

C
O
L
O
R
FIGURE 2: A 50-year-old male with PDAC. (a) Axial contrast-enhanced T1-weighted 3D gradient echo image (TE 5 1.4 msec,
TR 5 4.1 msec, and flip angle 5 108) in the arterial phase shows an ill-defined hypovascular mass in the pancreas body. (b) On axial
diffusion-weighted image (b 5 200 mm/sec2, the mass shows intermediate high signal intensity. (c) On Dslow parametric map, the
mass (arrow) showed slightly reduced value (Dslow 5 1.25 3 1023 mm2/sec). (d) On f parametric map, the mass (arrow) showed
markedly reduced f value (f 5 0.08) compared to normal pancreatic parenchyma.
computed tomography (CT) and MRI.28–32 NETs typically sis, the AUC of the Dfast was the highest among IVIM
show hypervascular enhancement due to rich capillary net- parameters in differentiating NET from NP and AcP from
works, while PDACs tend to be hypovascular due to malig- NP. Despite these promising results shown in our study, we
nant cell infiltration in the abundant desmoplastic stromas, are not confident about whether Dfast can be used as an
as well as decreased vascularity.6,12,28,29 The SPT, AcP, and imaging biomarker in clinical practice, as this parameter
AIP, which also showed reduced f value than NP in our showed a poor intraobserver agreement in our study. The
present study, are characterized by a reduced enhancement poor measurement reliability of Dfast has also been demon-
at the arterial phase compared with normal pancreas.30–32 A strated in a few previous studies.27,33–35 In the liver, the 95%
recent study25 demonstrated a correlation between f and his- confidence limit of Dfast has been reported to be as high as
tologic microvessel density in PDCA and NET, indicating 20 times the measured Dfast value, which would far exceed a
the dependency of f on the degree of tumor vascularity. clinically acceptable measurement error range.34 Our finding
Although our study revealed a significant difference in f showing poor intraobserver agreement of Dfast measurement
between PDCA and NP and a high AUC of f in distinction suggests that the clinical utility of Dfast in the pancreas may
between them, there was a considerable overlap in the f val- also be hampered by its limited measurement reliability.
ues for PDCA and those for NP. Given that, it should be In our study, ADC and Dslow did not allow for differ-
determined in a further research whether f truly has an entiation among the pancreatic diseases and normal pan-
incremental value in differentiating between PDCA and NP creas. Regarding the differentiation of PDCA from NP,
than does conventional anatomic imaging. conflicting results have been reported in the previous stud-
In terms of Dfast, PDAC, NET, SPT, and AcP were ies. Although earlier studies using two or three b-values con-
shown to have reduced Dfast than in NP. In the ROC analy- sistently reported significantly lower ADC in PDCA than
January 2017 265

C
O
L
O
R
FIGURE 3: A 63-year-old female with grade 2 NET. (a) Axial contrast enhanced T1-weighted 3D gradient echo image (TE 5 1.4
msec, TR 5 4.1 msec, and flip angle 5 108) in the arterial phase shows a hypervascular mass in the pancreas head. (b) On axial
diffusion-weighted image (b 5 200 mm/sec2, the mass appears slightly hyperintense to the normal pancreatic parenchyma. (c) On
Dslow parametric map, the mass (arrow) showed slightly reduced value (Dslow 5 1.27 3 1023 mm2/sec). (d) On f parametric map,
the mass (arrow) corresponds to high f value (f 5 0.28) which is similar to the surrounding pancreas parenchyma.
NP,6,14 the results were inconsistent in the recent studies forming pancreatitis without a statistically significant differ-
using 10 or more b-values.12,23,24 The reasons for the dis- ence. However, Klauss et al13 reported contrary results using
crepant results in the current and previous studies are poorly 11 b-values. In their study, ADC50-300 and f of the chronic
understood, but may include the different contribution of pancreatitis were significantly higher than those of PDAC. In
perfusion on ADC depending on the b-values used for another prior study, chronic pancreatitis had significantly
ADC calculation, uneven degree of necrosis and/or fibrosis higher and f and Dfast values than in PDAC.12 We assume
in PDAC across the studies, as well as potential sampling that conflicting results from ours and previous other studies
bias in the studies, owing to moderate number of lesions. may have stemmed from a different stage of pancreatitis
On the other hand, the current and previous studies consis- among the studies. Our study included AcP with active
tently reported no significant difference in the Dslow value inflammation and elevated serum amylase, while the previous
between PDCA and NP but significantly lower f value in studies mostly included chronic pancreatitis. Furthermore,
PDAC than in the normal pancreatic parenchyma, indicat- unlike our study, where AIP was separately evaluated, the pre-
ing that the difference in perfusion rather than diffusion vious study by Lee et al6 included both chronic pancreatitis
characterizes PDCA compared with NP. In the current and AIP in the category of mass-forming pancreatitis. Further
study, AcP and AIP showed decreased f and Dfast compared studies with a more clear definition of pancreatitis in terms
with NP, but there was no significant difference in any of of etiology and the stage of inflammation could confidently
the DWI-derived parameters between PDAC and AcP or disclose the usefulness of DWI-derived parameters in differ-
AIP. Our findings were at variance with the results of previ- entiating focal pancreatitis from PDAC.
ous studies that compared focal pancreatitis and PDAC. Lee In the current study we used a pixel-by-pixel fitting
et al6 reported that ADC and Dslow of mass-forming pancrea- method where fitting is first performed to obtain ADC and
titis were significantly lower than those of PDAC and that IVIM parameters in each pixel, followed by calculating the
the f of PDAC tended to be higher than that of mass- mean value of the pixels encompassed in the ROIs drawn

TABLE 3. Results of ROC Analysis for IVIM-Derived Parameters
Comparison AUCa Sensitivity (%)b Specificity (%)b Cutoff value
PDAC vs. NP
ADC 0.522 (0.414, 0.629) 30 (18/60) 90 (27/30) 1.34
Dslow 0.676 (0.569, 0.771) 48.3 (29/60) 90 (27/30) 1.27
Dfast 0.691 (0.584, 0.784) 70 (42/60) 70 (21/30) 58.7
f 0.919 (0.842, 0.966) 85 (51/60) 93.3 (28/30) 0.16
NET vs. NP
ADC 0.569 (0.413, 0.716) 40 (6/15) 83.3 (25/30) 1.39
Dslow 0.547 (0.391, 0.696) 60 (9/15) 63.3 (19/30) 1.08
Dfast 0.838 (0.698, 0.931) 86.7 (13/15) 76.7 (23/30) 47.6
f 0.640 (0.483, 0.778) 46.7 (7/15) 80 (24/30) 0.19
SPT vs. NP
ADC 0.537 (0.370, 0.698) 44.4 (4/9) 90.0 (27/30) 1.35
Dslow 0.574 (0.406, 0.731) 33.3 (3/9) 96.7 (29/30) 0.90
Dfast 0.896 (0.756, 0.971) 88.9 (8/9) 76.7 (23/30) 48.4
f 0.756 (0.592, 0.879) 55.6 (5/9) 100 (30/30) 0.11
AcP vs. NP
ADC 0.585 (0.424, 0.733) 76.9 (10/13) 50 (15/30) 1.51
Dslow 0.551 (0.392, 0.703) 61.5 (8/13) 60 (18/30) 1.18
Dfast 0.828 (0.682, 0.926) 84.6 (11/13) 76.7 (23/30) 49.8
f 0.815 (0.668, 0.917) 69.2 (9/13) 96.7 (29/30) 0.15
AIP vs. NP
ADC 0.681 (0.508, 0.824) 42.9 (3/7) 100 (30/30) 1.01
Dslow 0.624 (0.450, 0.777) 42.9 (3/7) 100 (30/30) 0.88
Dfast 0.833 (0.675, 0.935) 100 (7/7) 70 (21/30) 58.6
f 0.933 (0.800, 0.989) 85.7 (6/7) 100 (30/30) 0.14
PDAC vs. NET
ADC 0.541 (0.422, 0.657) 50 (30/60) 73.3 (11/15) 1.56
Dslow 0.720 (0.604, 0.818) 80 (48/60) 60 (9/15) 1.08
Dfast 0.624 (0.505, 0.734) 45 (27/60) 86.7 (13/15) 47.6
f 0.830 (0.726, 0.907) 86.7 (52/60) 73.3 (11/15) 0.17
PDAC vs. SPT
ADC 0.519 (0.395, 0.640) 100 (60/60) 22.2 (2/9) 1.04
Dslow 0.678 (0.554, 0.785) 78.3 (47/60) 66.7 (6/9) 1.09
Dfast 0.704 (0.582, 0.808) 71.7 (43/60) 66.7 (6/9) 21.8
f 0.637 (0.512, 0.749) 90 (54/60) 44.4 (4/9) 0.20
PDAC vs. AcP
ADC 0.578 (0.457, 0.693) 55 (33/60) 76.9 (10/13) 1.51
Dslow 0.631 (0.510, 0.741) 53.3 (32/60) 76.9 (10/13) 1.24
Dfast 0.624 (0.503, 0.735) 65 (39/60) 61.5 (8/13) 31.2
January 2017 267

TABLE 3: Continued
Comparison AUCa Sensitivity (%)b Specificity (%)b Cutoff value
f 0.610 (0.489, 0.722) 53.3 (32/60) 69.2 (9/13) 0.10

PDAC vs. AIP
ADC 0.688 (0.563, 0.796) 100 (60/60) 42.9 (3/7) 1.01
Dslow 0.724 (0.601, 0.826) 96.7 (58/60) 42.9 (3/7) 0.88
Dfast 0.543 (0.417, 0.665) 31.7 (19/60) 100 (7/7) 58.6
f 0.581 (0.454, 0.700) 40 (24/60) 85.7 (6/7) 0.09
a
Numbers in parentheses are 95% confidence intervals.
b
Numbers in parentheses are those used to calculate the percentage.
on the parametric maps. However, Fujima et al demonstrated ADC of PDCA compared to respiratory-triggered and free-
that IVIM parameters obtained via an ROI-based approach breathing acquisitions.37 Finally, our results obtained from
correlates better with perfusion parameters obtained with the pancreatic lesions 2 cm may not be generalized into
dynamic contrast-enhanced MRI in head and neck squamous smaller pancreatic lesions. In addition, as all DW images
cell carcinoma than those with the pixel-by-pixel approach.36 were evaluated by a single reader, we could not assess the
In ROI-based fitting, the signal intensities within the ROIs in interobserver variability of IVIM parameters.
each set of DW images are first averaged and then fitted to In conclusion, perfusion-related parameters f and Dfast
calculate IVIM parameters. ROI-based fitting may be advan- are more helpful in characterizing various neoplastic and
tageous in terms of the reliability of fitting result, as outliers inflammatory pancreatic diseases than ADC and Dslow. The
of signal intensity would be averaged before the actual fit- PDCA, AcP, and AIP were characterized by reduced f and
ting.36 However, this approach cannot adequately address a Dfast values compared with normal pancreas. The f value
heterogeneous tumor microenvironment. could help in differentiating between PDAC and NET.
There are several limitations to the present study. First,
since our study was a retrospective study, there may have Acknowledgments
been selection biases. Second, although we included various
Contract grant sponsor: Basic Science Research Program
benign and malignant pancreatic diseases, the sample sizes of
through the National Research Foundation of Korea (NRF)
the pancreatic diseases other than PDAC were not sufficiently
funded by the Ministry of Science, ICT and Future Plan-
large. Third, since IVIM DW images were acquired in a free-
ning; contract grant number: 2014R1A2A1A11052085
breathing manner, the ADC and IVIM parameters may have
been influenced by variances resulting from motion-induced
misalignment. However, as the pancreas is located in the ret-
roperitoneum, the effect of respiratory motion may not be as References
great as in the other abdominal organs.4 Furthermore, a pre- 1. Kartalis N, Lindholm TL, Aspelin P, Permert J, Albiin N. Diffusion-
vious study reported that the measurement reproducibility weighted magnetic resonance imaging of pancreas tumours. Eur
Radiol 2009;19:1981–1990.
and the values of ADC and IVIM parameters in the liver
2. Inan N, Arslan A, Akansel G, Anik Y, Demirci A. Diffusion-weighted
parenchyma were not significantly different between free- imaging in the differential diagnosis of cystic lesions of the pancreas.
breathing and respiratory-triggered IVIM DW imaging.33 AJR Am J Roentgenol 2008;191:1115–1121.
Also, the free-breathing technique is more time-efficient than 3. Fukukura Y, Takumi K, Kamimura K, et al. Pancreatic adenocarcinoma:
variability of diffusion-weighted MR imaging findings. Radiology 2012;
a respiratory-triggered technique. Breath-hold acquisition may 263:732–740.
be an effective method to suppress respiratory motion arti- 4. Barral M, Taouli B, Guiu B, et al. Diffusion-weighted MR imaging of
facts. However, this technique may not be an adequate tech- the pancreas: current status and recommendations. Radiology 2015;
nique for IVIM DW imaging, since a limited breath-hold 274:45–63.
time does not allow for multiple averaging to improve the 5. Kang BK, Kim JH, Byun JH, et al. Diffusion-weighted MRI: usefulness
for differentiating intrapancreatic accessory spleen and small hyper-
signal-to-noise ratio. Furthermore, multiple breath-hold scans vascular neuroendocrine tumor of the pancreas. Acta Radiol 2014;55:
are required to acquire DW images using multiple b-values, 1157–1165.
which may be subject to anatomic misregistration across 6. Lee SS, Byun JH, Park BJ, et al. Quantitative analysis of diffusion-
weighted magnetic resonance imaging of the pancreas: usefulness in
images at different b-values. A previous study demonstrated a characterizing solid pancreatic masses. J Magn Reson Imaging 2008;
poorer reliability of breath-hold DWI in the measurement of 28:928–936.

7. Erturk SM, Ichikawa T, Motosugi U, Sou H, Araki T. Diffusion-weighted son of apparent diffusion coefficient and intravoxel incoherent motion
MR imaging in the evaluation of pancreatic exocrine function before derived parameters. Invest Radiol 2009;44:769–775.
and after secretin stimulation. Am J Gastroenterol 2006;101:133–136.
24. Re TJ, Lemke A, Klauss M, et al. Enhancing pancreatic adenocarci-
8. Matsuki M, Inada Y, Nakai G, et al. Diffusion-weighed MR imaging of noma delineation in diffusion derived intravoxel incoherent motion f-
pancreatic carcinoma. Abdominal imaging 2007;32:481–483. maps through automatic vessel and duct segmentation. Magn Reson
Med 2011;66:1327–1332.
9. Iima M, Le Bihan D. Clinical intravoxel incoherent motion and diffusion
MR imaging: past, present, and future. Radiology 2016;278:13–32. 25. Klauss M, Mayer P, Bergmann F, et al. Correlation of histological ves-
sel characteristics and diffusion-weighted imaging intravoxel incoher-
10. Le Bihan D, Breton E, Lallemand D, Aubin ML, Vignaud J, Laval- ent motion-derived parameters in pancreatic ductal adenocarcinomas
Jeantet M. Separation of diffusion and perfusion in intravoxel incoher- and pancreatic neuroendocrine tumors. Invest Radiol 2015;50:792–
ent motion MR imaging. Radiology 1988;168:497–505. 797.
11. Koh D-M, Collins DJ, Orton MR. Intravoxel incoherent motion in body 26. Lee H-J, Rha SY, Chung YE, et al. Tumor perfusion-related parameter
diffusion-weighted MRI: reality and challenges. Am J Roentgenol of diffusion-weighted magnetic resonance imaging: Correlation with
2011;196:1351–1361. histological microvessel density. Magn Reson Med 2013:1554–1558.
12. Kang KM, Lee JM, Yoon JH, Kiefer B, Han JK, Choi BI. Intravoxel 27. Lee JH, Cheong H, Lee SS, et al. Perfusion assessment using intra-
incoherent motion diffusion-weighted MR imaging for characterization voxel incoherent motion-based analysis of diffusion-weighted mag-
of focal pancreatic lesions. Radiology 2014;270:444–453. netic resonance imaging: validation through phantom experiments.
13. Klauss M, Lemke A, Grunberg K, et al. Intravoxel incoherent motion Invest Radiol 2016 [Epub ahead of print].
MRI for the differentiation between mass forming chronic pancreatitis 28. Lewis RB, Grant E. Lattin J, Maj, Paal E. Pancreatic endocrine tumors:
and pancreatic carcinoma. Invest Radiol 2011;46:57–63. radiologic-clinicopathologic correlation. RadioGraphics 2010;30:1445–
1464.
14. Concia M, Sprinkart AM, Penner AH, et al. Diffusion-weighted mag-
netic resonance imaging of the pancreas: diagnostic benefit from an 29. Brennan DDD, Zamboni GA, Raptopoulos VD, Kruskal JB. Compre-
intravoxel incoherent motion model-based 3 b-value analysis. Invest hensive preoperative assessment of pancreatic adenocarcinoma with
Radiol 2014;49:93–100. 64-section volumetric CT. RadioGraphics 2007;27:1653–1666.
15. Thoeni RF. The Revised Atlanta Classification of Acute Pancreatitis: its 30. Cantisani V, Mortele KJ, Levy A, et al. MR imaging features of solid
importance for the radiologist and its effect on treatment. Radiology pseudopapillary tumor of the pancreas in adult and pediatric patients.
2012;262:751–764. Am J Roentgenol 2003;181:395–401.
16. Shimosegawa T, Chari ST, Frulloni L, et al. International consensus 31. O’Connor OJ, McWilliams S, Maher MM. Imaging of acute pancreati-
diagnostic criteria for autoimmune pancreatitis: guidelines of the Inter- tis. Am J Roentgenol 2011;197:W221–W225.
national Association of Pancreatology. Pancreas 2011;40:352–358.
32. Vlachou PA, Khalili K, Jang H-J, Fischer S, Hirschfield GM, Kim TK.
17. Patel J, Sigmund EE, Rusinek H, Oei M, Babb JS, Taouli B. Diagnosis IgG4-related sclerosing disease: autoimmune pancreatitis
of cirrhosis with intravoxel incoherent motion diffusion MRI and and extrapancreatic manifestations. RadioGraphics 2011;31:1379–1402.
dynamic contrast-enhanced MRI alone and in combination: Prelimi-
33. Lee Y, Lee SS, Kim N, et al. Intravoxel incoherent motion diffusion-
nary experience. J Magn Reson Imaging 2010;31:589–600.
weighted MR imaging of the liver: effect of triggering methods on
18. Guiu B, Petit JM, Capitan V, et al. Intravoxel incoherent motion regional variability and measurement repeatability of quantitative
diffusion-weighted imaging in nonalcoholic fatty liver disease: a 3.0-T parameters. Radiology 2015;274:405–415.
MR study. Radiology 2012;265:96–103. 34. Andreou A, Koh DM, Collins DJ, et al. Measurement reproducibility
19. Sigmund EE, Cho GY, Kim S, et al. Intravoxel incoherent motion of perfusion fraction and pseudodiffusion coefficient derived by intra-
imaging of tumor microenvironment in locally advanced breast can- voxel incoherent motion diffusion-weighted MR imaging in normal
cer. Magn Reson Med 2011;65:1437–1447. liver and metastases. Eur Radiol 2012;23:428–434.
20. Sigmund EE, Vivier PH, Sui D, et al. Intravoxel incoherent motion and 35. Pekar J, Moonen CT, van Zijl PC. On the precision of diffusion/perfu-
diffusion-tensor imaging in renal tissue under hydration and furose- sion imaging by gradient sensitization. Magn Reson Med 1992;23:
mide flow challenges. Radiology 2012;263:758–769. 122–129.
21. Bewick V, Cheek L, Ball J. Statistics review 13: receiver operating 36. Fujima N, Yoshida D, Sakashita T, et al. Intravoxel incoherent motion
characteristic curves. Crit Care 2004;8:508–512. diffusion-weighted imaging in head and neck squamous cell carci-
noma: assessment of perfusion-related parameters compared to
22. Kim SY, Lee SS, Byun JH, et al. Malignant hepatic tumors: short-term dynamic contrast-enhanced MRI. Magn Reson Imaging 2014;32:1206–
reproducibility of apparent diffusion coefficients with breath-hold and 1213.
respiratory-triggered diffusion-weighted MR imaging. Radiology 2010;
37. Kartalis N, Loizou L, Edsborg N, Segersvard R, Albiin N. Optimising
255:815–823.
diffusion-weighted MR imaging for demonstrating pancreatic cancer:
23. Lemke A, Laun FB, Klauss M, et al. Differentiation of pancreas carci- a comparison of respiratory-triggered, free-breathing and breath-hold
noma from healthy pancreatic tissue using multiple b-values: compari- techniques. Eur Radiol 2012;22:2186–2192.
January 2017 269

ORIGINAL RESEARCH
Evaluation of Antiangiogenic and

Antiproliferative Effects of Sorafenib by
Sequential Histology and Intravoxel
Incoherent Motion Diffusion-Weighted
Imaging in an Orthotopic Hepatocellular
Carcinoma Xenograft Model
Shuo-hui Yang, MD,1 Jiang Lin, PhD, MD,1* Fang Lu, MD,2 Zhi-hong Han, MD,3
Cai-xia Fu, MS,4 Peng Lv, PhD, MD,1 Hao Liu, BS,1 and Dong-mei Gao, PhD5
Purpose: To investigate the effectiveness of intravoxel incoherent motion (IVIM) in the assessment of the therapeutic
efficacy of sorafenib in an orthotopic hepatocellular carcinoma (HCC) xenograft model.
Materials and Methods: Thirty-five HCC nude mouse models were established. IVIM was performed on a 1.5T MR scan-
ner at baseline (n 5 5) and serially at 7, 14, and 21 days after sorafenib treatment. The apparent diffusion coefficient
(ADCtotal), true diffusion coefficient (D), pseudodiffusion coefficient (D*), and perfusion fraction (f) at these timepoints
were measured and compared between the treated (n 5 15) and control group (n 5 15). Differences in measurements
among different timepoints were evaluated. Correlations between IVIM parameters and histologic features including
necrotic fraction (NF) and microvessel density (MVD) were analyzed.
Results: Compared to the control group, ADCtotal and D were significantly higher at each timepoint (P 5 0.009), while f
significantly decreased at 7 days (P 5 0.009) and increased at 21 days (P 5 0.028) in the treated group. Serial measure-
ments in the treated group showed that both ADCtotal and D increased significantly at 7, 14, and 21 days compared to
baseline (P < 0.05), while f significantly declined at 7 days (P 5 0.016) and increased at 21 days (P 5 0.009). Significant
correlations were found between ADCtotal and NF (r 5 0.811, P < 0.001), D and NF (r 5 0.838, P < 0.001), and between
f and NF (r 5 0.528, P 5 0.017) in the treated group.
Conclusion: IVIM may provide useful biomarkers for evaluating the therapeutic effects of sorafenib on HCC.
T umor vasculature and tumor cell proliferation are essen-

tial for growth and metastases of hepatocellular carci-
noma (HCC), which suggests that they are targets for
initiation of antivascular treatments has been investigated as
an early response and predictive biomarker prior to reduc-
tion of the tumor size.3,4
anticancer treatment.1 The majority of the pharmaceuticals Unlike commonly used dynamic contrast-enhanced
that target tumor vasculature in HCC inhibit tumor angio- MRI (DCE MRI), intravoxel incoherent motion (IVIM)
genesis and thus lead to tumor cell death.2 Quantitative diffusion-weighted imaging (DWI) with a range of low and
magnetic resonance imaging (MRI) of tumor perfusion after high b-values can provide information separately reflecting
*Address reprint requests to: J.L., Department of Radiology, Zhongshan Hospital, Shanghai Medical College, Fudan University, and Shanghai Institute of
Medical Imaging, 180 Fenglin Road, Shanghai, 200032, China. E-mail: lin.jiang@zs-hospital.sh.cn
From the 1Department of Radiology, Zhongshan Hospital, Shanghai Medical College, Fudan University, and Shanghai Institute of Medical Imaging,
Shanghai, China; 2Department of Radiology, Shuguang Hosipital, Shanghai University of Traditional Chinese Medicine, Shanghai, China; 3Department of
Pathology, Shuguang Hosipital, Shanghai University of Traditional Chinese Medicine, Shanghai, China; 4Siemens Shenzhen Magnetic Resonance Ltd.,
Shenzhen, China; and 5Liver Cancer Institute, Zhongshan Hospital, Fudan University, Key Laboratory of Carcinogenesis and Cancer Invasion, Ministry of
Education, Shanghai, China.

V

Yang et al.: IVIM for Monitoring Sorafenib Effects on HCC
tissue microcapillary perfusion and tissue diffusivity without Orthotopic HCC Xenograft Model
injection of any exogenous contrast medium.5,6 By perform- This study was approved by our hospital’s ethics, animal care and use
ing IVIM-based biexponential analysis, it is possible to committee. Male BALB/C nu/nu mice (Shanghai Institute of Materia
derive the perfusion-sensitive parameters of perfusion frac- Medica, Chinese Academy of Sciences, Shanghai, China) weighing
tion (f ) and pseudodiffusion coefficient (D*), in addition to 23–25 g each were used in this study. The establishment of the ortho-
topic HCC xenograft model was described in detail elsewhere.15
the diffusion-sensitive parameter of true diffusion coefficient
Briefly, HCC-LM3 cells (5 3 106/0.2 ml/site) were inoculated sub-
(D), which can also be displayed as parametric maps.5
cutaneously in the left mediolateral region of axilla in these nude
Although with significant challenges and some uncertainties,
mice. When the tumor grew beyond 1 cm in diameter it was removed
this technology has been applied for tumor assessment and and cut into tumor blocks with a volume of 1 mm3, which were
monitoring the early responses of tumors to treat- implanted into the left lobe of the liver in 35 nude mice. At the 21st
ments.3,4,7–10 In two recently published articles, IVIM was day of tumor growth after its implantation, 5 out of 35 tumor-
used to evaluate therapeutic effects of vascular disrupting bearing nude mice were randomly chosen for a baseline MRI. The
agents in rabbit VX2 liver tumors with promising results.3,9 other 30 mice were randomized into a sorafenib-treated group
Unlike vascular-disrupting agents whose therapeutic effect (n 5 15) and a control group (n 5 15). Mice in the treated group
usually occurs within a few hours and disappears within 24– were further divided into three subgroups according to different treat-
48 hours after initiation of the treatment, the perfusion ment periods on days 7 (n 5 5), 14 (n 5 5), and 21 (n 5 5) after
change induced by more commonly used antiangiogenic treatment. Mice in the control group were also divided into three
drugs takes place days or weeks after treatment and lasts subgroups accordingly. After MRI studies at baseline or at each
different timepoint following treatment, mice in the corresponding
longer, which requires repeated follow-ups for a longer
groups were immediately sacrificed for histological examination and
period of time.
correlation.
Sorafenib is a multikinase inhibitor with antiangio-
genic and antiproliferative effects and shows profound anti- Sorafenib Preparation and Treatment
tumor activity in HCC.11,12 Currently, it represents the Sorafenib (Bayer Schering Healthcare Pharmaceuticals, Leverkusen,
only approved therapy for advanced HCC.12,13 Two pilot Germany) 200 mg per tablet was dissolved in 60 ml 0.9% saline,
clinical studies with a small number of patients have applied 10 ml 99.7% ethyl alcohol, and 10 ml castor oil (China National
IVIM to investigate the association between the effects of Medicines, Beijing, China).
sorafenib on advanced HCC and the parameters of IVIM, For the sorafenib-treated mice, the sorafenib solution was
administered daily by oral gavage at a dose of 30 mg/kg body
with possibly conflicting results.4,10 In one earlier study, per-
weight for 7, 14, and 21 days in the corresponding treated groups.
fusion fraction f was found significantly increased and corre-
This daily dosage was previously reported to produce complete
lated with early response to sorafenib, while apparent tumor growth inhibition in HCC tumor models.16 The control
diffusion coefficient (ADC) and D did not change substan- group mice received 0.2 ml 0.9% saline alone at the same schedule
tially.4 More recently, another study demonstrated that all and route of administration.
IVIM parameters of the patients with advanced HCC did
not change significantly after sorafenib treatment.10 How- MRI
ever, in these two studies the repeated assessment intervals All MRI examinations were performed with a 1.5T MR scanner
(Magnetom Aera; Siemens Healthcare, Erlangen, Germany)
using IVIM were different and an in-depth correlation
equipped with a 16-channel wrist coil. All nude mice were anesthe-
between IVIM and histology at multiple posttreatment
tized by intraperitoneal injection of 3% sodium pentobarbital at a
timepoints was not available in the clinical scenario to dose of 40 mg/kg body weight. The mice were placed in the prone
account for the different results. Hence, the aims of this position in an animal cradle with a belt to limit respiratory motion
study were to investigate the correlation between IVIM- during the experiment.
derived parameters and histologically derived microvessel The conventional MRI sequences and parameters were trans-
density (MVD) and the degree of tumor necrosis at multi- verse T1-weighted (T1W) turbo spin echo (TSE) (TR/TE, 480/13
ple timepoints after sorafenib treatment in an orthotopic msec, field of view [FOV], 80 mm, flip angle [FA], 908), transverse
HCC xenograft model and to further test the effectiveness T2-weighted (T2W) TSE (TR/TE, 4000/74 msec, FOV, 80 mm,
of IVIM in assessment of the therapeutic efficacy of FA, 1508), and coronal T2W TSE (TR/TE, 4000/74 msec, FOV,
sorafenib. 100 mm, FA, 1508). The section thickness and intersection gap
were 2 mm and 1 mm, respectively, for all conventional sequences.
Total scan time for the above sequences was 5 minutes 36 seconds.
Axial IVIM DWI scans were performed using a free-breathing
Materials and Methods single-shot echo-planar sequence with the following imaging parame-
HCC Cell Lines ters: TR/TE, 5000/64 msec; FOV, 295 mm; section thickness, 2 mm;
HCC-LM3, a human HCC cell line, was established and maintained intersection gap, 0.4 mm; voxel size, 1.5 3 1.5 3 2.0 mm3; band-
according to a previously reported study.14 width, 1402 Hz/Px; eight b-values (number of excitations), 0 (1), 25
January 2017 271

signal without diffusion weighting, and S corresponds to the signal

with diffusion weighting.
To generate three parametric images (D, D*, and f ), the
above biexponential model was fit using the following approach:
initial estimation of D using a reduced set of b-values larger than a
predetermined value (200 s/mm2) and then using the resulting D
as a fixed parameter to fit the missing parameters D* and f similar
to what was described in a previous study.18
Regions of interest (ROIs) were separately drawn on all para-
metric images by two experienced radiologists with 10 and 23
years’ experience of MR image interpretation who were blinded to
information regarding the baseline, treatment vs. control, treatment
schedule, and histological examination results. The evaluations
were done on a commercially available workstation (Syngo Multi-
modality Workplace; Siemens Healthcare).
ROIs of the tumor were drawn on ADCtotal images to out-
line the border of the tumor on the slice showing the longest
dimension of the tumor and then these ROIs were copied to the
corresponding IVIM images for D, D*, and f measurements (Figs.
1, 2).9 The drawing was done twice at the same slice and the F1 F2
measured values were averaged by each reader. After a 4-week
interval, these measurements were repeated and the final averaged
values were calculated by each reader.
A set of source IVIM DWI images was transferred to a PC
and the ROI based fitted curves was drawn by using an in-house
developed software based on MatLab (MathWorks, Natick, MA),
where the fitting algorithm was the same with that implemented in
the Siemens IVIM prototype software (Fig. 3). F3
Tumor size defined by the longest dimension on axial T2W
C images was measured and recorded.9 Images from the conventional
O sequences including T1WI and T2WI were also reviewed side-by-
L
O side to exclude hemorrhage or blood products in the tumor. If
R there was hemorrhage on the longest dimension slice, the most
FIGURE 1: IVIM images with ROI, T2WI axial image, and patho- adjacent slice would be analyzed.
logic image of nude mouse xenograft HCC in the treated
group at 14 days after sorafenib treatment. a: ADCtotal map Histologic Examination
(tumor ADCtotal 5 0.779 3 1023 mm2 /s). b: D* map (tumor
After MR scanning at indicated timepoints, the mice were sacri-
D* 5 20.25 3 1023 mm2 /s). c: D map (tumor D 5 0.733 3
1023 mm2 /s). d: f map (tumor f 5 5.26%). e: T2WI axial image ficed by means of cervical dislocation. The tumor was removed
shows the tumor (arrowhead). f: Hematoxylin and eosin stain- and fixed in a 10% buffered formalin solution for at least 24
ing shows the necrotic area (black arrows) in the tumor (magni- hours. After fixation, the tumor was sliced in the central transverse
fication 2003). g: Anti-CD31 immunohistochemistry shows section with a slice thickness of 2 mm to match IVIM images.
intratumoral microvessels (white arrows) (magnification 1003).
Then hematoxylin eosin (HE) staining was performed on 3 lm
(1), 50 (1), 80 (1), 150 (1), 300 (2), 500 (2), and 800 (3) s/mm2; sections for measuring the necrotic fraction (NF) of the tumor.9,19
and scan time, 3 minutes 7 seconds. Generalized autocalibrating The measurement was done by a pathologist with 7 years of expe-
partially parallel acquisition with an acceleration factor of 2 was rience in histologic analysis using an Olympus microscope (Olym-
pus CKX41, Japan) at 2003 magnification with software Image-
employed.
Pro Plus 6.1 at five randomly selected FOVs and the results were
averaged.19 In order to evaluate MVD, an antibody against CD31
Imaging Analysis (Abcam, Cambridge, UK) was used for tumor microvessel immu-
ADCtotal images were automatically calculated inline by the MR nostaining. Tumor MVD was obtained twice and the results were
scanner immediately after acquisition by using all eight b-values averaged on anti-CD 31 sections.20 This was performed by the
with a monoexponential fit of signal intensity.17 All IVIM DWI same pathologist using light microscopy. In brief, areas with the
data were postprocessed by using a Siemens IVIM work-in- densest CD 31-positive vessel in areas of HCC-LM3 tumor were
progress package after acquisition to extract the IVIM parameters, considered to be hot spots at low power (403 and 1003) first.
including D, D*, and f. For IVIM calculation, the biexponential Three hot spots were then chosen and vessels were manually
model was mathematically expressed as follows: S/S0 5 (1-f) counted in a high-power field (4003) and results were averaged
3exp(-bD) 1 f 3exp(-b(D* 1 D)), where S0 corresponds to the and expressed as the highest number of microvessels.

uated by using intraclass correlation coefficients (ICCs) with a

two-way mixed consistency model in the treated group and base-
line (n 5 20). All P values were 2-tailed, with 0.05 as the criterion
for statistical significance. Measurement data were expressed as
(means 6 SD). The statistical analysis was performed using a statis-
tic software (SPSS v. 18.0; IBM, Armonk, NY).
Results
Our liver imaging protocol including IVIM was successfully
completed in all 35 mice. The tumors could be seen clearly
C
O
L
O
R
FIGURE 2: IVIM images with ROI, T2WI axial image, and patho-
logic image of nude mouse xenograft HCC in the control group at
14 days. a: ADCtotal map (tumor ADCtotal 5 0.574 3 1023 mm2 /s).
b: D* map (tumor D* 5 16.09 3 1023 mm2 /s). c: D map (tumor
D 5 0.536 3 1023 mm2 /s). d: f map (tumor f 5 4.53%). e: T2WI
axial image shows the tumor (arrowhead). f: Hematoxylin and
eosin staining shows the necrotic area (black arrows) in the tumor
(magnification 2003). g: Anti-CD31 immunohistochemistry shows
intratumoral microvessels (white arrows) (magnification 1003).
Differences in tumor size, ADCtotal, and all IVIM parameters, NF,
and tumor MVD at each timepoint between sorafenib-treated sub-
groups and the corresponding control subgroups were evaluated
with the Mann–Whitney U-test. Differences in ADCtotal and all
IVIM parameters among baseline and three subgroups of both
sorafenib-treated and control mice were estimated by the Kruskal–
Wallis test and Mann–Whitney U-test. The Spearman rank correla-
tion test was performed to evaluate the correlation between ADCto-
tal and all IVIM parameters and corresponding histologic features
C
including NF and MVD of all the tumors from baseline to 21 O
days follow-up. The correlation coefficient rho (r) was obtained to L
define the degree of correlation as follows: poor or no relationship O
R
if 0 jrj < 0.2, fair if 0.2 jrj 0.4, moderate if 0.4 < FIGURE 3: DWI images (b 5 0, 25, 50, 80, 150, 300, 500, and
jrj0.6, good if 0.6 < jrj 0.8, and excellent if jrj > 0.8.21 The 800 s/mm2) and ROI-based fitted curves of a nude mouse xeno-
intra- and interobserver agreement in ROI measurements was eval- graft HCC in the control (a,b) and treated (c,d) group at 7 days.
January 2017 273

on ADCtotal and all IVIM maps. Intratumoral hemorrhage Effect of Sorafenib Treatment on
foci were found in two mice of the control group (one at Tumor Size, NF, and MVD
day 14 and one at day 21) and in four mice of the treated The tumor size at baseline was 1.062 6 0.070 cm. Signifi-
group (two at day 7, one at day 14, and one at day 21). cantly smaller tumor size was found in the treated group
However, none of them was observed on the longest dimen- than the control group at each timepoint (7 days, 1.091 6
sion slice of the tumor used for IVIM analysis. 0.086 cm vs. 1.591 6 0.140 cm, P 5 0.009; 14 days,
1.183 6 0.107 cm vs. 1.877 6 0.168 cm, P 5 0.009; 21
Comparison of ADCtotal and IVIM Parameters days, 1.320 6 0.108 cm vs. 2.347 6 0.203 cm, P 5 0.009).
Between the Control and Treated Groups The NF at baseline was 11.050 6 1.754%. Signifi-
The mean values of ADCtotal and IVIM parameters for both cantly greater NF was observed in the treated group than
treated and control groups at each timepoint are summar- the control group at each timepoint (7 days, 36.892 6
T1 ized in Table 1 and Fig. 4. Both ADCtotal (7 days, P 5 4.917% vs. 12.202 6 2.181%, P 5 0.009; 14 days,
F4 0.009; 14 days, P 5 0.009; 21 days, P 5 0.009) and D (7 44.506 6 5.067% vs. 34.888 6 2. 302%, P 5 0.009; 21
days, P 5 0.009; 14 days, P 5 0.009; 21 days, P 5 0.009) days, 63.474 6 11.188% vs. 46.480 6 1.911%, P 5 0.016)
values were significantly higher in the treated group than in (Figs. 1, 2).
the control group at each timepoint. Compared to the con- The MVD at baseline was 12.700 6 1.121. Signifi-
trol group, f values significantly decreased in the treated cantly lower MVD was found in the treated group than
group at 7 days follow-up (P 5 0.009) and then slightly and the control group at each timepoint (7 days, 8.067 6 0.630
significantly increased at 14 days (P 5 0.251) and 21 days vs. 12.633 6 1.330, P 5 0.009; 14 days, 10.700 6 0.721 vs.
follow-up (P 5 0.028). No significant difference was found 13.200 6 1.440, P 5 0.012; 21 days, 11.167 6 1.219 vs.
in D* at any follow-up timepoints between the treated and 13.367 6 1.376, P 5 0.036) (Figs. 1, 2).
control groups (7 days, P 5 0.465; 14 days, P 5 0.465; 21
days, P 5 0.251). Correlation of Histologic Features (NF and MVD)
With ADCtotal and IVIM Parameters
Sequential Measurements of ADCtotal and IVIM In the treated group, ADCtotal, D, and f showed a signifi-
Parameters in the Treated and Control Groups cant positive correlation with NF (Fig. 5). Both ADCtotal F5
as Compared With Baseline and D demonstrated an excellent relationship with NF
Results of the Mann–Whitney U-test for sequential meas- (ADCtotal, r 5 0.811, P < 0.001; D, r 5 0.838, P < 0.001).
urements of ADCtotal and IVIM parameters of both the A moderate relationship was observed between f and NF
treated and control groups are summarized in Table 1 and (r 5 0.528, P 5 0.017). However, no correlation was
Fig. 4. Sequential measurements of ADCtotal and D in the observed between D* and NF (r 5 –0.095, P 5 0.691).
treated group showed that both ADCtotal and D increased There was no correlation between ADCtotal, IVIM parame-
significantly at 7, 14, and 21 days compared with the values ters and histological MVD either (ADCtotal, r 5 –0.115,
at baseline (ADCtotal, P 5 0.047, 0.009, and 0.009; D, P 5 0.629; D, r 5 –0.156, P 5 0.512; D*, r 5 –0.115,
P 5 0.028, 0.009, and 0.009, respectively). A significant P 5 0.629 and f, r 5 0.437, P 5 0.054).
increase in these two parameters was found at 14 days and In the control group, only f showed significant positive
21 days vs. 7 days (ADCtotal, P 5 0.028 and 0.009; D, correlation with MVD, with moderate relationship
P 5 0.047 and 0.009, respectively) and at 21 days vs. 14 (r 5 0.568, P 5 0.009) (Fig. 5). No correlations were
days (ADCtotal, P 5 0.009; D, P 5 0.009). f was found sig- observed between ADCtotal, D, D*, and MVD (ADCtotal,
nificantly declined at 7 days (P 5 0.016), recovered, and r 5 0.145, P 5 0.543; D, r 5 0.130, P 5 0.586 and D*,
slightly increased at 14 days (P 5 0.175) and markedly r 5 0.281, P 5 0.230). There was no correlation between
increased at 21 days (P 5 0.009) compared with the values ADCtotal, IVIM parameters, and NF (ADCtotal, r 5 –0.176,
at baseline. f was significantly higher at both 14 and 21 P 5 0.458; D, r 5 –0.177, P 5 0.454; D*, r 5 –0.018,
days (P 5 0.016 and 0.009, respectively) than that at 7 P 5 0.940 and f, r 5 0.329, P 5 0.156) (Fig. 5).
days. But there were no significant changes found in D* in
Intra- and Interobserver Agreement on
the treated group (v2 5 0.669, P 5 0.881). In the control ADCtotal and All IVIM Parameters
group, both ADCtotal and D decreased slightly at 7 days High intra- and interobserver ICCs for measuring ADCtotal,
and recovered a little at 14 and 21 days; however, not sig- D, D*, and f in the treated group and baseline were
nificantly among all different timepoints (ADCtotal, v2 5 observed (Table 2). T2
6.680, P 5 0.083; D, v2 5 7.229, P 5 0.065). Neither D*
(v2 5 0.931, P 5 0.818) nor f (v2 5 1.709, P 5 0.635) Discussion
showed any significant changes among all different time- Our study showed that IVIM could be used with excellent
points in the control group. intra- and interobserver agreement to detect sequential changes

TABLE 1. Sequential Measurements of ADCtotal and IVIM Parameters From Baseline (n 5 5), the Treated Group (n 5 15), and Control Group (n 5 15) and P Val-
January 2017
ues for Comparisons Between Different Timepoints
Treated group Control group
P valuesa P valuesa
Parameter Measurement 7 days 14 days 21 days Measurement 7 days 14 days 21 days
ADCtotal ( 3 1023 mm2 /s)

baseline 0.609-0.685 (0.654 6 0.030) 0.047 0.009 0.009 0.609-0.685 (0.654 6 0.030) NS NS NS
7 days 0.649-0.757 (0.708 6 0.041) 0.028 0.009 0.504-0.643 (0.582 6 0.057) NS NS
14 days 0.732-0.822 (0.773 6 0.037) 0.009 0.541-0.637 (0.592 6 0.040) NS
21 days 0.965-1.201 (1.110 6 0.099) 0.556-0.655 (0.600 6 0.040)
23 2
D ( 3 10 mm /s)
baseline 0.562-0.632 (0.609 6 0.028) 0.028 0.009 0.009 0.562-0.632 (0.609 6 0.028) NS NS NS
7 days 0.622-0.731 (0.679 6 0.043) 0.047 0.009 0.447-0.605 (0.543 6 0.066) NS NS
14 days 0.700-0.785 (0.739 6 0.032) 0.009 0.515-0.606 (0.554 6 0.039) NS
21 days 0.811-1.127 (1.017 6 0.146) 0.523-0.610 (0.561 6 0.035)
23 2
D* ( 3 10 mm /s)
baseline 13.815-30.450 (17.939 6 7.054) NS NS NS 13.815-30.450 (17.939 6 7.054) NS NS NS
7 days 11.175-25.040 (18.224 6 6.357) NS NS 11.150-17.520 (15.154 6 2.694) NS NS
14 days 10.680-27.525 (19.140 6 6.298) NS 11.070-20.640 (16.249 6 3.657) NS
21 days 9.100-27.105 (17.230 6 6.535) 10.765-17.190 (14.657 6 2.624)
f (%)
baseline 4.010-4.660 (4.390 6 0.260) 0.016 0.175 0.009 4.010-4.660 (4.390 6 0.260) NS NS NS
7 days 3.115-4.035 (3.532 6 0.361) 0.016 0.009 4.070-4.865 (4.411 6 0.347) NS NS

14 days 4.020-5.655 (4.917 6 0.671) 0.530 4.295-4.810 (4.547 6 0.216) NS
21 days 4.725-5.605 (5.249 6 0.370) 4.240-4.870 (4.579 6 0.264)
Stage:
a
Data were tested with Kruskal-Wallis test followed by Mann-Whitney U-test in case of statistical significance. “NS” indicates nonsignificant according to Kruskal-Wallis test; ADCtotal,
apparent diffusion coefficient; IVIM, intravoxel incoherent motion.

275
Page: 275
FIGURE 4: Boxplots show sequential measurements of ADCtotal (a), D (b), D* (c), and f (d) in both treated and control groups.
Comparisons between control and treated groups were performed by using the Mann–Whitney U-test (䊏 indicates P < 0.01;
~ indicates 0.01 < P < 0.05). Sequential data of both treated and control groups were tested with the Kruskal–Wallis test
followed by Mann–Whitney U-test in case of statistical significance (**P < 0.01; *0.01 < P < 0.05). Center line represents median;
upper and lower margins of box are upper and lower quartiles; 䊊 indicates outlier.
in diffusion and perfusion in an HCC animal model under sor- platelet-derived growth factor receptor, and serine/threonine
afenib treatment. Parameters obtained by IVIM might be able kinases c-RAF and BRAF.16 Its antiangiogenic and antiproli-
to simultaneously reflect tumor necrosis and tumor vascular ferative effects on HCC was confirmed in our xenograft
inhibition resulting from the effects of sorafenib. models by smaller tumor size, greater NF, and lower MVD
Although the treatment with sorafenib has proved in the treated than the control mice.
effective in different studies, only a portion of patients ADCtotal, calculated by using all b-values, was a reflec-
actually benefit from it.12,22 And its use frequently causes tion of both diffusion and perfusion effects, while D was
side effects and is costly.12 An imaging technology being the slow component of diffusion reflecting both intra- and
able to determine the therapeutic efficacy of sorafenib at an extracellular molecular diffusion.24 Both ADC and D may
early stage is desirable because it helps in selecting respon- increase with persistent cell death or tumor necrosis follow-
sive patients to sorafenib shortly after the start of treat- ing treatment.25 In line with this assumption, ADCtotal and
ment.23 It may also enable a better understanding of the D in this study showed a significant increase after sorafenib
changes of HCC neoangiogenesis and proliferation during treatment and an excellent correlation with pathologic NF.
sorafenib treatment, which are incompletely understood. This result confirmed that these two IVIM parameters could
Sorafenib acts by suppressing the receptor tyrosine be used to reflect the degree of necrosis after sorafenib treat-
kinases of vascular endothelial growth factor receptor, ment. Moreover, after removal of perfusion contamination,

FIGURE 5: Correlation between ADCtotal, IVIM parameters, and NF, MVD in the treated and control groups. In the treated group,
significantly positive correlations between ADCtotal (a), D (b), f (c) and histological NF are found (ADCtotal, r 5 0.811, P < 0.001; D,
r 5 0.838, P < 0.001; f, r 5 0.528, P 5 0.017). (d) In the control group, f shows significantly positive correlation with MVD
(r 5 0.568, P 5 0.009). No correlations were observed between ADCtotal (e), D (f) and MVD in the treated (ADCtotal, r 5 –0.115,
P 5 0.629 and D, r 5 –0.156, P 5 0.512) and in the control group (ADCtotal, r 5 0.145, P 5 0.543 and D, r 5 0.130, P 5 0.586).
D would be more beneficial than ADCtotal for revealing the more strongly with D than with ADCtotal. However, in a
decreased cellular density and increased necrotic area.17 This previous clinical study in advanced HCC patients under
could be used to explain why the histological NF correlated sorafenib treatment, no substantial changes in ADCtotal and
January 2017 277

TABLE 2. Intra- and Interobserver Agreement on ADCtotal and all IVIM Parameters in Baseline and the Treated
Group (n 5 20)
Reader 1 Reader 2 Interobserver

Parameter ICCs P ICCs P ICCs P
ADCtotal 0.962 0.000 0.975 0.000 0.978 0.000

D 0.966 0.000 0.973 0.000 0.973 0.000
D* 0.899 0.000 0.912 0.000 0.907 0.000
f 0.968 0.000? 0.972 0.000 0.963 0.000
ICCs, intraclass correlation coefficients; ADCtotal, apparent diffusion coefficient; IVIM, intravoxel incoherent motion.
D values were found. The authors attributed the lack of the diffusion gradients might be used to strengthen the robust-
changes in theses parameters to the lack of marked necrotic ness of D*.32,33
effects at their follow-up intervals.4 Contrary to the treated group, moderate correlation
In the treated group, the parameter f, which character- was found between f and MVD in the control group. This
izes the vascular volume fraction in the tumor, showed a sig- correlation has been reported in pancreatic tumors in
nificant decrease followed by an increase, with no patients,34 in colorectal tumors in nude mice,35 and in VX2
correlation found with MVD. However, a moderate positive liver tumors in rabbits.36 Hence, we believe f could serve as
correlation between f and NF was observed. This trend of a noninvasive marker of tumor vascularity in its natural
change in f might be explained by the antiangiogenic effect course of development without drug intervention. However,
of sorafenib, in which first destruction and then normaliza- no correlation was found between ADCtotal, D, and NF in
tion of the abnormal structure and function of tumor vascu- the control group as opposed to the treated group. We have
lature takes place. The tumor vessels are transiently rendered a possible explanation for this result. Theoretically, the
close to normal with improved basement membrane, greater increase of cellularity due to proliferation of tumor cells
pericyte coverage, decreased vascular permeability and inter- caused obliteration of the extracellular spaces, resulting in a
stitial fluid pressure,26 which led to the increase in f at a decrease of ADCtotal and D; while tumor cell death and
later time. This similar normalization effect has been necrosis could account for an increase in ADCtotal and D.37
reported and histologically confirmed by Zwick et al using In other words, prior to the antiangiogenic therapy, the bal-
DCE MRI and vessel size imaging to monitor antiangio- ance between tumor cell proliferation and necrosis instead
genic therapy in an animal tumor model.27 In another clini- of the necrosis (ie, NF) alone determine the values of ADC-
cal study, Lewin et al also described the cause of the total and D. When this balance was broken after antiangio-
increased f as an increased perfusion rate by normalization genic therapy or chemotherapy, both ADCtotal and D would
of tumor blood vessels.4 The substantial increase in f might correlate with the degree of tumor necrosis, as shown in this
also be attributed to the necrosis and markedly decreased study and in a previous study on liver metastasis from colo-
cell density induced by sorafenib with reduced pressure that rectal cancer.25
tumor cells exert against vessels in the tumor, with resultant Our study has several limitations. First, the histopatho-
higher perfusion.28 This postulation was validated by a sig- logic section of the tumor with 3-lm thickness may not
nificant positive correlation demonstrated between f and NF completely match that on IVIM with a thickness of 2 mm.
in the treated group. MVD and NF obtained at this section may not be represen-
This study did not find significant changes in D* tative of the whole tumor vessels and necrosis because tumor
value, which represents the tumor microcirculation in the growth is very heterogeneous. Second, the animal sample
capillary network, at any time intervals in the treated size was relatively small, although five mice per subgroup
groups. Similar results were reported recently in a clinical both in the control and treated groups were used. Further
study.10 The underlying causes remained unknown and war- investigation with a larger number of animals is warranted
rant further investigation. The inherent poor signal-to-noise to confirm the interesting findings in this study and their
ratio of D*,17,29–31 and subsequently decreased precision in potential benefits in the future. Third, although utilizing 10
its measurements with larger standard deviation as shown in or more b-values is not unusual, which would provide more
this study, might be the critical issues interfering with its data support for evaluation of various parameters, it trans-
accurate reflection of tumor microcirculation. Hence, lates into very long measurement times and is unlikely to be
further technical improvements such as flow-compensated clinically feasible.7 Hence, we used eight b-values, with five

lower than 200 s/mm2 to enable extraction of perfusion- 9. Joo I, Lee JM, Han JK, Choi BI. Intravoxel incoherent motion
diffusion-weighted MR imaging for monitoring the therapeutic efficacy
sensitive information with high reproducibility and in a rea- of the vascular disrupting agent CKD-516 in rabbit VX2 liver tumors.
sonably short acquisition time. Fourth, tumor hemorrhage Radiology 2014;272:417–426.
was excluded in this study since bleeding could potentially 10. Shirota N, Saito K, Sugimoto K, Takara K, Moriyasu F, Tokuuye K.
Intravoxel incoherent motion MRI as a biomarker of sorafenib treat-
affect the accuracy of IVIM analysis. Fifth, because many
ment for advanced hepatocellular carcinoma: a pilot study. Cancer
different factors acting on the intra- and extracellular spaces Imaging 2016;16:1.
can induce modifications of water diffusion, IVIM cannot 11. Llovet JM, Ricci S, Mazzaferro V, et al. Sorafenib in advanced hepato-
differentiate between intra- and extracellular water molecule cellular carcinoma. N Engl J Med 2008;359:378–390.
diffusion.38 Lastly, although a similar MRI platform and 12. Colagrande S, Regini F, Taliani GG, Nardi C, Inghilesi AL. Advanced
hepatocellular carcinoma and sorafenib: Diagnosis, indications, clinical
human HCC xenograft models were used in this study, the and radiological follow-up. World J Hepatol 2015;7:1041–1053.
degree to which results from animal studies can be trans-
13. Llovet JM, Ducreux M, Lencioni R, et al. EASL-EORTC clinical practice
lated into clinical practice requires further validation. guidelines: management of hepatocellular carcinoma. J Hepatol
In conclusion, our results suggest that ADCtotal, D, 2012;56:908–943.
and f obtained by IVIM may be useful as noninvasive bio- 14. Tian J, Tang ZY, Ye SL, et al. New human hepatocellular carcinoma
(HCC) cell line with highly metastatic potential (MHCC97) and its
markers for prediction of HCC responses to sorafenib. Our expressions of the factors associated with metastasis. Br J Cancer
HCC tumor model and IVIM may potentially be used as a 1999;81:814–821.
preclinical platform in validating the effects of newly devel- 15. Sun FX, Tang ZY, Lui KD, et al. Establishment of a metastatic model
oped antiangiogenic treatments. of human hepatocellular carcinoma in nude mice via orthotopic
implantation of histologically intact tissues. Int J Cancer 1996;66:239–
243.
Acknowledgment 16. Liu L, Cao Y, Chen C, et al. Sorafenib blocks the RAF/MEK/ERK path-
way, inhibits tumor angiogenesis, and induces tumor cell apoptosis in
Contract grant sponsor: National Natural Science Founda- hepatocellular carcinoma model PLC/PRF/5. Cancer Res 2006;66:
tion of China; contract grant number: 81371542; Contract 11851–11858.
grant sponsor: Shanghai Health Committee; contract grant 17. Yoon JH, Lee JM, Yu MH, Kiefer B, Han JK, Choi BI. Evaluation of
number: XBR2013115. hepatic focal lesions using diffusion-weighted MR imaging: compari-
son of apparent diffusion coefficient and intravoxel incoherent
Dr. Hua Lv from Shuguang Hosipital, Shanghai University motion-derived parameters. J Magn Reson Imaging 2014;39:276–285.
of Traditional Chinese Medicine, provided statistical advice 18. Luciani A, Vignaud A, Cavet M, et al. Liver cirrhosis: intravoxel inco-
for this study. herent motion MR imaging—pilot study. Radiology 2008;249:891–
899.
19. Geng W, Ng KT, Sun CK, et al. The role of proline rich tyrosine kinase
2 (Pyk2) on cisplatin resistance in hepatocellular carcinoma. PLoS One
2011;6:e27362.
References
20. Weidner N, Folkman J, Pozza F, et al. Tumor angiogenesis: a new sig-
1. Pircher A, Medinger M, Drevs J. Liver cancer: Targeted future options. nificant and independent prognostic indicator in early-stage breast
World J Hepatol 2011;3:38–44. carcinoma. J Natl Cancer Inst 1992;84:1875–1887.
2. Sampat KR, O’Neil B. Antiangiogenic therapies for advanced hepato- 21. Fujima N, Yoshida D, Sakashita T, et al. Intravoxel incoherent motion
cellular carcinoma. Oncologist 2013;18:430–438. diffusion-weighted imaging in head and neck squamous cell carcinoma:
3. Joo I, Lee JM, Grimm R, Han JK, Choi BI. Monitoring vascular disrupt- assessment of perfusion-related parameters compared to dynamic
ing therapy in a rabbit liver tumor model: relationship between tumor contrast-enhanced MRI. Magn Reson Imaging 2014;32:1206–1213.
perfusion parameters at IVIM diffusion-weighted MR imaging and 22. Abou-Alfa GK, Amadori D, Santoro A, et al. Safety and efficacy of sor-
those at dynamic contrast-enhanced MR imaging. Radiology 2016; afenib in patients with hepatocellular carcinoma (HCC) and Child-
278:104–113. Pugh A versus B cirrhosis. Gastrointest Cancer Res 2011;4:40–44.
4. Lewin M, Fartoux L, Vignaud A, Arrive L, Menu Y, Rosmorduc O. The 23. Schraml C, Kaufmann S, Rempp H, et al. Imaging of HCC—current
diffusion-weighted imaging perfusion fraction f is a potential marker state of the art. Diagnostics (Basel) 2015;5:513–545.
of sorafenib treatment in advanced hepatocellular carcinoma: a pilot
study. Eur Radiol 2011;21:281–290. 24. Guiu B, Petit JM, Capitan V, et al. Intravoxel incoherent motion
diffusion-weighted imaging in nonalcoholic fatty liver disease: a 3.0-T
5. Le Bihan D, Breton E, Lallemand D, Aubin ML, Vignaud J, Laval- MR study. Radiology 2012;265:96–103.
Jeantet M. Separation of diffusion and perfusion in intravoxel incoher-
ent motion MR imaging. Radiology 1988;168:497–505. 25. Chiaradia M, Baranes L, Van Nhieu JT, et al. Intravoxel incoherent
motion (IVIM) MR imaging of colorectal liver metastases: are we only
6. Le Bihan D, Breton E, Lallemand D, Grenier P, Cabanis E, Laval-
looking at tumor necrosis? J Magn Reson Imaging 2014;39:317–325.
Jeantet M. MR imaging of intravoxel incoherent motions: application
to diffusion and perfusion in neurologic disorders. Radiology 1986; 26. Jain RK. Normalization of tumor vasculature: an emerging concept in
161:401–407. antiangiogenic therapy. Science 2005;307:58–62.
7. Koh DM, Collins DJ, Orton MR. Intravoxel incoherent motion in body 27. Zwick S, Strecker R, Kiselev V, et al. Assessment of vascular remodel-
diffusion-weighted MRI: reality and challenges. AJR Am J Roentgenol ing under antiangiogenic therapy using DCE-MRI and vessel size
2011;196:1351–1361. imaging. J Magn Reson Imaging 2009;29:1125–1133.
8. Granata V, Fusco R, Catalano O, et al. Early assessment of colorectal 28. Ferretti S, Allegrini PR, O’Reilly T, et al. Patupilone induced vascular
cancer patients with liver metastases treated with antiangiogenic disruption in orthotopic rodent tumor models detected by magnetic
drugs: the role of intravoxel incoherent motion in diffusion-weighted resonance imaging and interstitial fluid pressure. Clin Cancer Res
imaging. PLoS One 2015;10:e0142876. 2005;11:7773–7784.
January 2017 279

29. Kakite S, Dyvorne H, Besa C, et al. Hepatocellular carcinoma: short- 34. Klau M, Mayer P, Bergmann F, et al. Correlation of histological vessel
term reproducibility of apparent diffusion coefficient and intravoxel characteristics and diffusion-weighted imaging intravoxel incoherent
incoherent motion parameters at 3.0T. J Magn Reson Imaging 2015; motion-derived parameters in pancreatic ductal adenocarcinomas and
41:149–156. pancreatic neuroendocrine tumors. Invest Radiol 2015;50:792–797.
30. Guo Z, Zhang Q, Li X, Jing Z. Intravoxel incoherent motion diffusion 35. Lee HJ, Rha SY, Chung YE, et al. Tumor perfusion-related parameter
weighted MR imaging for monitoring the instantly therapeutic efficacy of diffusion-weighted magnetic resonance imaging: correlation with
of radiofrequency ablation in rabbit VX2 tumors without evident links histological microvessel density. Magn Reson Med 2014;71:1554–
between conventional perfusion weighted images. PLoS One 2015; 1558.
10:e0127964.
36. Wu H, Liu H, Liang C, et al. Diffusion-weighted multiparametric MRI
31. Dyvorne H, Jajamovich G, Kakite S, Kuehn B, Taouli B. Intravoxel for monitoring longitudinal changes of parameters in rabbit VX2 liver
incoherent motion diffusion imaging of the liver: optimal b-value sub- tumors. J Magn Reson Imaging 2016 [Epub ahead of print] doi:
sampling and impact on parameter precision and reproducibility. Eur 10.1002/jmri.25179.
J Radiol 2014;83:2109–2113.
37. Wu H, Hu M, Liu Z, et al. Intravoxel incoherent motion diffusion-
32. Wetscherek A, Stieltjes B, Laun FB. Flow-compensated intravoxel weighted MRI for characterizing regional variability and monitoring
incoherent motion diffusion imaging. Magn Reson Med 2015;74:410– serial changes of parameters in rabbit VX2 liver tumors. J Magn
419. Reson Imaging 2016;43:173–180.
33. Fujita N, Harada K, Sakurai K, Akai Y, Kozuka T. Separation of diffu- 38. Ackerman JJ, Neil JJ. The use of MR-detectable reporter molecules
sion and slow flow effects by use of flow rephasing and dephasing. and ions to evaluate diffusion in normal and ischemic brain. NMR
Magn Reson Med 1992;24:109–122. Biomed 2010;23:725–733.

ORIGINAL RESEARCH
Gadoxetic Acid-Enhanced MRI for the

Characterization of Hepatocellular
Carcinoma: A Systematic Review and
Meta-Analysis
Joanna K. Duncan, PhD,1* Ning Ma, BHlthSc, BMaCompSc,1
Thomas D. Vreugdenburg, PhD,1 Alun L. Cameron, PhD,1 and
Guy Maddern, MD, PhD1,2
Purpose: To establish the relative diagnostic accuracy of gadoxetic acid-enhanced magnetic resonance imaging (GA-MRI)
compared with contrast-enhanced computed tomography (CE-CT), dynamic MRI (D-MRI), gadopentetic acid-enhanced
MRI (GP-MRI), or gadobenic acid-enhanced MRI (GB-MRI) in the characterization of hepatocellular carcinoma (HCC).
Materials and Methods: PubMed, EMBASE, the Cochrane Library, and the University of York CRD databases were
searched to February 29 2016 for any studies that compared the diagnostic accuracy of GA-MRI to CE-CT, D-MRI, GP-
MRI, or GB-MRI in patients with known or suspected HCC. Diagnostic accuracy outcomes (true positive, true negative,
false positive, false negative) were extracted and analyzed using the bivariate model of Reitsma et al (2005).
Results: In studies comparing GA-MRI to CE-CT in patients with any-sized lesions, estimated sensitivities were 0.881 (95%
confidence interval [CI] 5 0.766, 0.944) and 0.713 (95% CI 5 0.577, 0.819) respectively. Estimated specificities were 0.926
(95% CI 5 0.829, 0.97) and 0.918 (95% CI 5 0.829, 0.963), respectively. This difference was not statistically significant.
In studies including patients with small lesions GA-MRI was superior to CE-CT, with estimated sensitivities of 0.919
(95% CI 5 0.834, 0.962) and 0.637 (95% CI 5 0.565, 0.704 and estimated specificities of 0.936 (95% CI 5 0.882, 0.966) and
0.971 (95% CI 5 0.937, 0.987), respectively. In studies comparing GA-MRI to D-MRI in patients with any-sized lesions
estimated sensitivities were 0.907 (95% CI 5 0.870, 0.934) and 0.820 (95% CI 5 0.776, 0.857); estimated specificities were
0.929 (95% CI 5 0.877, 0.961) and 0.934 (95% CI 5 0.881, 0.964).
Conclusion: GA-MRI has superior diagnostic ability to CE-CT in patients with small lesions. In patients with any-sized
lesions there is no evidence that GA-MRI is superior to either CE-CT to D-MRI.
H epatocellular carcinoma (HCC) is the sixth most com-

mon cancer worldwide and the third most common
cause of cancer-related death.1 The preferred treatment for
stage disease is treated with palliative intent and survival is
markedly lower. The median survival rate for intermediate
stage HCC (multinodular lesions) following transcatheter
primary malignant liver lesions is dependent on tumor stage arterial chemoembolization (TACE) is 20 months (range
and patient symptoms. Very early stage (single tumor 14–45 months) and the median 5-year survival following
<2 cm) and early stage (3 nodules 3 cm each) HCC can advanced stage HCC (portal invasion) treated with chemo-
be treated with curative intent using transplant, resection, therapy is 11 months (range 6–14).2
and/or ablation, providing there will be sufficient liver func- Less than 30% of HCC patients are eligible for cura-
tion following the procedure. The 5-year survival rate in tive procedures at the time of diagnosis5,6 and hepatic recur-
this population is between 70 and 80%.2–4 In contrast, later rence occurs in more than 50% of patients treated with
*Address reprint requests to: J.D., 199 Ward Street, North Adelaide SA 5006, Australia. E-mail: college.asernip@surgeons.org
From the 1Australian Safety and Efficacy Register of New Interventional Procedures, Surgical (ASERNIP-S), Royal Australasian College of Surgeons,
Adelaide, South Australia, Australia; and 2Discipline of Surgery, University of Adelaide and the Queen Elizabeth Hospital, Adelaide, South Australia,
Australia.

V 281

curative measures.7–10 More accurate detection of HCC at an (surgical specimen included explanted liver, percutaneous biopsy,
earlier stage of disease may reduce the risk of tumor recurrence or clinical and radiological follow-up) to confirm the diagnosis.
and this is the focus of much diagnostic research.11,12 Comparator imaging techniques included: CE-CT, D-MRI, GP-
Current imaging modalities for characterizing liver MRI, and GB-MRI. Studies were excluded if they included
lesions include contrast-enhanced computed tomography patients with known or suspected pathology other than HCC, if
imaging was used to assess treatment efficacy rather than for diag-
(CE-CT) and magnetic resonance imaging (MRI).13 Dynamic
nosis, if diagnostic accuracy data (true positives, false positives, true
MRI (D-MRI) imaging should include arterial phase, portal
negatives, and false negatives) was not reported and could not be
venous phase, and delayed phase imaging, which can be con- calculated, or if they were studies of diagnostic yield with no refer-
ducted using extracellular contrast agents, for example, gado- ence standard. Studies that did not report true-negative data were
pentetic acid (Gd-DTPA, GP). Additional hepatobiliary excluded post-hoc as these studies could not be included in the
phase images may be obtained if an intracellular agent such as bivariate analyses of diagnostic accuracy. Searches were date-limited
gadoxetic acid (Gd-EOB-DTPA, GA) or gadobenic acid (Gd- to only retrieve articles published since 1991, as this was the year
BOPTA, GB) is used.2 gadoxetic acid was first mentioned in the literature.
There have been a number of systematic reviews Data were extracted by one evaluator and checked by a sec-
concerned with the diagnosis of HCC by GA-MRI14–19; ond using standardized data extraction tables developed a priori.
however, these reviews either calculated the diagnostic accu- Any disagreements or discrepancies in the extracted data were
racy of GA-MRI without including a head-to-head compari- resolved through discussion. Diagnostic accuracy data (true posi-
tives, false positives, true negatives, and false negatives) were
son to existing imaging modalities,14,16–18 or did not separate
extracted or calculated for all studies, where available.
results for GA-MRI from other MRI techniques.15 Wu et al
Diagnostic accuracy studies were appraised using the
did compare GA-MRI to CE-CT in studies published before
QUADAS-II tool.22 Included studies on diagnostic accuracy were
April 2012 and therefore does not include recently published examined in four domains: patient selection, intervention, compar-
literature.19 The aim of this systematic review was to establish ator, reference standard, and study design. Each domain was
the relative diagnostic accuracy of GA-MRI compared to judged to have a low, high, or unclear risk of bias and a low, high,
CE-CT, D-MRI, GP-MRI, and MRI with hepatobiliary phase or unclear concern of applicability. Critical appraisal was conducted
images using Gd-BOPTA (GB-MRI). by one reviewer and checked by a second.
Data from studies where patients were included with any-sized
Materials and Methods tumors was analyzed separately to data extracted from studies where
The protocol for this review, including additional research ques- only patients with tumors less than 3 cm in diameter were enrolled.
tions not presented here, underwent external peer-review and con- Data from studies that included patients with lesions of any size and
sultation. It is accessible from the website of the Australian presented a subgroup analysis of results for tumors less than 3 cm in
Government Department of Health.20 This systematic review dif- diameter were not included in the analysis of small lesions, as it could
fers from the protocol with respect to the comparator tests which not be confirmed whether detection of small lesions was influenced
have been updated to be CE-CT, D-MRI, GP-MRI, and GB-MRI by the presence of a larger lesion in the patient.
in line with the American College of Gastroenterology guidelines Both bivariate models and hierarchical summary receiver operat-
for diagnosing HCC.13 This systematic review adheres to the ing characteristic (HSROC) analyses were conducted to compare GA-
reporting guidelines of the PRISMA statement.21 MRI to any comparator (D-MRI, GP-MRI, GB-MRI, CE-CT) where
Literature searches were conducted in four bibliographic data- three or more studies reported both sensitivity and specificity data.
bases: PubMed, EMBASE, the Cochrane Library, and the Centre for Each comparator was analyzed separately and only studies that directly
Reviews and Dissemination (CRD) of the University of York on Feb- compared GA-MRI to the comparator technique were included in
ruary 19 2015. Searches were updated on February 29 2016. A com- each analysis. The mixed modeling approach described by Reitsma
prehensive search strategy using a range of relevant search terms was et al was subsequently used to provide estimated summaries of sensitiv-
used, including key words, phrases, Medical Subject Headings ity and specificity and the corresponding 95% confidence ellipses.23
(MeSH), and Emtree terms (Supplementary Table 1). The bibliogra- The HSROC curve described by Rutter and Gatsonis was generated
phies of all included studies were hand-searched for any relevant and the associated area under the curve (AUC) was compared across
references that may have been missed by the literature searches. imaging techniques.24 Meta-analyses were conducted in R i386 v. 3.1.2
Title and abstracts were screened by one researcher to exclude using the “mada” package.25 Publication bias was not assessed due to
studies that obviously did not meet the prespecified inclusion criteria. the inherent difficulty in estimating publication bias for diagnostic
All other studies were retrieved for full-text assessment, which was con- studies and inaccuracy in interpretation of results.26
ducted independently by two researchers. Studies were included to
address the research question if they met the prespecified criteria. Any
Results
discrepancies between the two reviewers were resolved by consensus. Literature Identification and Description
Studies were included if they compared the diagnostic accu- From a total of 10,464 studies identified through literature
racy of GA-MRI to a comparator imaging technique in patients searches, 35 studies were identified that compared the diagnostic
with known or suspected HCC and used a valid reference standard accuracy of GA-MRI to CE-CT, D-MRI, GP-MRI, or GB-

Duncan et al.: Gadoxetic Acid for HCC Characterization
MRI. Of these, 19 were excluded for failing to report complete no significant difference between GA-MRI and C-CT were
diagnostic accuracy data, leaving 16 studies with 1138 patients found.
and 1536 lesions included in this review (Supplementary Fig. 1). Three studies compared GA-MRI to CE-CT for diag-
Twelve studies included patients with lesions of any nosing lesions less than 3 cm in diameter; three of these
size,27–38 while four studies restricted enrolment to patients reported both sensitivity and specificity data. From the
with lesions smaller than 3 cm in diameter.39–42 bivariate analysis, GA-MRI had an estimated sensitivity and
All of the included studies enrolled patients with specificity of 0.919 (95% CI 5 0.834, 0.962) and 0.936
known or suspected HCC who underwent imaging in a (95% CI 5 0.882, 0.966). CE-CT had an estimated sensi-
nonsurveillance setting. The baseline characteristics of the tivity and specificity of 0.637 (95% CI 5 0.565, 0.704) and
T1 included patients for each study are described in Table 1, 0.971 95% CI 5 0.937, 0.987). There was no overlap in
and are in line with known risk factors for HCC.43 the 95% confidence ellipses reflecting the statistical signifi-
CE-CT was a comparator in 10 studies, with 4, 8, 16, cance of the results (Fig. 2). A moderate amount of hetero- F2
40, or 64 channel multidetector CT (MD-CT) machines geneity was observed in the data for GA-MRI; there was no
used.28,29,32–35,38,40–42 The contrast agents used for CE-CT apparent heterogeneity in the CE-CT data. The complete
were all nonionic, iodine-based contrasts, including iopro- separation of the prediction intervals reflects that future
mide, iomeprol, and iopamidol. D-MRI was a comparator studies are predicted to confirm the diagnostic superiority of
in seven studies.27–31,37,39 GP-MRI was a comparator in GA-MRI over CE-CT for diagnosing small lesions. No sub-
one study.36 No studies were identified where GB was a group analysis was undertaken due to the small number of
comparator and full diagnostic accuracy data were reported. studies including patients with small lesions.
Quality Appraisal
Comparison Between GA-MRI and D-MRI
Following assessment with the QUADAS-II tool, the quality
Six studies that included patients with lesions of any size
of the included studies was mixed and no study was judged
evaluated the additional benefit conferred by including hep-
to have a low risk of bias across all domains (Supplementary
atobiliary images (GA-MRI) compared to dynamic imaging
Table 2 and Supplementary Fig. 2). All studies were judged
alone (D-MRI). Bivariate model analyses were undertaken
to have included patients who matched the review questions
and HSROC curves were generated (Fig. 3 and Table 2). F3
of this report, and there were no applicability issues identi-
GA-MRI had an estimated sensitivity of 0.907 (95%
fied in any study.
CI 5 0.870, 0.934) and an estimated specificity of 0.929
(95% CI 5 0.877, 0.961). D-MRI had an estimated sensi-
Comparison Between GA-MRI and CE-CT
tivity of 0.820 (95% CI 5 0.776, 0.857) and specificity of
Seven studies compared GA-MRI with CE-CT in lesions of
0.934 (95% CI 5 0.881, 0.964). There was a small overlap
any size. Bivariate model analysis was undertaken and
between the 95% confidence ellipses of GA-MRI and D-
F1 HSROC curves were generated (Fig. 1 and Table 2).
MRI, indicating that the differences in diagnostic accuracy
T2 GA-MRI had an estimated sensitivity of 0.881 (95% confi-
do not reach the level of statistical significance (a 5 0.05).
dence interval [CI] 5 0.766, 0.944) and an estimated speci-
No heterogeneity was observed; the prediction ellipses over-
ficity of 0.926 (95% CI 5 0.829, 0.97). CE-CT had an
lay the confidence ellipses. Subgroup analyses (Table 2)
estimated sensitivity of 0.713 (95% CI 5 0.577, 0.819) and
found that for studies reporting that authors had no conflict
specificity of 0.918 (95% CI 5 0.829, 0.963). While the
of interest and those not using a consistent reference stand-
summary estimate of GA-MRI is well separated from that
ard for all patients, GA-MRI had significantly higher esti-
of CE-CT, reflecting the greater diagnostic ability of
mated sensitivity then D-MRI.
GA-MRI, the overlap between the 95% confidence ellipses
No analysis was undertaken comparing GA-MRI to
indicates that the differences in diagnostic accuracy do not
D-MRI for small lesions due to a lack of data. One study39
reach the level of statistical significance (a 5 0.05). The
reported this comparison in patients with lesions less than
widths of the prediction ellipses reflect that there is moder-
2 cm in diameter. Golfieri et al39 reported no difference in
ate heterogeneity present and the overlapping prediction
sensitivity or specificity between GA-MRI and D-MRI.
intervals reflect uncertainty that future studies will report
significant differences between the two techniques.
Post-hoc subgroup analyses are reported in Table 2. In Comparison Between GA-MRI and GP-MRI
studies reporting that they enrolled consecutive patients and One study reported diagnostic accuracy data for the com-
in studies not using a consistent reference standard for all parison between GA-MRI and GP-MRI.36 Pahade et al36
patients, GA-MRI was associated with significantly higher found no statistically significant difference between GA-
estimated sensitivity than CE-CT. For all other subgroups MRI and GP-MRI in patients with lesions of any size.
January 2017 283

284
TABLE 1. Study, Patient, and Imaging Characteristics of the Diagnostic Imaging Studies
Study ID Study designc Patients/lesions Indication prior Age (years) Comparator Reference Conflict
lesion size tests male/female contrast standard of interest
Studies including patients with any sized lesion

Ahn et al (2010) Retrospective, 59/113 2.8 cm [0.4-11] Suspected HCC 57 [29-75] 50/9 D-MRI:1.5T or PB or SS or FU NR
Consecutive PT: US, CT or SM 3.0T GA (0.1 ml/kg) ( 12 months)
Baek et al (2012) Retrospective, 51/110 2.98 cm (0.2-10) Suspected HCC NR [32-80] 43/8 CE-CT: 64, 16 or 4 PB or SS or FU NR
Consecutive PT: US and/or SM channel MDCT ( 12 monthsb)
Non-ionic (2 ml/kg).
D-MRI: 3.0T GA
(0.1 ml/kg)
Chen et al Retrospective 139/111 NR Known or suspected 68 6 11 101/38 D-MRI: 1.5T GA SS, PB, FU None
(2016) Consecutive HCC PT: NR (0.1 ml/kg) CE-CT: (>12 months)
64 channel MDCT
Omnipaqe, Iomeron
or Iopamiron 600
mgL/kg
Faletti et al Retrospective, 28/50 NR Known HCC 54.7 6 5.9 27/1 D-MRI: 1.5T GA SS NR
(2015) NR if consecutive PT: NR (0.1 ml/kg)
Filippone et al Retrospective, 34/39 3.2 cm (0.8-7.5) Suspected HCC 58.76NR 27/7 D-MRI: 1.5T GA PB, SS, FU None
(2010) NR if consecutive PT: NR (0.1 ml/kg) (>12 months)
Hwang et al Retrospective, 108/162 1.6 cm (0.7-5.1) Known HCC PT: NR (37-86) 81/27 CE-CT: 40 or 64 PB or SS or FU NR
(2010) Consecutive imaging and/or SM channel MDCT (1-6 months)
Iopamidol (120 ml)
Kakihara et al Retrospective, 15/63 1.3 cm [0.6-6.6] Suspected HCC 55 [45-71] 11/4 CE-CT: 64 channel SS NR
(2013) NR if consecutive PT: NR MDCT Iopamidol
(1.62 ml/kg)
Maiwald et al Prospective, 50/50 Known or suspected 60.6 (29-84) 42/8 CE-CT: 64 channel PB or SS or FU Yesd
(2014) NR if consecutive HCC PT: US, SM MDCT Iopromide (6 months)
100 ml
Nishie et al Retrospective, 12/26 1.2 cm [0.5-3.4] Suspected HCC 60 [46-72] 3/9 CE-CT: 64 channel SS NR
(2015) Consecutive PT: NR MDCT Iopamidol
(1.62 ml/kg, maxi-
Stage:
mum 100ml)

Volume 45, No. 1
Page: 284
January 2017
TABLE 1: Continued
Study ID Study designc Patients/lesions Indication prior Age (years) Comparator Reference Conflict
lesion size tests male/female contrast standard of interest
Pahade et al Retrospective, 30/20 2.4 cm (1.0-8.0) Liver transplant 56 (35-75) 26/4 GP-MRI: 1.5T GP SS None
(2016) NR if consecutive PT: NR (0.1 mmol/kg)
Phongkitkarun Retrospective, 100/59 1.9 cm (1-12) Suspected HCC PT: 59.5 (27-89) D-MRI: 1.5T GA SS, PB or FU NR
et al (2013) Consecutive SM 71/29 (0.1 ml/kg) ( 6 months)
Raman et al NR if prospective, 177/269 NR Suspected or known 54.2 (22-81)a CECT: singe detec- PB, FU Yese
(2010) NR if consecutive HCC PT: NR 124/111a tor scanners. iodin- ( > 12 months)
ated, non-ionic
(100-200 ml )
Studies including patients with lesions < 3 cm in diameter
Golferi et al Prospective, 127/173 1.6 cm Suspected HCC PT: 54 (31-77) 127/0 D-MRI: 1.5T GA SS or PB NR
(2011) Consecutive (0.4-2.0) US (0.1 ml/kg)
Haradome Retrospective, 75/86 1.74 cm 6 0.65 Suspected HCC PT: 54.7 (42-67) CE-CT:16 channel PB or SS NR
(2011) Consecutive NR 60/15 MDCT Non-ionic
contrast (1.8 ml/kg)
Sano et al Retrospective, 64/108 1.27 cm HCC resection PT: Male: 65.6 6 9.5 CE-CT: 16 channel SS None
(2011) Consecutive (0.4-2.0) NR Female 71.0 6 MDCT Iodine (600
8.6 47/17 mgkg)
Sun et al (2010) Retrospective, NR if 69/97 1.37 cm 6 0.41 Suspected HCC PT: 55.8 (39-73) CE-CT: 8, 16 or 64 SS < PB or NR
consecutive US or clinical 56/13 channel MDCT FU ( 6
suspicion Iopromide months)
(90-150 ml)
CE-CT 5 contrast-enhanced CT. CT 5 computed tomography. D-MRI 5 dynamic MRI. DW-MRI 5 diffusion-weighted MRI. FU 5 clinical and/or radiological follow-up. GA 5 gadoxetic
acid. GA-MRI 5 gadoxetic acid-enhanced MRI. GP-MRI 5 gadopentetic acid enhanced MRI. HCC 5 hepatocellular carcinoma. MDCT 5multidetector CT. MRI 5 magnetic resonance
imaging. NR 5 not reported. NR 5 not reported. PB 5 percutaneous biopsy. PT 5 prior test. SM 5 serum markers. SS 5 surgical specimen. US 5 ultrasound.
a
For total cohort of enrolled patients, not those analyzed.
b
Follow-up for15 patients.
c
Was the study prospective or retrospective? Consecutive or nonconsecutive?
d
Supported by a grant from Bayer.
e
Stage:
Funding and editorial support from Bayer.

Data reported as: mean 6 SD, mean (range), median [range], median [IQR].

285
Page: 285
Further, the direct survival benefits for GA-MRI com-

pared to other MRI techniques are not established. One
study of 823 patients compared survival of patients with a
known single-nodular HCC following either CE-CT (377
patients), GA-MRI plus CE-CT (323 patients), or other
MRI (predominantly GB-MRI) plus CE-CT (123
patients).46 Kim et al46 reported that GA-MRI plus CE-CT
was associated with higher overall survival (82% vs. 74%,
P 5 0.001) and recurrence-free survival (56% vs. 42%,
P 5 0.005) than CE-CT alone. It is not clear what propor-
tion of the survival benefit is due to the increase in sensitiv-
ity gained by introducing a secondary imaging evaluation.
C There was no statistical difference in either overall survival
O (82% vs. 78%, P 5 0.46) and recurrence-free patient sur-
L
O vival (P 5 0.35) between GA-MRI plus CE-CT and other
R MRI techniques plus CE-CT.
FIGURE 1: HSROC curves of GA-MRI and CE-CT in patients
A second study of 147 patients with suspected early
with lesions of any size. The summary receiver operating curve
(SROC) of gadoxetic acid-enhanced MRI (GA-MRI) is shown as a HCC compared survival following GA-MRI compared to
solid red line. The SROC for contrast-enhanced CT (CE-CT) is MRI without GA contrast.47 Matsuda et al47 reported no
shown as a dashed blue line. The summary estimate for the difference in overall 1-, 3-, and 5-year survival rates between
diagnostic accuracy of GA-MRI is shown as a red square, while
the summary estimate for the diagnostic accuracy of CE-CT is patients imaged with GA-MRI compared to conventional
shown as a blue triangle. The 95% confidence ellipses are MRI (P 5 0.38). However, GA-MRI was associated with
shown around the summary estimates in blue and red. The statistically higher 1-, 3-, and 5-year recurrence-free survival
95% prediction intervals are shown in black.
rates.
The available data suggest that any changes in patient
outcomes associated with the use of GA-MRI were due to
Discussion the improved detection of lesions. Improved lesion detection
In this review we present a direct comparison of the diag- may result in changes in management for small numbers of
nostic accuracy of GA-MRI with CE-CT and other MRI patients, including more accurate resection margins, reduced
techniques for the characterization of HCC lesions. For delay in the treatment of lesions, and avoidance of addi-
lesions of any size, differences in accuracy between GA-MRI tional surgical procedures or follow-up imaging.48–50 Liver
and CE-CT or D-MRI do not reach the level of statistical resection has a relatively high mortality rate of 1 in 10051;
significance. For small lesions, GA-MRI has a higher diag- given these circumstances, accurate diagnosis and avoidance
nostic accuracy than CE-CT. of unnecessary surgery are particularly important in optimiz-
One issue with the evidence base for this review was ing patient outcomes.
the number of studies that did not report true-negative The results of this systematic review should be viewed
numbers due to the inherent difficulty in confirming nega- in the light of the limitations of the included primary stud-
tive cases. This prevented a number of studies from being ies. The diagnostic accuracy studies were judged to have a
included in the bivariate analysis. moderate risk of bias based on the QUADAS-II quality
Analysis was further limited by a very small number of appraisal tool.22 Several sources of bias were commonly
studies reporting results comparing GA-MRI to GP-MRI identified in the included studies. In the majority of studies,
and GB-MRI. Studies comparing GA-MRI to D-MRI in not all patients received the same reference standard. This is
patients with small lesions was similarly lacking. Of particu- clinically appropriate, as patients with diagnosed HCC can
lar interest was the comparison between the two hepatobili- undergo resection and have pathological confirmation of
ary contrast agents (GA and GB). Both contrast agents are diagnosis, whereas for lesions diagnosed as benign, appropri-
hepatobiliary-specific, albeit with different degrees of hepatic ate reference standard is clinical and radiological follow-up.
uptake and delays before hepatobiliary imaging can be per- All studies used appropriate reference standards. Subgroup
formed.44,45 Surprisingly, there is a distinct lack of compara- analyses found that in studies using more than one reference
tive evidence on the relative diagnostic accuracy of GA-MRI standard, GA-MRI had higher sensitivity than comparator
compared to GB-MRI. In light of the differing clinical and imaging techniques, whereas in studies using surgical speci-
economic factors associated with these contrast agents, we men as the reference standard for all patients there was no
suggest that the comparison between these two agents is a difference in estimated sensitivity between GA-MRI and
topic that warrants further research. comparator techniques. Therefore, the bias identified in the

TABLE 2. Summary of Results From the Subgroup Analyses
All sized lesions GA-MRI All sized lesions CE-CT
January 2017
Subgroup Number of Sensitivity Specificity Sensitivity Specificity
studies estimate estimate estimate estimate
[95% CI] [95% CI] [95% CI] [95% CI]
All studies 7 0.881 [0.766, 0.94] 0.926 [0.829, 0.97] 0.713 [0.577, 0.819] 0.918 [0.829, 0.963]
a
Retrospective studies 5 0.872 [0.733, 0.944] 0.941 [0.857, 0.977] 0.689 [0.539, 0.807] 0.935 [0.859, 0.971]
b
Consecutive recruitment 4 0.901 [0.834, 0.943] 0.951 [0.91, 0.974] 0.698 [0.571, 0.801] 0.956 [0.917, 0.977]
NR if consecutive recruitment 3 0.853 [0.501, 0.971] 0.894 [0.538, 0.984] 0.738 [0.403, 0.922] 0.876 [0.583, 0.973]
RS: Surgical specimen only 2 0.667 [0.35, 0.879] 0.972 [0.866, 0.995] 0.616 [0.228, 0.898] 0.963 [0.893, 0.988]
b
RS: mixed 5 0.919 [0.875, 0.949] 0.894 [0.757, 0.958] 0.751 [0.916, 0.848] 0.894 [0.755, 0.96]
Reported conflict of interest 2 0.934 [0.836, 0.975] 0.764 [0.665, 0.841] 0.852 [0.740, 0.921] 0.764 [0.665, 0.841]
NR if conflict of interest 4 0.826 [0.603, 0.937] 0.959 [0.92, 0.979] 0.637 [0.440, 0.798] 0.965 [0.931, 0.982]
All sized lesions GA-MRI All sized lesions D-MRI

Subgroup Number of Sensitivity Specificity Sensitivity Specificity
studies estimate estimate estimate estimate
[95% CI] [95% CI] [95% CI] [95% CI]
All studies 6 0.907 [.870, 0.934] 0.929 [0.877, 0.961] 0.820 [0.776, 0.857] 0.934 [0.881, 0.964]
Consecutive recruitment 4 0.905 [0.863, 0.936] 0.942 [0.882, 0.972] 0.821 [0.759, 0.870] 0.943 [0.878, 0.975]
NR if consecutive 2 0.935 [0.853, 0.973] 0.927 [0.749, 0.982] 0.831 [0.738, 0.896] 0.968 [0.803, 0.995]
a b
RS: mixed 5 0.908 [0.869, 0.936] 0.941 [0.886, 0.97] 0.812 [0.759, 0.856] 0.934 [0.875, 0.966]
b
No conflict of interest 2 0.927 [0.846, 0.967] 0.946 [0.809, 0.987] 0.761 [0.658, 0.840] 0.948 [0.813, 0.987]
NR if conflict of interest 4 0.903 [0.860, 0.934] 0.931 [0.869, 0.965] 0.842 [0.793, 0.881] 0.93 [0.866, 0.965]
a
If there was only one study in a particular subgroup, then results from this group were not reported. Instead, analysis for the remaining studies was reported as a sensitivity analysis remov-
ing the unique study and should be compared to the entire cohort of studies. Estimates of sensitivity and specificity are from the bivariate model.
b
Stage:
Results are statistically significant and favor GA-MRI.

CE-CT 5 contrast-enhanced computed tomography. CI 5 confidence interval. D-MRI 5 dynamic MRI, GA-MRI 5 MRI with gadoxetic acid contrast, NR 5 not reported, RS 5 reference
standard.

287
Page: 287
this potential bias did not appear to affect the findings of

this review. Most studies failed to report whether the refer-
ence standard results were interpreted without knowledge of
the index and comparator tests; however, all but one studies
explicitly stated that index and comparator images were
reviewed without knowledge of other tests and with prede-
fined thresholds for lesion characterization, and the effect of
this bias was thought to be minimal. Patient selection crite-
ria was unclear in eight of the studies; therefore, it was
unclear whether any patients had been inappropriately
excluded. This may have increased the reported sensitivities
by excluding hard to diagnose cases. A lack of transparency
C in the timing of the reference standard analyses compared to
O
L the index and comparator tests was a commonly identified
O limitation; however, as all included studies performed both
R
index and comparator imaging on the same groups of
FIGURE 2: HSROC curves of GA-MRI and CE-CT in patients
with lesions less than 3 cm in diameter. The summary receiver patients, it is possible that any sources of bias would simi-
operating curve (SROC) of gadoxetic acid-enhanced MRI (GA- larly affect all images and therefore the effect on the overall
MRI) is shown as a solid red line. The SROC for contrast-
comparative results would be minimal.
enhanced CT (CE-CT) is shown as a dashed blue line. The sum-
mary estimate for the diagnostic accuracy of GA-MRI is shown While conducting this review, no attempt was made to
as a red square, while the summary estimate for the diagnostic assess publication bias. The authors decided not to assess
accuracy of CE-CT is shown as a blue triangle. The 95% confi-
any potential bias for this outcome due to difficulties in
dence ellipses are shown around the summary estimates in
blue and red. The 95% prediction intervals are shown in black. assessing and interpreting any funnel plot asymmetry for
analyses of fewer than 10 studies.28
reference standard may have led to an overestimation in the In conclusion, GA-MRI is superior to CE-CT in
superiority of GA-MRI. However, it should be noted that patients with lesions less than 3 cm in diameter. However, the
the overall results of the review found no differences in sen- results of this review do not show any superiority of GA-MRI
sitivity between GA-MRI and CE-CT or D-MRI; therefore, compared to CE-CT or D-MRI in patients with lesions of any
size. More research into the optimum imaging modality for
characterizing HCC is required. Particularly, future research
should focus on comparing GA-MRI to D-MRI and GB-MRI
and into the whether any improvements in diagnostic accuracy
confer improvement in patient outcomes.
Acknowledgments
The authors thank Tamsin Garrod, Anje Scarfe, and Dag-
mara Riitano for their contribution to the data extraction
and quality appraisal for this review and Associate Professor
Wendy Babidge for reviewing the draft article.
C Conflict of Interest
O No funding was received for the completion of this article.
L
O The authors report no financial interests or other competing
R
interests with respect to this publication. No manufacturer
FIGURE 3: HSROC curves of GA-MRI and D-MRI in patients
with lesions of any size. The summary receiver operating curve funded this research or played any role in conducting the
(SROC) of gadoxetic acid-enhanced MRI (GA-MRI) is shown as a study, collection of data, or presentation of results.
solid red line. The SROC for dynamic-MRI (D-MRI) is shown as a
dotted green line. The summary estimate for the diagnostic
accuracy of GA-MRI is shown as a red square, while the sum-
References
mary estimate for the diagnostic accuracy of D-MRI is shown as 1. Forner A, Llovet JM, Bruix J. Hepatocellular carcinoma. Lancet 2012;
a green circle. The 95% confidence ellipses are shown around 379:1245–1255.
the summary estimates in red and green. The 95% prediction 2. Bruix J, Sherman M. Management of hepatocellular carcinoma: an
intervals are shown in black. update. Hepatology 2011;53:1020–1022.

3. Parikh S, Hyman D. Hepatocellular cancer: a guide for the internist. summary measures in diagnostic reviews. J Clin Epidemiol 2005;58:
Am J Med 2007;120:194–202. 982–990.
4. Saraswat VA, Pandey G, Shetty S. Treatment algorithms for managing 24. Rutter CM, Gatsonis CA. A hierarchical regression approach to meta-analysis
hepatocellular carcinoma. J Clin Exp Hepatol 2014;4:S80–S89. of diagnostic test accuracy evaluations. Stat Med 2001;20:2865–2884.
5. Bialecki ES, Di Bisceglie AM. Diagnosis of hepatocellular carcinoma. 25. Doebler P, Holling H. Meta-analysis of diagnostic accuracy with mada.
HPB (Oxford) 2005;7:26–34. 2012 CRAN. Available at: http://cran.mtu.edu/web/packages/mada/
vignettes/mada.pdf. Accessed 1 October 2015.
6. Delis SG, Dervenis C. Selection criteria for liver resection in patients
with hepatocellular carcinoma and chronic liver disease. World J Gas- 26. Macaskill P, Gatsonis C, Deeks J, Harbord R, Takwoingi Y. Cochrane
troenterol 2008;14:3452–3460. handbook for systematic reviews of diagnostic test accuracy. Chapter
10: analysing and presenting results. 2010 Cochrane Collaboration;
7. Poon R, Fan S-T, Lo C-M, Liu C-L, Wong J. Intrahepatic recurrence
accessed 1 October 2015 http://srdta.cochrane.org/.
after curative resection of hepatocellular carcinoma: long-term results
of treatment and prognostic factors. Ann Surg 1999;229:216. 27. Ahn SS, Kim MJ, Lim JS, Hong HS, Chung YE, Choi JY. Added value
of gadoxetic acid-enhanced hepatobiliary phase MR imaging in the
8. Poon RT-P, Fan ST, Lo CM, Liu CL, Wong J. Long-term survival and
diagnosis of hepatocellular carcinoma. Radiology 2010;255:459–466.
pattern of recurrence after resection of small hepatocellular carcinoma
in patients with preserved liver function: implications for a strategy of 28. Baek CK, Choi JY, Kim KA, et al. Hepatocellular carcinoma in patients
salvage transplantation. Ann Surg 2002;235:373–382. with chronic liver disease: a comparison of gadoxetic acid-enhanced
MRI and multiphasic MDCT. Clin Radiol 2012;67:148–156.
9. Portolani N, Coniglio A, Ghidoni S, et al. early and late recurrence
after liver resection for hepatocellular carcinoma: prognostic and ther- 29. Chen N, Motosugi U, Morisaka H, et al. Added value of a gadoxetic acid-
apeutic implications. Ann Surg 2006;243:229–235. enhanced hepatocyte-phase image to the LI-RADS system for diagnosing
10. Shah SA, Cleary SP, Wei AC, et al. Recurrence after liver resection for hepatocellular carcinoma. Magn Reson Med Sci 2016;15:49–59.
hepatocellular carcinoma: risk factors, treatment, and outcomes. Sur- 30. Faletti R, Cassinis MC, Fonio P, et al. Multiparametric Gd-EOB-DTPA
gery 2007;141:330–339. magnetic resonance in diagnosis of HCC: dynamic study, hepatobili-
11. Imamura H, Matsuyama Y, Tanaka E, et al. Risk factors contributing to ary phase, and diffusion-weighted imaging compared to histology
early and late phase intrahepatic recurrence of hepatocellular carci- after orthotopic liver transplantation. Abdom Imaging 2015;40:46–55.
noma after hepatectomy. J Hepatol 2003;38:200–207. 31. Filippone A, Cianci R, Patriarca G, Sabatino F, Tartaro A, Cotroneo
12. Soong R-S, Yu M-C, Chan K-M, et al. Analysis of the recurrence risk AR. The value of gadoxetic acid-enhanced hepatospecific phase MR
factors for the patients with hepatocellular carcinoma meeting Univer- imaging for characterization of hepatocellular nodules in the cirrhotic
sity of California San Francisco criteria after curative hepatectomy. liver. Eur J Clin Med Oncol 2010;2.
World J Surg Oncol 2011;9. 32. Hwang J, Kim SH, Kim YS, et al. Gadoxetic acid-enhanced MRI versus
13. Marrero JA, Ahn J, Rajender Reddy K. ACG clinical guideline: the multiphase multidetector row computed tomography for evaluating
diagnosis and management of focal liver lesions. Am J Gastroenterol the viable tumor of hepatocellular carcinomas treated with image-
2014;109:1328–1347; quiz 1348. guided tumor therapy. J Magn Reson Imaging 2010;32:629–638.
14. Chen L, Zhang L, Liang M, et al. Magnetic resonance imaging with 33. Kakihara D, Nishie A, Harada N, et al. Performance of gadoxetic acid-
gadoxetic acid disodium for the detection of hepatocellular carci- enhanced MRI for detecting hepatocellular carcinoma in recipients of
noma: a meta-analysis of 18 studies. Acad Radiol 2014;21:1603–1613. living-related-liver-transplantation: comparison with dynamic multide-
tector row computed tomography and angiography-assisted com-
15. Chou R, Cuevas C, Fu R, et al. Imaging techniques for the diagnosis puted tomography. J Magn Reson Imaging 2013;40:1112–1120.
of hepatocellular carcinoma: a systematic review and meta-analysis.
Ann Intern Med 2015;162:697–711. 34. Maiwald B, Lobsien D, Kahn T, Stumpp P. Is 3-Tesla Gd-EOB-DTPA-
enhanced MRI with diffusion-weighted imaging superior to 64-slice
16. Junqiang L, Yinzhong W, Li Z, et al. Gadoxetic acid disodium (Gd- contrast-enhanced CT for the diagnosis of hepatocellular carcinoma?
EOBDTPA)-enhanced magnetic resonance imaging for the detection PLoS One 2014;9:e111935.
of hepatocellular carcinoma: a meta analysis. J Magn Reson Imaging
2014;39:1079–1087. 35. Nishie A, Kakihara D, Asayama Y, et al. Detectability of hepatocellular
carcinoma on gadoxetic acid-enhanced MRI at 3 T in patients with
17. Lee YJ, Lee JM, Lee JS, et al. Hepatocellular carcinoma: diagnostic severe liver dysfunction: clinical impact of dual-source parallel radio-
performance of multidetector CT and MR imaging—a systematic frequency excitation. Clin Radiol 2015;70:254–261.
review and meta-analysis. Radiology 2015:140690.
36. Pahade JK, Juice D, Staib L, et al. Is there an added value of a hepa-
18. Liu X, Zou L, Liu F, Zhou Y, Song B. Gadoxetic acid disodium- tobiliary phase with gadoxetate disodium following conventional MRI
enhanced magnetic resonance imaging for the detection of hepato- with an extracellular gadolinium agent in a single imaging session for
cellular carcinoma: a meta-analysis (Provisional abstract). PLoS One detection of primary hepatic malignancies? Abdom Radiol (NY) 2016
2013;8:e70896.
[Epub ahead of print].
19. Wu LM, Xu JR, Gu HY, et al. Is liver-specific gadoxetic acid-enhanced 37. Phongkitkarun S, Limsamutpetch K, Tannaphai P, Jatchavala J. Added
magnetic resonance imaging a reliable tool for detection of hepato-
value of hepatobiliary phase gadoxetic acid-enhanced MRI for diag-
cellular carcinoma in patients with chronic liver disease? Dig Dis Sci
nosing hepatocellular carcinoma in high-risk patients. World J Gastro-
2013;58:3313–3325.
enterol 2013;19:8357–8365.
20. MSAC. 1372: MRI liver — scan for the detection and characterisation
38. Raman SS, Leary C, Bluemke DA, et al. Improved characterization of
of focal liver lesions. 2014 Medical Services Advisory Committee;
focal liver lesions with liver-specific gadoxetic acid disodium-
accessed 1 October 2015 http://www.msac.gov.au/internet/msac/
enhanced magnetic resonance imaging: a multicenter phase 3 clinical
publishing.nsf/Content/1372-public.
trial. J Comput Assist Tomogr 2010;34:163–172.
21. Liberati A, Altman DG, Tetzlaff J, et al. The PRISMA statement for report-
39. Golfieri R, Renzulli M, Lucidi V, Corcioni B, Trevisani F, Bolondi L.
ing systematic reviews and meta-analyses of studies that evaluate health-
Contribution of the hepatobiliary phase of Gd-EOB-DTPA-enhanced
care interventions: explanation and elaboration. BMJ 2009;339:b2700.
MRI to Dynamic MRI in the detection of hypovascular small ( 2 cm)
22. Whiting PF, Rutjes AW, Westwood ME, et al. QUADAS-2: a revised HCC in cirrhosis. Eur Radiol 2011;21:1233–1242.
tool for the quality assessment of diagnostic accuracy studies. Ann
40. Haradome H, Grazioli L, Tinti R, et al. Additional value of gadoxetic acid-
Intern Med 2011;155:529–536.
DTPA-enhanced hepatobiliary phase MR imaging in the diagnosis of
23. Reitsma JB, Glas AS, Rutjes AW, Scholten RJ, Bossuyt PM, Zwinderman early-stage hepatocellular carcinoma: comparison with dynamic triple-
AH. Bivariate analysis of sensitivity and specificity produces informative phase multidetector CT imaging. J Magn Reson Imaging 2011;34:69–78.
January 2017 289

41. Sano K, Ichikawa T, Motosugi U, et al. Imaging study of early hepato- 47. Matsuda M, Ichikawa T, Amemiya H, et al. Preoperative gadoxetic
cellular carcinoma: usefulness of gadoxetic acid-enhanced MR imag- acid-enhanced MRI and simultaneous treatment of early hepatocellu-
ing. Radiology 2011;261:834–844. lar carcinoma prolonged recurrence-free survival of progressed hepa-
tocellular carcinoma patients after hepatic resection. HPB Surg 2014;
42. Sun HY, Lee JM, Shin CI, et al. Gadoxetic acid-enhanced magnetic
2014.
resonance imaging for differentiating small hepatocellular carcinomas
(<2 cm in diameter) from arterial enhancing pseudolesions: Special 48. Cha DI, Lee MW, Kim YK, et al. Assessing patients with hepatocellular
emphasis on hepatobiliary phase imaging. Invest Radiol 2010;45:96– carcinoma meeting the Milan criteria: is liver 3 Tesla MR with gadox-
103. etic acid necessary in addition to liver CT? J Magn Reson Imaging
2014;39:842–852.
43. Bosch FX, Ribes J, Dıaz M, Cl
eries R. Primary liver cancer: worldwide
incidence and trends. Gastroenterology 2004;127:S5–S16. 49. Hammerstingl R, Huppertz A, Breuer J, et al. Diagnostic efficacy of
gadoxetic acid (Primovist)-enhanced MRI and spiral CT for a therapeu-
44. Reimer P, Schneider G, Schima W. Hepatobiliary contrast agents for
tic strategy: comparison with intraoperative and histopathologic find-
contrast-enhanced MRI of the liver: properties, clinical development
ings in focal liver lesions. Eur Radiol 2008;18:457–467.
and applications. Eur Radiol 2004;14:559–578.
50. Yoo SH, Choi JY, Jang JW, et al. Gd-EOB-DTPA-enhanced MRI is
45. Van Beers BE, Pastor CM, Hussain HK. Primovist, Eovist: what to
better than MDCT in decision making of curative treatment for hepa-
expect? J Hepatol 2012;57:421–429.
tocellular carcinoma. Ann Surg Oncol 2013;20:2893–2900.
46. Kim H-D, Lim Y-S, Han S, et al. Evaluation of early-stage hepatocellu-
51. Belghiti J, Hiramatsu K, Benoist S, Massault PP, Sauvanet A, Farges O.
lar carcinoma by magnetic resonance imaging with gadoxetic acid
Seven hundred forty-seven hepatectomies in the 1990s: an update to
detects additional lesions and increases overall survival. Gastroenter-
evaluate the actual risk of liver resection. J Am Coll Surg 2000;191:38–46.
ology 2015;148:1371–1382.

ORIGINAL RESEARCH
Acute Kidney Damage Induced by

Low- and Iso-osmolar Contrast Media in
Rats: Comparison Study With Physiologic
MRI and Histologic-Gene Examination
Chen-Jiang Wu, MM,1 Mei-Ling Bao, MM,2 Qing Wang, MM,1
Xiao-Ning Wang, MD,1 Xi-Sheng Liu, MD,1 Hai-Bin Shi, MD,1 and
Yu-Dong Zhang, MD., PhD1*
Purpose: To investigate the physiopathological effects of low- and iso-osmolar contrast media (CM) on renal function
with physiologic MRI and histologic-gene examination.
Materials and Methods: Forty-eight rats underwent time-course DWI and DCE-MRI at 3.0 Tesla (T) before and 5–15 min
after exposure of CM or saline (Iop.370: 370 mgI/mL iopromide; Iod.320: 320 mgI/mL iodixanol; Iod.270: 270 mgI/mL iodix-
anol; 4 gI/kg body weight). Intrarenal viscosity was reflected by apparent diffusion coefficient (ADC). Renal physiologies
were evaluated by DCE-derived glomerular filtration rate (GFR), renal blood flow (RBF), and renal blood volume (RBV).
Potential acute kidney injury (AKI) was determined by histology and the expression of kidney injury molecule 1 (Kim-1).
Results: Iop.370 mainly increased ADC in inner-medulla (䉭ADCIM: 12.3 6 11.1%; P < 0.001). Iod.320 and Iod.270 mainly
decreased ADC in outer-medulla (䉭ADCIM; Iod.320: 16.8 6 7.5%; Iod.270: 18.1 6 9.5%; P < 0.001) and inner-medulla
(䉭ADCIM; Iod.320: 28.4 6 9.3%; Iod.270: 30.3 6 6.3%; P < 0.001). GFR, RBF and RBV were significantly decreased by
Iod.320 (䉭GFR: 45.5 6 24.1%; 䉭RBF: 44.6 6 19.0%; 䉭RBV: 35.2 6 10.1%; P < 0.001) and Iod.270 (33.2 6 19.0%;
38.1 6 15.6%; 30.1 6 10.1%; P < 0.001), while rarely changed by Iop.370 and saline. Formation of vacuoles and increase
in Kim-1 expression was prominently detected in group of Iod.320, while rarely in Iod.270 and Iop.370.
Conclusion: Iso-osmolar iodixanol, given at high-dose, produced prominent AKI in nonhydrated rats. This renal dysfunc-
tion could be assessed noninvasively by physiologic MRI.
C ontrast-induced acute kidney injury (CI-AKI) is an iat-

rogenic complication caused by exposure to iodinated
contrast media (CM).1,2 As the use of iodinated CM
low-osmolar CM (LOCM; e.g., iopromide), and (ii) high-
viscous, iso-osmolar CM (IOCM; e.g., iodixanol). It need
reminded that, even called low-osmolality CM, iopromide
increases, CI-AKI has become the third leading cause for actually has 2–3 times the osmolality of serum; while iodix-
hospital acquired acute renal failure.3–5 Although there is anol has the same osmolality as plasma.12 Therefore,
relatively low prevalence of CI-AKI in healthy populations, LOCM actually has a higher osmolality than that of
the consequences of kidney damage can be more severe in IOCM. However, there are still unsolved questions and con-
patients with pre-existing chronic renal insufficiency.6–9 troversies on their hazardness on kidney.13–18
Osmolality and viscosity are two crucial physicochemi- Early in 2003, the NEPHRIC study demonstrated
cal properties of iodinated CMs that contribute to the pro- that IOCM can produce lower incidence of CI-AKI than
gression of CI-AKI.10,11 The two general types of CMs are LOCM in risk patients19; however, this is still debated more
widely used in current clinical practice are: (i) low-viscous, than a decade later.20–22 Theoretically, IOCM has low
Received May 4, 2016, Accepted for publication May 31, 2016.
*Address reprint requests to: Y.-D.Z., Department of Radiology, the First Affiliated Hospital with Nanjing Medical University 300, Guangzhou Road, Nanjing,
Jiangsu Province, China, 210009. E-mail: njmu_wcj@163.com
From the 1Department of Radiology, the First Affiliated Hospital with Nanjing Medical University, Nanjing, China; and 2Department of Pathology, the First
Affiliated Hospital with Nanjing Medical University, Nanjing, China.

V 291

TABLE 1. Viscosity and osmolality of different solutions of different concentration of CM
Brand name Compound Concentration Osmolality Viscosity

(mg iodine/mL) (mOsm/kg at 37 8C)a (mPa.s at 37 8C)a
Visipaque 320 Iodixanol 320 290 11.8

Visipaque 270 Iodixanol 270 290 6.3
Ultravist 370 Iopromide 370 779 9.5
a
Data from manufacturer’s sheets.
osmolality that is approximately equal to the blood plasma, rats were divided in four groups (see Table 1): (i) iodixanol-320 T1
leading to lower cytotoxicity. However, its viscosity is as (Iod.320, 320 mgI/mL, GE Healthcare), (ii) iodixanol-270
two-fold high as that of LOCM, which has been recognized (Iod.270, 270 mgI/mL, GE Healthcare), (iii) iopromide-370
as a more adverse factor than osmolality for renal safety. (Iop.370, 370 mgI/mL, Bayer Healthcare), and (iv) control (0.9%
Preclinical studies have demonstrated that IOCM featured saline, 3 mL per rat).
with high viscosity can produce a series of renal disturbances The animals were kept under standard laboratory conditions
at a temperature of 22 8C with a dark/light rhythm of 12 h. The
such as medullary hypoperfusion, hypoxia and decreased
animals were fed with standard food and water until 2 h before the
vasa recta function.16,23,24 Evidence from clinical trials sug-
experiment. No special hydration protocol was applied. During the
gests that patients who received IOCM experienced clini- imaging procedures, the animals were anesthetized with 3% pento-
cally relevant renal failure two- to three-fold as often as barbital sodium (2 mL/kg body mass isofluran inhalation) by
patients who received LOCM.25 Likewise, another registry means of intraperitoneal injection.
study in 58,957 patients found CI-AKI incidence signifi-
cantly higher following the IOCM versus LOCM.26 The MR Protocols
reason for more risk of CI-AKI produced by IOCM was All examinations were conducted on a 3.0 Tesla (T) MR scanner
ascribed to the fact that many patients in these groups were with a dedicated eight-channel small animal coil (Verio Tim; Sie-
not sufficiently hydrated. However, the underlying relation- mens, Erlangen, Germany). The coil was set with a frequency of
ship of CM viscosity and hydration with the renal safety is 123.3 MHz, and an automatically adjusted fast Fourier transform
scale factor, pulse amplitude and gain factor. Following the acquisi-
somewhat complex and as of yet, unclear.
tion of scout images, the T2-weighted, single-shot fast spin-echo
In everyday clinical practice, it is difficult to determine
images of the kidneys were obtained in the axial and coronal
differences in fluid administration in an individual patient. planes, with the following parameters: repetition time (TR) ms/
Second, because the heterogeneity of experimental design in echo time (TE) ms, 3500/105; section thickness, 3 mm; field of
different trials, there are many different CM types, dosages, view (FOV), 12 cm; and matrix, 256 3 224. On the basis of these
methods of injection. Additionally, in some studies CMs images, a three-dimensional (3D), axial spoiled gradient-recalled
were preheated, and not all studies used preventive hydra- echo protocol encompassing both kidneys and descending aortas
tion, etc.27 Moreover, the outcome measure in most studies was prescribed to acquire data for subsequent baseline T1 estimates
is serum creatinine concentration, which is an insensitive using a metric of driven equilibrium single pulse observation of T1
and indirect measure of kidney function.27–29 (DESPOT1).30 This protocol was performed with the following
Therefore, this animal experience was purposed to parameters: TR/TE, 3.1/0.9; flip angle, 3 8 and 15 8; number of
investigate the acute kidney response to the exposure of sections, 6; section thickness, 3 mm; FOV, 12 cm; and matrix,
128 3 160.
three-type CMs featured with different physicochemical
A novel readout segmentation of long variable echo-trains
properties: low-viscous LOCM (iopromide, 370 mgI/mL),
(RESOLVE) sequence was prescribed to acquire axial DWI of the
high-viscous IOCM (iodixanol, 320 mgI/mL), and a low-
kidney. Diffusion was measured by using b values of 0, 50, and
viscous IOCM (iodixanol, 270 mgI/mL), by expending a 800 s/mm2; TR/TE, 2000/54; flip angle, 90 8; number of readout
physiologic MRI and histologic-gene examination. segments, 5; section thickness, 3 mm; number of sections, 13;
FOV/matrix, 12.8 3 12.8 cm2/ 128 3 128, producing a
Materials and Methods 1.0 3 1.0 mm in-plane resolution. DCE-MRI sequence involved a
Animals and Experimental Protocols 3D nonselective saturation-recovery prepulse followed by a centri-
This study was approved by Local Animal Care and Use Commit- cally ordered fast gradient-echo readout with parameters: TR/TE,
tee. The experiments were performed in male 3- to 4-month-old 428.1/1.98 ms; inversion time, 300 ms; flip angle, 12 8; slice thick-
Sprague-Dawley rats (200–300 g; average, 250 g; Nanjing Medical ness, 3 mm; number of sections, 6; FOV, 12 cm; matrix,
University Laboratory Animal Center, Nanjing, China). Forty-eight 128 3 160, and parallel imaging GRAPPA acceleration factor, 2.

Wu et al.: Functional MRI in Acute Kidney Injury
C
O
L
O
R
FIGURE 1: Overall experimental diagram showing the timing of image acquisitions before and after administration of iodinated
contrast media (CM) or saline (central yellow bar). High-resolution DWI (blue bars) is performed 24 h before, 5 min, 10 min, and
15 min after administration of high-dose iodinated CMs or saline. Dynamic contrast-enhanced MRI (green bars) is performed 24-h
before, and approximately 20 min after the CM administration. Rats are killed 24h after the imaging experiment to undergo kid-
ney histopathologic and gene examination. Gd-DTPA 5 gadolinium-diethylenetriamine penta-acetic acid; Kim-1 5 kidney-induced
molecule 1.
The temporal resolution of DCE-MR sequence was 1.72 s for each MR Image Postprocessing
set of six sections. Ten image sets were acquired before the contrast All data were transferred in Digital Imaging and Communications
agent injection to obtain an accurate baseline measurement. DCE- in Medicine format and postprocessed offline with in-house soft-
MRI involved the intravenous injection of 0.025 mmol/kg (proxi- ware (FireVoxel; Center for Advanced Imaging Innovation and
mate 0.5 mL/rat) of gadolinium-diethylenetriamine penta-acetic Research [CAI2R], New York University School of Medicine, New
acid (Gd-DTPA, Magnevist, Schering AG, Berlin, Germany). The York, NY). Two fellowship-trained radiologists (C.W., 3 years of
contrast was hand-injected as a bolus at an approximate speed of experience in interpreting cross-sectional images and Y.Z., 6 years
1.0 mL/s followed by an equal volume of saline. The DCE imaging of postfellowship experience in interpreting abdominal cross-
was continued for 4.3 min after the contrast agent injection, yield- sectional images) were recruited for image postprocessing: C.W.,
ing a total of 160 serial datasets. for DWI and Y.D., for DCE imaging data, respectively.
DWI and DCE-MRI was performed at 24 h before the The renal ADC maps were constructed with by fitting the
administration of high-dose iodinated CMs to determine a pre- signal intensity values to a mono-exponential model. For regions of
CM renal function. The iodinated CMs were kept at a preheating interest (ROIs), first, whole-kidney ADC histogram distribution
temperature (37 8C) and administered into the tail vein as a single was determined by performing a voxel analysis of each kidney. The
dosage of 4.0 gI/kg (and 3 mL 0.9% saline for control group) histogram parameters included: (a) mean; (b) median; (c) entropy;
within 30–40 s by using a 24 G catheter. Serial post-CM DW (d) kurtosis; (e) skewness; (f ) an inhomogeneity index (INH),
images were obtained at 5 min, 10 min, and 15 min after the CM which is the standard deviation divided by the average signal inten-
and saline administration. Post-CM DCE-MRI was examined sity across all voxels. Second, to assess the ADC difference between
immediately after the post-CM DWI (approximately 20 min after kidney cortex and medulla, ROIs encompassing each kidney were
F1 CM injection). Our detailed protocols are listed in Figure 1. automatically segmented into the region of cortex (CO), outer
C
O
L
O
R
FIGURE 2: MR images of a representative case to indicate the image fusion and regions of interest (ROIs) segmentation. First,
arterial phase of dynamic contrast enhanced (DCE) images is identified by finding the peak of time-dependent curve (a; white
arrow), this work can be nicely completed by in-house Firevoxel software. Then the arterial contrast-enhanced image was fused to
the diffusion-weighted (DW) image (b) using a DICOM-tags technique. The DWI using RESOLVE sequence allows to produce minor
geometry deformation, thus provides excellent spatial registration between T1-weighted and DW images. The signal difference in
kidney cortex (CO), outer medulla (OM) and inner medulla (IM) can be well identified on the fused image (c), which is to facilitate
the ROI-drawing.
January 2017 293

C
O
L
O
R
FIGURE 3: Dynamic contrast-enhanced (DCE) MRI of a representative case to indicate the image processing procedure. a: Vector
region of interest (ROI) was sampled in the abdominal aorta (green) and the whole kidney (red) to extract arterial input function
(AIF) and single kidney time activity curve (TAC), respectively. b: A dynamic multi-agent clustering (DMAC) was performed to
extract voxel-wised TACs of the kidney. c: Regions of cortex (CO) and outer medulla (OM) were then semiautomatically selected
from the collecting system, renal pelvis, and surrounding tissue. And regions with motion artifact were removed. d: Single kidney
GFR was estimated from the motion-free kidney TACs using a two-compartment plot. e: pixel-wised perfusion maps (RBF, RBV,
and mean transmit time [MTT]) were constructed by expending a deconvolution algorithm.
medulla (OM), and inner medulla (IM), respectively, based on a plasma volume; K (GFR) is the extraction-flow product by glomeru-
fused contrast-enhanced T1 image and DWI slice-by-slice. Image lar filtration membranes, and Kt is the tubular outflow rate. The
fusion was performed by expending a DICOM-tags technique pro- component in Eq. [2] expresses the tracer retention in kidney by a
F2 vided by FireVoxel software (Fig. 2). convolution of Ca (t) as input with a retention function, R (t). F is
The tracer-kinetic model for calculating renal hemodynamics the plasma blood flow, and * denotes convolution. V in Eq. [3] is
from DCE-MRI is based on a Toft two-compartment (2C) phar- the plasma blood volume. K and F was normalized as a unit of mL/
macokinetic model.31 This model describes the kidney as a combi- g/min, and V was normalized as a unit of mL/g. Here, we used a
nation of a prefiltration (intravascular) compartment measured by constant kidney density of 1 g/mL and Hct of 0.42 for all subjects.
sampling the arterial signal, and postfiltration (tubular) compart- Processing of DCE-MR images began by coregistration of
ment. Split kidney glomerular filtration rate (GFR) is identified serial images using a mutual information algorithm provided by
with the flow from the prefiltration to postfiltration compartments. FireVoxel software. Kidney parenchyma (excluding the collecting
Because an earlier study32 suggested that Toft 2C-derived Va or system and renal pelvis; median volume of 0.245 cm3, range of
Kep (here is Kt) are not as effective as Ktrans (here is GFR), there- 0.164–0.386 cm3) was then segmented from the surrounding tis-
fore, we used a deconvolution algorithm to assess renal blood flow sue. Aortic signal was sampled in the abdominal aorta at the level
(RBF) and renal blood volume (RBV). The complete equations of of renal arteries to extract the arterial input function. All input
Toft 2C model and deconvolution algorithm are described in detail function ROIs had a standardized size (1 voxel) and were defined
in Eqs. [1-3]: at the height of the exit of the aorta. The model was fitted to renal
time activity curves (TAC) by nonlinear least squares Levenberg-
ðt
Marquardt algorithm. The goodness of fit was assessed by the
CðtÞ5VaCaðtÞ1K CaðtÞe2Kt dt (1)
0
residual (root-mean-square error) and expressed as percentage of
the mean measured concentration (Fig. 3). F3
CðtÞ5FCaðtÞ RðtÞ (2) Lastly, the kidney boundary was manually outlined on a
Ðt high-resolution, 3D T1 weighted imaging slice-by-slice, then the
CðtÞdt in-house software automatically measured the whole volume of the
V5 Ð t0 ; (3)
outlined kidney. Therefore, total 13 imaging parameters including
0 CaðtÞdt
whole-kidney ADC histogram distribution (mean, median, entropy,
where C ðtÞ is the number of moles of contrast agent (CA) in renal kurtosis, skewness, and INH), segmented kidney ADCs (ADCCO,
parenchyma divided by the total tissue volume as a function of ADCOM, and ADCIM), SKGFR, RBF, RBV, and kidney volume
time; Ca ðtÞ is the plasma concentration of the CA; Va is the were calculated from each rat kidney.

TABLE 2. Baseline Renal ADCs and Hemodynamic Characteristicsa
Group Whole ADCs ADCCO ADCOM ADCIM GFR RBF RBV
Iod.320 1.86 6 0.11 1.81 6 0.13 1.79 6 0.11 1.93 6 0.21 0.59 6 0.13 1.67 6 0.17 0.59 6 0.06
Iod.270 1.83 6 0.15 1.80 6 0.16 1.76 6 0.14 1.92 6 0.20 0.62 6 0.14 1.70 6 0.19 0.60 6 0.05
Iop.370 1.85 6 0.18 1.83 6 0.15 1.85 6 0.21 1.86 6 0.21 0.60 6 0.15 1.68 6 0.18 0.56 6 0.07
Saline 1.80 6 0.15 1.78 6 0.13 1.76 6 0.12 1.82 6 0.16 0.58 6 0.14 1.67 6 0.18 0.53 6 0.07
Total 1.82 6 0.13 1.80 6 0.14 1.79 6 0.16 1.89 6 0.21 0.60 6 0.14 1.68 6 0.18 0.57 6 0.07
a 23 2
ADCs is united with 3 10 mm /s, GFR and RBF are united with mL/g/min, and RBV is united with mL/g.
CM 5 contrast media; CO 5 cortex; OM 5 outer medulla; IM 5 inner medulla; ADC 5 apparent diffusion coefficient; GFR 5
glomerular filtration rate; RBF 5 renal blood flow; RBV 5renal blood volume.
Histopathologic and Gene Examination Data were excluded because of: (i) Notable susceptibil-
The animals were killed 24 h after the MR examination. Kidney ity artifact from adjacent bowel on DWI (two kidneys from
tissues were fixed in 10% neutral buffered formalin and embedded one rat in Iod.320 group, two kidneys from one rat in
in paraffin for light microscopic study. Sections of 2-mm thickness Iop.370 group, and two kidneys from one rat in saline
were stained with hematoxylin-eosin, periodic acid-Schiff reagent, group); (ii) Prominent motion artifact on DCE-MR imag-
or phosphotungstic acid-hematoxylin. The tissue was reviewed
ing (one kidney from one rat in Iod.270 group, one kidney
independently by one pathologist (M.L.B.) with more than 5 years
from one rat in saline group).
of experience in urological pathologic analysis. For gene expression
analysis, kidney tissue Kim-1 levels were determined in duplicate
Baseline ADCs and quantitation of DCE-MR renogra-
by solid phase enzyme-linked immunosorbent assay techniques phy is shown in Table 2. Two-way ANOVA showed signifi- T2
(Westang Bio-Tech Co., Ltd., Shanghai, China). cant main effect of kidney region (CO/IM/OM; F 5 12.07;
P < 0.001) and no effect for experimental group (iod.320,
Statistical Analysis iod.270, iop.370 and saline; F 5 2.57; P 5 0.073). There
All statistical analyses were performed by SPSS 16.0 software pack- was no region/group interaction effect for ADCs (F 5 1.31;
age (SPSS, Chicago, Ill). The observations were the histogram and P 5 0.292). Bonferroni test showed ADCIM was signifi-
segmented renal ADCs, GFR, RBF, RBV and post-CM change
cantly higher than ADCCO (P < 0.001) and ADCOM
percent (䉭ADC, 䉭GFR, 䉭RBF, and 䉭RBV). Change percent,
(P < 0.001), there was no difference between ADCCO and
e.g., for ADC, can be expressed as:
ADCOM (P 5 0.574). It showed that pre-CM GFR
ADCpost 2ADCpre (F 5 0.480; P 5 0.697), RBF (F 5 0.205; P 5 0.892) and
䉭ADC5 (2) RBV (F 5 0.345; P 5 0.769) was not significantly different
ADCpre
between four experimental groups.
Statistical comparisons of pre-CM observations among the groups
and kidney regions (unpaired data) were made with two-way analy- Renal ADCs
sis of variance (AVOVA) followed by Bonferroni test for multiple Whole-kidney histogram analysis showed that Iop.370
parametric comparisons. Comparisons of unpaired data (䉭ADC%, increased histogram mean and median ADCs by
䉭GFR, 䉭RBF, and 䉭RBV) among the groups was made by one- 8.7 6 3.1% and 8.3 6 3.4%, respectively, 15 min after the
way ANOVA and Bonferroni test. Comparisons within a given CM-administration (both P values < 0.001). Iod.320
group (paired data; used to compare serial ADCs before and after decreased histogram mean and median ADCs by
CM was given) were performed with the Friedman test (repeated
8.9 6 3.7% and 9.4 6 4.1% (P < 0.001). Iod.270 decreased
measurement analysis of variance on ranks). Comparisons of GFR,
histogram mean and median ADCs by 13.7 6 4.8% and
RBF, and RBV within a given group before and after CM adminis-
tration were made using paired Student t-test. Relationships in
13.4 6 5.2% (P < 0.001). The histogram entropy, skewness,
䉭GFR, 䉭RBF, and 䉭ADCs were determined by Pearson’s correla- kurtosis, and INH did not reflect significant difference
tion analysis. Differences with P values of less than 0.05 were con- between pre- and post-CM in group of Iop.370, Iod.320,
sidered significant. and Iod.270 (results were not shown). However, the pre-
versus post-CM histogram distribution shows remarkable
RESULTS differences in group of Iop.370, Iod.320, and Iod.270 (Fig.
Baseline Characteristics 4). All histogram parameters were not significantly changed F4
A total of 90 kidneys from 45 rats were included in DWI by the administration of saline (results were not shown).
analysis and a total of 92 kidneys from 46 rats in DCE- The time-course change in ADCCO, ADCOM and
MRI analysis. ADCIM responding to the CM-exposure is shown in Figure
January 2017 295

C
O
L
O
R
FIGURE 4: Comparison of whole-kidney ADC histogram distribution in two rat kidneys pre- versus postadministration of Iop.370
(a) and Iod.320 (b). The baseline kidney ADC histogram distribution shifts significantly to the right axis by Iop.370 (a), while shifts
significantly to the left axis by Iod.320 (b), suggesting significantly reverse hemodynamic effects of two CMs on renal function.
The arrowhead may indicate the histogram distribution of kidney IM, which shows difference between post-Iop.370 and post-
Iod.320.
5. It shows that ADCCO was mildly increased by Iop.370 and 15-min (Iod.320: 28.4 6 9.3%; Iod.270: 30.3 6 6.3%; F5
(post-15min, 5.9 6 8.8%, P 5 0.015), mildly decreased by P < 0.001) after the CM-administration (Fig. 6). F6
Iod.320 (post-5min: 3.1 6 7.1%; P 5 0.008), while was not
significantly changed by Iod.270 and saline (Friedman test, Renal Hemodynamics
P > 0.05) after the CM-administration. ADCOM was mildly Kidney volume was significantly increased by the exposure
increased by Iop.370 (post-15min, 4.9 6 10.5%, P 5 0.043) of Iod.320 (pre-CM: 1.03 6 0.26 cm3 versus post-CM:
and saline (post-15min, 4.1 6 9.8%, P 5 0.047), while was 1.54 6 0.30 cm3; P < 0.001) and Iod.270 (pre-CM: 1.13 6
significantly decreased by Iod.320 and Iod.270 at 5-min 0.17 cm3 versus post-CM: 1.43 6 0.19 cm3; P < 0.001; Fig.
(Iod.320: 7.6 6 7.1%; Iod.270: 10.4 6 7.2%; P < 0.001), 7). Kidney volume did not reflect significant difference F7
10-min (Iod.320: 14.1 6 6.1%; Iod.270: 13.3 6 6.6%; between pre- and postexposure of Iop.370 (pre-CM: 1.09 6
P < 0.001), and 15-min (Iod.320: 16.8 6 7.5%; Iod.270: 0.14 cm3 versus post-CM: 1.25 6 0.17 cm3; P 5 0.081) and
18.1 6 9.5%; P < 0.001) after the CM-administration. saline (1.03 6 0.11 cm3 versus post-CM: 1.13 6 0.12 cm3;
ADCIM was serially increased by Iop.370 at 10-min P 5 0.083).
(4.9 6 8.1%; P 5 0.035) and 15-min (12.3 6 11.1%; GFR, RBF, and RBV were significantly decreased by
P < 0.001) after the CM-administration, while was serially Iod.320 (䉭GFR: 45.5 6 24.1%, P 5 0.002; 䉭RBF:
decreased by Iod.320 and Iod.270 at 5-min (Iod.320: 44.6 6 19.0%, P < 0.001; 䉭RBV: 35.2 6 10.1%,
18.3 6 8.4%; Iod.270: 17.7 6 6.8%; P < 0.001), 10-min P < 0.001) and Iod.270 (䉭GFR: 33.2 6 19.0%, P < 0.001;
(Iod.320: 21.2 6 10.2%; Iod.270: 26.4 6 8.7%; P < 0.001), 䉭RBF: 38.1 6 15.6%, P < 0.001; 䉭RBV: 30.1 6 10.1%,
C
O
L
O
R
FIGURE 5: the time-course change in ADCCO (a), ADCOM (b), and ADCIM (c) caused by iod.320, iod.270, iop.370 and saline, respec-
tively. Note. *P < 0.05 versus pre-CM ADC; **P < 0.001 versus pre-CM ADCs.

C
O
L
O
R
FIGURE 6: Comparison of pixel-wised pre-CM versus post-CM ADCs between Iod.320 and Iop.370: the pre-CM ADC maps (a–a0 ),
the corresponding post-CM ADC maps (b–b0 ), the subtraction of a–b (c), and the inverse subtraction of (a0 –b0 ) (c0 ). It shows that
Iod.320 and Iop.370 have reverse hemodynamic effects on renal ADCs, where Iod.320 remarkably decreases the renal ADCs,
while Iop.370 remarkably increases the renal ADCs, especially in the regions of inner medullary kidney (arrow).
P < 0.001), while were rarely changed by Iop.370 (䉭GFR: glomeruli and tubules in all groups were fairly intact. No
2.4 6 24.3%, P 5 0.236; 䉭RBF: 2.6 6 19.2%, P 5 0.314; necrosis or apoptosis was detected in kidney tubules. How-
䉭RBV: 1.4 6 12.2%, P 5 0.317; Fig. 7) and saline ever, formation of vacuoles in proximal tubular cells was
(䉭GFR: -2.9 6 10.7%, P 5 0.176; 䉭RBF: -1.5 6 9.2%, prominent in group of Iod.320, slightly less following the
P 5 0.227; 䉭RBF: -2.3 6 10.9%, P 5 0.325). Iod.270, while sparse after Iop.370 and saline (arrows in
F8 The scatterplots (Fig. 8) and Pearson correlation test show Fig. 9a). In the rats treated with Iod.320, we observed a F9
that 䉭GFR is highly correlated with 䉭RBF (Pearson correla- substantially increased number of Kim-1/Havcr1 transcripts
tion coefficient r 5 0.842; P < 0.001), moderately correlated (5.62 6 1.68 ng/mL) compared with the average number of
with 䉭ADCIM (r 5 0.668; P < 0.001) and 䉭ADCOM transcripts observed in the saline control (2.25 6 0.57 ng/
(r 5 0.545; P 5 0.001), while rarely correlated with 䉭ADCOM mL; P < 0.001), Iod.270 (2.03 6 0.41 ng/mL, P < 0.001),
(r 5 0.234; P > 0.05). relations of 䉭GFR versus 䉭RBF and Iop.370 (2.97 6 0.32, P < 0.001), respectively. In the
animals treated with Iod.270 and Iop.370, no significant
Histology and Kim-1 Gene Expression changes in Kim-1/Havcr1 mRNA expression compared with
As compared with the kidney in controls, the kidneys in the control groups (P > 0.05) were found (Fig. 9b). Thus
group of Iod.320 and Iod.270 had predominantly enlarged Iod.320 significantly induced molecular marker of kidney
volume. Darkened incisal surface of the specimens was injury, whereas Iop.370 and Iod.270 had no significant
detected. Light microscopic examinations showed that the effect on this parameter.
January 2017 297

C
O
L
O
R
FIGURE 7: a–b0 : Comparison of typical kidney morphology and hemodynamics observed in two individual animals between
Iod.320 and Iop.370. The illustration includes high-resolution structural image (vol.), ROI-based GFR and pixel-wised RBF map. The
structural image reflect volume change of the kidney. The ROI-based time activity curves (TACs) reflect hemodynamic function of
the kidney, which is characteristic with a vascular peak (arrow 1), an equilibrium phase, and a slow excretory phase. The colors on
RBF maps represent the relative levels of intrarenal perfusion. It shows that kidney volume (vol.) in post-Iob.320 is significantly
enlarged. The TACs in post Iod.320 is featured with a flat vascular peak (arrow 2) and delayed tracer wash-out, suggesting a
reduced renal function. For this case, left single kidney GFR and RBF are reduced by approximate 50%. In the second case
injected with Iop.370, the post-CM kidney volume and TACs are rarely changed, and GFR and RBF is not significantly reduced due
to the CM-administration.
Discussion Iop.370, even given in high dose, rarely took adverse effect
In this study, we used DWI and DCE-MR renography to on renal function or rarely resulted in acute injury in the
assess acute renal response following experimental perturba- kidney tubule cells. Conversely, IOCMs, Iod.320, and
tions of kidney function using three CMs regarding differ- Iod.270, both produced significant decrease in single-kidney
ent osmolality and viscosity. We observed that LOCM GFR, renal perfusion and medullary ADCs. Additionally,
C
O
L
O
R
FIGURE 8: Scatterplots and relations of 䉭GFR versus 䉭RBF (a), 䉭GFR versus 䉭ADCIM (b) and 䉭GFR versus 䉭ADCOM (c), respec-
tively. Note. *P < 0.05 by Pearson test. 䉭GFR is plotted in Y axial, 䉭RBF, 䉭ADCIM, and 䉭ADCOM is plotted in X axial, respectively.

C
O
L
O
R
FIGURE 9: Exemplary hematoxylin-eosin staining (a) and renal tissue expression of Kim-1 (b) to assess formation of vacuoles and
acute kidney injury in proximal tubular cells, all taken 24 h after injection of either Iod.320, Iod.270, Iop.370, or saline. Formation
of vacuoles (yellow arrows) in proximal tubular cells was prominent for the Iod.320, slightly less following the Iod.270, while
sparse after Iop.370 and saline (not illustrated in this figure). Kim-1 level was significantly increased in group of Iod.320
(**P < 0.001) compared with the transcripts observed in the saline, Iod.270 and Iop.370, respectively. In the animals treated with
Iod.270 and Iop.370, no significant changes in Kim-1/Havcr1 mRNA expression compared with the control groups (P > 0.05) were
found.
the Iod.320 significantly induced molecular markers (Kim- leading to adverse effect on renal function. Also Jost et al
1) of kidney injury, whereas Iop.370 and Iod.270 had no monitored a strong decrease in renal ADCs by Iod.320,
significant effect on this parameter. Therefore, we concluded while a mild increase in renal ADCs by iopromide (300
that, the LOCM and IOCM, featured with different physi- mgI/mL).33 Recently longer time-course (0–72 h or 0–18
cochemical properties, have reverse physiopathological days) in vivo studies by Zhang et al41 and Wang et al,42
effects on kidney function, which could be noninvasively respectively, reported an administration of high-dose LOCM
determined by physiologic MRI. Furthermore, we make the or IOCM can produce successively reduction in intrarenal
hypothesis that IOCM-produced acute kidney injury might ADCs, furthermore, IOCM had more adverse effects on
be caused by renal hypoperfusion or increased tubular fluid kidney function than LOCM. This phenomenon is likely
viscosity. explained as that Iop.370 has remarkably high osmolality
Jost et al33 initially used a single-shot EPI DWI to (774 mOsm/kg H2O, manufacturer’s specification), produc-
investigate the intrarenal CM viscosity during their excretion ing osmotic diuresis that prevents excessive tubular viscosity
because of the exponential viscosity-ADC relationship. They levels. This mechanism has been comprehensively investi-
reported a prominent decrease in whole-kidney ADCs gated by previous vitro and in vivo studies.14,16,24
caused by IOCM. Similar results were observed in this DCE-MRI with optimal pharmacokinetic model has
study by expending a novel RESOLVE DWI, which has been reported as an effective way to estimate the split renal
shown previously better ability in improvement of image function.31,43,44 As reported previously by Seeliger et al,14,16
quality than traditional ss-EPI.34–40 Benefited from these the vascular response to CM can be different between
individual features, signal difference between the kidney cor- LOCM and IOCM: High-osmolality iopromide transiently
tex and medulla can be well monitored: we found that increased renal blood flow, markedly increased urine pro-
ADCCO was rarely affected by the administration of any duction, and rarely affected GFR. However, high-viscosity
CM, while the ADC in medullary kidney was significantly iodixanol predominantly decreased the renal medullary
changed by means of administration of different CMs. blood flux, significantly enhanced urine viscosity, and transi-
Additionally, the LOCM and IOCM had significantly ently decreased GFR. Partly in agreement with their studies,
different physiologic effects on medullary ADCs: in group we observed LOCM Iop.370, even given in high dose,
of Iop.370, LOCM produced mild increase in ADCOM and rarely took adverse effect on RBF and GFR, and rarely
ADCIM, suggesting its harmless potential; while in group of resulted in histologic-gene expression level. Conversely,
Iod.320 and Iod.270, IOCM produced prominent decrease IOCMs Iod.320 and Iod.270 significantly decreased RBF
in ADCOM and ADCIM, which is associated with a physio- and GFR, and increased the Kim-1 gene expression. The
logically increased intrarenal fluid viscosity, thus potentially reason for GFR reduction with administration of IOCM
January 2017 299

was previously explained as an increased tubular fluid Our study had several limitations. First, this is initial
viscosity. result for an animal experiment with relatively small sample
However, the viscosity difference between Iod.320 and numbers. The dose of CMs was higher than that used in
Iop.370 is only approximately 20% (11.8 versus 9.5 mPa.s) percutaneous cardiac interventions or contrast-enhanced CT
and the viscosity of Iod.270 is even less than Iop.370. The angiography. It remains to be seen whether differences
LOCM featured with high osmolality causing an osmotic would be significant for more clinical relevant dose. Addi-
diuresis can be an important contributor for less GFR tionally, the pathophysiologic response of the kidney to the
reduction.14,16 Conversely, high viscosity in IOCM is identi- two type CMs under nonhydrated versus hydrated state is
fied as adverse factor for CIAKI, which is associated with yet unknown. This would be interesting as a future direc-
consecutive renal disturbances, i.e., reducing in medullary tion. Second, the measure in renal ADCs may only partly
descending vasa recta luminal diameter, increasing intra- reflect the intrarenal viscosity of CM future more precise
tubular pressure, and prolonging tubuloglomerular feedback measure in intrarenal viscosity is needed. Third, we used
response.16,23,24 Furthermore, we monitored close relation- gadolinium-based CM, which is potentially nephrotoxic.
ship between the dropped GFR and renal hypoperfusion. However, our saline group serves as control to correct for
The hypoperfusion can reduce directly the preglomerular the potential confounding results. Last, the imaging analysis
arterial pressure, and second, sharpen the oxygen burden of was performed on a human but not a small animal MR
Henle’s thick ascending limbs, as high oxygen requirements scanner, even considerable effort was devoted, uncertainty
while weak oxygen delivery in this region.16,41 still cannot be ruled out in some cases.
The drop in GFR was observed with moderate rela- In conclusion, the primary purpose of this study is to
tionship with decrease in medullary ADCs. The increase in compare the impact of three iodinated CMs with different
medullary ADCs may, first, indicate high tubular fluid vis- physicochemical properties on renal function using DCE-
cosity, and second, indicate a cytotoxic edema of the MR renography and DWI. The high-viscous, iso-osmolar
involved kidney tissue.42 High tubular fluid viscosity caused iodixanol (Iod.320), given in high dose and without suffi-
by IOCM produces high postglomerular tubular pressure, cient hydration, can cause predominant renal dysfunction
which contributes directly to the drop in GFR. Also it can and acute kidney injury as compared with the low-viscous,
prolong the retention of iodine ions in renal tubules, thus low-osmolar iopromide (Iop.370). The mechanism of acute
increasing the nephrotoxicity potentially. Our hypothesis kidney response is probably associated with increased med-
can be nicely confirmed by post-CM histopathologic and ullary viscosity and hypoperfusion. Second, we monitored
gene examination as the involved kidneys had more nephric Iod.270 did not alleviate the extent of adverse renal response
swelling, significantly higher expression of Kim-1 than these than Iod.320 and Iop.370. This finding may have clinical
of LOCM. implications, while the underlying mechanism needs further
Iod.270 was commercially designed to reduce the high investigation.
viscosity and to alleviate the underlying renal injury. Theo-
retically, Iod.270 has lower osmolality than Iop.370 and
lower viscosity than Iod.320, and should produce less renal Acknowledgments
injury. Even the Kim-1 gene expression of the kidney tissue
Contract grant sponsor: National Natural Science Founda-
was not significantly increased by Iod.270, the extent of
tion of China; contract grant number: 81301191; Contract
acute kidney physiologic response evaluated by MRI was grant sponsor: the Priority Academic Program Develop-
not significantly reduced as compared with Iod.320 or ment of Jiangsu Higher Education Institutions; contract
Iop.370. The mechanism for this phenomenon is unclear: it grant number: JX10231801.
is probably associated with lacking osmotic diuresis or We thank Prof. Henry Rusinek (Department of Radiology,
caused by a delayed excretion of iodixanol. This is true as New York University School of Medicine, New York, NY)
both dimeric iodixanol and monomeric iopromide are for language assistance and in-house software assistance.
excrete almost exclusively by means of kidney and are hav-
ing similar plasma half-live times. However, lower iodine
concentration in the kidney was observed after administra-
tion of the monomeric iopromide against the dimeric iodix- References
anol in several preclinical experiments,11,45 indicating a 1. Tepel M. Preventing nephropathy induced by contrast medium. N
Engl J Med 2006;354:1853–1855; author reply 1853–1855.
faster excretion of the monomeric iopromide. Nevertheless,
2. Finn WF. The clinical and renal consequences of contrast-induced
our finding has very important clinical implications: sug- nephropathy. Nephrol Dial Transplant 2006;21:i2–i10.
gesting that Iod.270 may not offer better renal safety than 3. McCullough PA. Contrast-induced acute kidney injury. J Am Coll Car-
either of the other two CMs under nonhydrated condition. diol 2008;51:1419–1428.

4. Bouzas-Mosquera A, Vazquez-Rodriguez JM, Calvino-Santos R, et al. 23. Wang YC, Tang A, Chang D, Zhang SJ, Ju S. Significant perturbation
[Contrast-induced nephropathy and acute renal failure following emer- in renal functional magnetic resonance imaging parameters and con-
gent cardiac catheterization: incidence, risk factors and prognosis]. trast retention for iodixanol compared with iopromide: an experimen-
Rev Esp Cardiol 2007;60:1026–1034. tal study using blood-oxygen-level-dependent/diffusion-weighted
magnetic resonance imaging and computed tomography in rats.
5. Goldenberg I, Matetzky S. Nephropathy induced by contrast media:
Invest Radiol 2014;49:699–706.
pathogenesis, risk factors and preventive strategies. CMAJ 2005;172:
1461–1471. 24. Jost G, Pietsch H, Lengsfeld P, Hutter J, Sieber MA. The impact of
the viscosity and osmolality of iodine contrast agents on renal elimina-
6. James MT, Samuel SM, Manning MA, et al. Contrast-induced acute tion. Invest Radiol 2010;45:255–261.
kidney injury and risk of adverse clinical outcomes after coronary angi-
ography: a systematic review and meta-analysis. Circ Cardiovasc 25. Liss P, Persson PB, Hansell P, Lagerqvist B. Renal failure in 57 925
Interv 2013;6:37–43. patients undergoing coronary procedures using iso-osmolar or low-
osmolar contrast media. Kidney Int 2006;70:1811–1817.
7. Calvin AD, Misra S, Pflueger A. Contrast-induced acute kidney injury
and diabetic nephropathy. Nat Rev Nephrol 2010;6:679–688. 26. Reed MC, Moscucci M, Smith DE, et al. The relative renal safety of
iodixanol and low-osmolar contrast media in patients undergoing per-
8. Briguori C, Donnarumma E, Quintavalle C, Fiore D, Condorelli G. cutaneous coronary intervention. Insights from Blue Cross Blue Shield
Contrast-induced acute kidney injury: potential new strategies. Curr of Michigan Cardiovascular Consortium (BMC2). J Invasive Cardiol
Opin Nephrol Hypertens 2015;24:145–153. 2010;22:467–472.
9. Kim JH, Yang JH, Choi SH, et al. Predictors of outcomes of contrast- 27. Seeliger E, Sendeski M, Rihal CS, Persson PB. Contrast-induced kid-
induced acute kidney injury after percutaneous coronary intervention ney injury: mechanisms, risk factors, and prevention. Eur Heart J
in patients with chronic kidney disease. Am J Cardiol 2014;114:1830– 2012;33:2007–2015.
1835.
28. Solomon R, Dauerman HL. Contrast-induced acute kidney injury. Cir-
10. Pannu N, Wiebe N, Tonelli M. Prophylaxis strategies for contrast- culation 2010;122:2451–2455.
induced nephropathy. JAMA 2006;295:2765–2779.
29. Waikar SS, Bonventre JV. Creatinine kinetics and the definition of
11. Jost G, Pietsch H, Sommer J, et al. Retention of iodine and acute kidney injury. J Am Soc Nephrol 2009;20:672–679.
expression of biomarkers for renal damage in the kidney after applica-
tion of iodinated contrast media in rats. Invest Radiol 2009;44:114– 30. Deoni SC, Rutt BK, Peters TM. Rapid combined T1 and T2 mapping
using gradient recalled acquisition in the steady state. Magn Reson
123.
Med 2003;49:515–526.
12. Singh J, Daftary A. Iodinated contrast media and their adverse reac-
31. Bokacheva L, Rusinek H, Zhang JL, Chen Q, Lee VS. Estimates of glo-
tions. J Nucl Med Technol 2008;36:69–74; quiz 76-77.
merular filtration rate from MR renography and tracer kinetic models.
13. Rudnick MR, Goldfarb S, Wexler L, et al. Nephrotoxicity of ionic and J Magn Reson Imaging 2009;29:371–382.
nonionic contrast media in 1196 patients: a randomized trial. The
32. Oto A, Kayhan A, Jiang Y, et al. Prostate cancer: differentiation of
Iohexol Cooperative Study. Kidney Int 1995;47:254–261.
central gland cancer from benign prostatic hyperplasia by using
14. Seeliger E, Becker K, Ladwig M, Wronski T, Persson PB, Flemming B. diffusion-weighted and dynamic contrast-enhanced MR imaging. Radi-
Up to 50-fold increase in urine viscosity with iso-osmolar contrast ology 2010;257:715–723.
media in the rat. Radiology 2010;256:406–414.
33. Jost G, Lenhard DC, Sieber MA, Lengsfeld P, Hutter J, Pietsch H.
15. Davenport MS, Khalatbari S, Dillman JR, Cohan RH, Caoili EM, Changes of renal water diffusion coefficient after application of iodin-
Ellis JH. Contrast material-induced nephrotoxicity and intravenous ated contrast agents: effect of viscosity. Invest Radiol 2011;46:796–
low-osmolality iodinated contrast material. Radiology 2013;267:94– 800.
105.
34. Robson MD, Anderson AW, Gore JC. Diffusion-weighted multiple
16. Seeliger E, Flemming B, Wronski T, et al. Viscosity of contrast shot echo planar imaging of humans without navigation. Magn Reson
media perturbs renal hemodynamics. J Am Soc Nephrol 2007;18: Med 1997;38:82–88.
2912–2920.
35. Porter DA, Heidemann RM. High resolution diffusion-weighted imag-
17. Heinrich MC, Haberle L, Muller V, Bautz W, Uder M. Nephrotoxicity ing using readout-segmented echo-planar imaging, parallel imaging
of iso-osmolar iodixanol compared with nonionic low-osmolar contrast and a two-dimensional navigator-based reacquisition. Magn Reson
media: meta-analysis of randomized controlled trials. Radiology 2009; Med 2009;62:468–475.
250:68–86. 36. Koyasu S, Iima M, Umeoka S, et al. The clinical utility of reduced-
18. Bruce RJ, Djamali A, Shinki K, Michel SJ, Fine JP, Pozniak MA. Back- distortion readout-segmented echo-planar imaging in the head and
ground fluctuation of kidney function versus contrast-induced nephro- neck region: initial experience. Eur Radiol 2014;24:3088–3096.
toxicity. AJR Am J Roentgenol 2009;192:711–718. 37. Yeom KW, Holdsworth SJ, Van AT, et al. Comparison of readout-
19. Aspelin P, Aubry P, Fransson SG, et al. Nephrotoxic effects in high- segmented echo-planar imaging (EPI) and single-shot EPI in clinical
risk patients undergoing angiography. N Engl J Med 2003;348:491– application of diffusion-weighted imaging of the pediatric brain. AJR
499. Am J Roentgenol 2013;200:W437–W443.
20. Mehran R, Nikolsky E, Kirtane AJ, et al. Ionic low-osmolar versus non- 38. Park JY, Shin HJ, Shin KC, et al. Comparison of readout segmented
ionic iso-osmolar contrast media to obviate worsening nephropathy echo planar imaging (EPI) and EPI with reduced field-of-VIew diffu-
after angioplasty in chronic renal failure patients: the ICON (Ionic version-weighted imaging at 3t in patients with breast cancer. J Magn
sus non-ionic Contrast to Obviate worsening Nephropathy after Reson Imaging 2015;42:1679–1688.
angioplasty in chronic renal failure patients) study. JACC Cardiovasc 39. Medved M, Soylu-Boy FN, Karademir I, et al. High-resolution
interv 2009;2:415–421. diffusion-weighted imaging of the prostate. AJR Am J Roentgenol
2014;203:85–90.
21. McCullough PA, Bertrand ME, Brinker JA, Stacul F. A meta-analysis of
the renal safety of isosmolar iodixanol compared with low-osmolar 40. Tokoro H, Fujinaga Y, Ohya A, et al. Usefulness of free-breathing
contrast media. J Am Coll Cardiol 2006;48:692–699. readout-segmented echo-planar imaging (RESOLVE) for detection of
malignant liver tumors: comparison with single-shot echo-planar imag-
22. Jo SH, Youn TJ, Koo BK, et al. Renal toxicity evaluation and compari-
ing (SS-EPI). Eur J Radiol 2014;83:1728–1733.
son between visipaque (iodixanol) and hexabrix (ioxaglate) in patients
with renal insufficiency undergoing coronary angiography: the 41. Zhang Y, Wang J, Yang X, et al. The serial effect of iodinated contrast
RECOVER study: a randomized controlled trial. J Am Coll Cardiol media on renal hemodynamics and oxygenation as evaluated by ASL
2006;48:924–930. and BOLD MRI. Contrast Media Mol Imaging 2012;7:418–425.
January 2017 301

42. Wang J, Zhang Y, Yang X, et al. Time course study on the effects of 44. Yamamoto A, Zhang JL, Rusinek H, et al. Quantitative evaluation of
iodinated contrast medium on intrarenal water transport function acute renal transplant dysfunction with low-dose three-dimensional
using diffusion-weighted MRI. J Magn Reson Imaging 2012;35:1139– MR renography. Radiology 2011;260:781–789.
1144.
45. Ueda J, Nygren A, Sjoquist M, Jacobsson E, Ulfendahl HR, Araki Y.
43. Zhang YD, Wang J, Zhang J, Wang X, Jiang X. Effect of Iodine concentrations in the rat kidney measured by X-ray microanaly-
iodinated contrast media on renal function evaluated with sis. Comparison of concentrations and viscosities in the proximal
dynamic three-dimensional MR renography. Radiology 2014;270:409– tubules and renal pelvis after intravenous injections of contrast media.
415. Acta Radiol 1998;39:90–95.

ORIGINAL RESEARCH
Persistent T2*-Hypointensity of the Liver

Parenchyma After Irradiation to the SPIO-
Accumulated Liver: An Imaging Marker for
Responses to Radiotherapy in Hepatic
Malignancies
Toshihiro Furuta, MD, PhD,1,2 Masayuki Yamaguchi, MD, PhD,1*
Manabu Minami, MD, PhD,3 Kuni Ohtomo, MD, PhD,2 and
Hirofumi Fujii, MD, PhD1
Purpose: To determine whether T2*-weighted MRI has the ability to visualize the irradiated liver parenchyma and liver
tumor after irradiation to the previously superparamagnetic iron oxide (SPIO)-accumulated liver.
Materials and Methods: We examined 24 liver tumor-bearing rats. Nine rats (Group 1) received 20 mmol Fe/kg SPIO
and subsequent 70-Gy irradiation to the tumor-bearing liver lobe. Four rats (Group 2) received SPIO and sham irradia-
tion. Six rats (Group 3) received saline and irradiation. Finally, five rats (Group 4) received saline and sham irradiation.
We acquired sequential 3 Tesla T2*-weighted images of the liver on day 7, and assessed MR image findings including
signal intensity of the tumors and tumor-bearing liver lobes.
Results: In six rats in Group 1, tumors shrunk by 39–100% (303–0 mm3), and severely, well-defined hypointense irradiat-
ed areas were observed. In the other two rats, tumors enlarged by 25 and 172% (595 and 1148 mm3), and hypointense
rings surrounded the tumors. The normalized relative signal intensity of the irradiated areas was significantly lower than
that of the tumor (0.53 6 0.06 versus 0.94 6 0.06; P < 0.05). The severely, well-defined hypointense areas were not
observed in the other groups. Histologically, necrotic regions dominated and minimal nonnecrotic tumor cells remained
in irradiated areas. The number of CD68-positive cells was higher in irradiated areas than in nonirradiated areas.
Conclusion: T2*-weighted MR imaging visualized the irradiated liver parenchyma as markedly, well-defined hypointense
areas and liver cancer lesions as hyperintense areas only when SPIO was administered before irradiation. The visualiza-
tion of the hypointense area was associated with tumor regression after irradiation.
M ORE THAN 700,000 new liver cancer cases were

reported worldwide in 2012.1 While surgical resection
is the current standard therapy for resectable liver cancer,
or safety margin status of liver resection specimens is an
established, useful indicator for this assessment because a
positive margin is a risk factor for local recurrence.4 By
radiotherapy is a less invasive option in selected patients.2,3 applying this concept to radiotherapy, the safety margin
In surgery and radiotherapy, assessing the risk of local recur- assessment may also provide a reliable indicator of the risk
rence is considered to be very important for predicting the of local recurrence if irradiated liver parenchyma and liver
prognosis of these patients. In surgery, the surgical margin cancer lesions are visualized using in vivo imaging during
Received Feb 8, 2016, Accepted for publication Jun 2, 2016.

**Address reprint requests to: M.Y., Magnetic Resonance Imaging Section, Division of Functional Imaging, Exploratory Oncology Research & Clinical Trial
Center, National Cancer Center, Kashiwanoha 6-5-1, Kashiwa, Chiba 277-8577, Japan. E-mail: masyamag@east.ncc.go.jp
From the 1Division of Functional Imaging, Exploratory Oncology Research & Clinical Trial Center, National Cancer Center, Kashiwa, Japan; 2Department of
Radiology, Graduate School of Medicine, University of Tokyo, Tokyo, Japan; and 3Department of Radiology, Graduate School of Comprehensive Human
Sciences, University of Tsukuba, Tsukuba, Japan

V 303

the treatment period or soon after its completion. While Animal Model
radiation oncologists use sophisticated computer programs Animal experiments were conducted in compliance with the proto-
to optimally define the planning target volume and secure col approved by the Institutional Animal Experimental Committee.
the margin2, such an imaging technique may assist in ascer- Twenty-four female Sprague-Dawley rats (8 weeks of age) were
taining the same irradiated areas as planned. purchased from Japan SLC, Inc. (Hamamatsu, Japan), and housed
There is an emerging need to explore this technique two or three per cage under controlled lightning conditions (12-h
light/dark cycles). Room temperature and humidity were main-
especially in stereotactic body radiotherapy (SBRT) for liver
tained at 22 6 2 8C and 50%, respectively. After 1 week of acclima-
malignancies because liver motion due to respiration and
tization, 4 3 106 N1-S1 hepatoma cells were inoculated into the
cardiac pulsation may produce mis-registration between the
left liver lobe as described previously.15 This N1-S1 hepatoma
target volumes and actual irradiated areas.3,5,6 A clear and model has been commonly used as a liver cancer animal mod-
simultaneous visualization of the irradiated liver parenchyma el.16–20 One week later, when the hepatoma had reached a maxi-
and cancer lesions remains challenging with currently avail- mum diameter of approximately 10 mm, 24 rats were divided into
able imaging techniques, including contrast-enhanced com- four groups: nine rats that received SPIO and subsequent hepatic
puted tomography (CT) and MR imaging.7–13 Radiation- irradiation (Group 1), four that received SPIO and subsequent
induced changes in liver tissues hinder this visualization, sham irradiation (Group 2), six that received saline and subsequent
particularly in acute and subacute phases after irradiation. hepatic irradiation (Group 3), and five that received saline and
Radiation-induced hyperemia, for instance, has been shown subsequent sham irradiation (Group 4). Their body weights ranged
to frequently produce enhanced areas that are indiscernible between 199 and 238g at that time. We measured the body weight
of rats on day 7. In addition, we carefully monitored adverse reac-
from residual liver cancer lesions.13
tions of irradiation to the gastrointestinal tracts, including gastroin-
Furuta and co-workers recently reported a new MR
testinal bleeding throughout the experimental period.
imaging technique that is potentially capable of visualizing
the irradiated liver parenchyma and cancer lesions irrespective SPIO Administration and MR Examination
of these postradiation changes.14 In this technique, superpara- Schedule
magnetic iron oxide (SPIO) was administered to label Kupffer Each rat in Groups 1 and 2 was intravenously injected with
cells (KCs) in the liver parenchyma before the radiation dose 20 mmol iron/kg body weight of ferucarbotran (Resovist; Bayer
was delivered in this area. The irradiated liver parenchyma Yakuhin, Ltd., Osaka, Japan) at a volume of 0.3 mL via the tail
was visualized as a hypointense area on T2*-weighted MR vein. In Groups 3 and 4, each rat was intravenously injected with
images for at least 7 days because the washout of SPIO was an equivalent volume of saline. MR images of the upper abdomen
were acquired before and 10 min after the administration of SPIO
delayed. In contrast, the nonirradiated liver parenchyma
or saline using the scan protocol described below (see the MR
appeared as a relatively hyperintense area during the period
Imaging section). Additional MR examinations were performed 3
that SPIO was being washed out. Theoretically, irradiated liv- and 7 days after irradiation without the additional administration
er cancer lesions may be visualized as hyperintense areas sur- of SPIO or saline.
rounded by the hypointense irradiated liver parenchyma
because they are devoid of KCs and do not accumulate SPIO. Irradiation
However, it has not yet been established whether cancer Four hours after the administration of SPIO or saline, each rat in
lesions and the surrounding irradiated liver parenchyma are all groups was anesthetized with 1–2% isoflurane, placed in the
contrasted based on differences in the accumulation of SPIO supine position, and secured with adhesive tapes on a cork board.
After a midline abdominal incision was made, the hepatoma-
before the occurrence of radiation-induced changes.
bearing liver lobe was pulled out through the incision and placed
Therefore, the purpose of the present study was to
on a 3-mm-thick sterile lead board (30 3 40 mm2) to avoid the
determine whether T2*-weighted MR imaging has the abili-
organs underneath being exposed to radiation. We then covered
ty to visualize the irradiated liver parenchyma and liver can- the whole body of the rat with a lead half-cylinder (diameter of
cer lesions after irradiation to the previously SPIO- 85 mm, length of 260 mm, and thickness of 3 mm) with a
accumulated liver. 15 3 15 mm2 window and placed the window just above the hepa-
toma (Supporting Fig. S1, Supporting Information 1, which is
Materials and Methods available online and explains the relationships among the irradiated
Tumor Cell Line and Culture field, liver tumor, and surrounding liver parenchyma).
The N1-S1 hepatoma cell line was purchased (CRL-1604; the Each rat in Groups 1 and 3 was placed in a 160-kV x-ray
American Type Culture Collection [ATCC], Manassas, VA) and unit (Model CP-160; Faxitron X-Ray Corp., Wheeling, IL), typi-
cultured in Iscove’s Modified Dulbecco’s Medium (Gibco 12440; cally for 10 min, and treated with 70 Gy irradiation at a dose rate
Life Technologies, Grand Island, NY) supplemented with 10% of 107–121 mGy/s. We applied high dose irradiation to effectively
fetal bovine serum (Gibco 10099; Life Technologies). Cells were suppress tumor growth of highly malignant N1-S1 hepatomas. The
maintained in suspension culture dishes at 37 8C in a humidified x-ray beam was filtered with a 0.1-mm copper layer. This proce-
atmosphere containing 5% CO2 and passaged twice a week. dure enabled a high dose of irradiation to be delivered to the left

Furuta et al.: T2*-Hypointensity in Irradiated Livers
liver lobe while minimizing the radiation dose to the other parts of Supporting Information 3 shows hypointense areas observed on T2*-
the body. Each rat in Groups 2 and 4 was placed into the x-ray weighted images at 9.4T, which help readers to interpret the hypoin-
unit for 10 min without x-ray exposure (sham irradiation). In a tense areas shown on 3.0T MR images. The volume of the hypoin-
preliminary study, we confirmed that this irradiation procedure tense areas was estimated on T2*-weighted images using the following
produced T2*-hypointense areas in the previously SPIO- equation: Volume of hypointense area (mm3) 5 D1 3 D2 3 D3,
accumulated, non–tumor-bearing liver lobes of rats on day 7 (Sup- where D1 and D2 are the maximum orthogonal diameters of the
porting Fig. S2, Supporting Information 2, which explains tempo- hypointense area. D3 is the product of the number of slices in which
ral T2*-weighted signal changes in the irradiated, non–tumor- the hypointense area is observed and 2.4 (i.e., the sum of the slice
bearing liver lobe). thickness and inter-slice gap).
Tumor volumes (mm3) were estimated on T2-weighted
MR Imaging images using the ellipsoid formula (pi/6 3 D1t 3 D2t 3 D3t,
Imaging was performed with a 3 Tesla (T) scanner (Signa HDx; where D1t and D2t are the maximum orthogonal diameters and
GE Healthcare, Milwaukee, WI) and a multi-receiver coil dedicat- D3t is the product of the number of slices on which the tumor is
ed for multiple animal MR imaging.15 In each MR imaging ses- displayed and 2.4 [i.e., the sum of the slice thickness and interslice
sion, four rats were anesthetized using 1–2% isoflurane and a gas gap]). We then calculated % increases in tumor volumes by divid-
mixture of nitrous oxide and oxygen (1:1) through nose masks. ing tumor volumes on days 3 and 7 by those on day 0 and then
They were subsequently placed in either the supine or prone posi- multiplying by 100.
tion on acrylic animal beds, and two of the four rats were placed
on the upper coil frames in a supine position, and the other two Pathological Analysis
on the lower coil frames in a prone position, with their abdominal Each rat was euthanized by CO2 asphyxiation immediately after the
surfaces close to coil elements. Respiratory rates were continuously last MR imaging session. The liver was excised and fixed in 10%
monitored using a monitoring system (Model 1025; SA Instru- neutral-buffered formalin, embedded in paraffin, and cut into thin
ment, Stony Brook, NY). sections. The sections were stained with hematoxylin and eosin
We acquired MR images of the liver using T2*-weighted (H&E). CD68 immunohistochemical staining was also performed.
gradient echo sequence (repetition time [TR] 5 440 ms, echo time The number of CD68-positive cells was counted in five different and
[TE] 5 11 ms, flip angle [FA] 5 30 8, and number of excitations randomly selected 0.1-mm2 areas in the left and right liver lobes
[NEX] 5 4–8), T2-weighted fast-spin-echo (FSE) sequence with using an image analyzer (BZ-9000; Keyence, Osaka, Japan).
periodically rotated overlapping parallel lines with enhanced recon-
struction (PROPELLER) technique (TR 5 4000 ms, TE 5 86 ms,
Tukey’s test or the Tukey-Kramer method was performed to com-
echo train length [ETL] 5 24, FA 5 90 8, and NEX 5 4.5), and
pare the normalized relative intensity and number of CD68-
proton density-weighted FSE sequence (TR 5 4000 ms, TE 5 6
positive cells using SPSS (version 22.0, IBM, Armonk, NY). The
ms, FA 5 90 8, ETL 5 24, and NEX 5 1). Field-of-view and recon-
significance level was set at P < 0.05. Pearson product-moment
struction matrices were set at 120 3 120 mm2 and 512 3 512.
correlation was performed to determine the relationship between
Twelve to fifteen 2.0-mm-thick slice sections with 0.4-mm gaps
the normalized relative intensity of the left liver lobe on day 7 and
were acquired.
changes in the hepatoma size in Group 1.
Image Analysis
Image data were outputted into a Digital Imaging and Communi-
Results
cations in Medicine (DICOM) format and transferred to a One rat in Group 1 failed to establish tumor growth and
Windows-based computer for analyses. A signal inhomogeneity was excluded from the analysis. Furthermore, one rat in
correction was performed on T2*- and T2-weighted images with Group 3 was excluded because of anesthesia-related death
an in-house program using the signal distribution on proton-densi- while being irradiated in the x-ray unit. Therefore, 22 rats
ty-weighted images as described previously.21 A radiologist with 12 were included in the analysis. The median body weights (g)
years of experience (T.F.) subsequently placed 4 mm2 regions of in Groups 1, 2, 3, and 4 rats were 216.5, 228.5, 199, and
interest (ROIs) on the left liver lobe, right liver lobe, tumor, and 218, respectively, on day 7. Body weight loss was mild on
paraspinal muscle and recorded the signal intensity (SI) of these irradiated rats compared with control rats, except for one
ROIs. Relative intensity was calculated by dividing the SI of the
rat which showed more than 20% weight loss. No rats had
left liver lobe, right liver lobe, and tumor by that of the paraspinal
sign of gastrointestinal bleeding in all groups on day 7.
muscle. Normalized relative intensity was then calculated by divid-
Each MRI session, in which four rats were simultaneously
ing the relative intensity at each time point by that at the first
time point (i.e., before the administration of SPIOs or saline) to
examined, was completed within 21–78 min. The time from
evaluate temporal SI changes. anesthesia induction to the end of anesthesia in each MRI
We also assessed the presence or absence of hypointense areas session was 57–150 min. It was 53–118 min in each radia-
in tumor-bearing left liver lobes. If hypointense areas were tion session.
observed, we further classified these hypointense areas based on Percent increases in tumor volumes in Groups 1, 2, 3,
severity of hypointensity (mild or severe) and the shape of the and 4 were -30 6 32, 411 6 74, -82 6 13, and 148 6 113
hypointense area (such as well-defined, ill-defined, or ring-like). (average 6 standard error [SE]), respectively, on day 7.
January 2017 305

TABLE 1. Appearance of T2*-weighted Hypointense Areas in Left Liver Lobes on Day 7
Group 1 2 3 4
SPIO Administration 4 hrs before irradiation 1 1 2 2

Irradiation on Day 0 1 2 1 2
Types of hypointense areas (signal strength/shape) Residual tumor
Type 1 (severely, well-defined) 1 4a 0 0 0
2 2 0 0 0
Type 2 (mildly, ill-defined) 1 0 0 2 0
- 0 0 1 0
Type 3 (mildly or severely, ring-like) 1 2 4 2 5
2 0 0 0 0
Type 1 hypointense areas were only observed in Group 1 rats.
a
Two out of four rats showed not only Type 1, but also Type 2 hypointense areas. Of the 24 rats used in the study, 22 were included
in the table (1 inoculation failure, 1 anesthesia-related death).
Percent increases were significantly lower in Groups 1 and 3 In the latter, two rats showed Type 1 hypointense areas
than in Group 2 (P < 0.05). at the same slice level at which the residual tumor was dis-
In the MR examination on day 7, we found three pat- played. The other two rats showed Type 1 hypointense areas
terns of hypointense areas with or without residual tumors at different slice levels (Fig. 2A), and instead showed Type 2
T1 in the left liver lobes of rats (Table 1): a severely, well- hypointense areas at the same slice levels at which the resid-
defined hypointense area (hereafter called Type 1, represen- ual tumors were displayed (Fig. 2B).
F1 F2 tative cases are shown in Figs. 1 and 2A), a mildly, ill- Figure 3 shows the relationship between the presence F3
defined hypointense area (hereafter called Type 2, a repre- of hypointense areas and tumor responses to radiotherapy in
sentative case is shown in Fig. 2B), and a ring-like hypoin- Group 1. In all rats that showed Type 1 hypointense areas,
tense area (hereafter called Type 3, a representative case is tumors decreased in size. In contrast, tumors increased in
shown in Fig. 2C). Type 1 hypointense areas were observed size in two rats that showed Type 3 hypointense areas sur-
only in Group 1 rats (6 of 8) that received SPIO and subse- rounding the tumors.
quent hepatic irradiation (Fig. 1). They ranged from 58 to Type 2 hypointense areas were also observed in three
593 (average 6 SE, 347 6 89, n 5 6) mm3. irradiated rats not administered SPIO in Group 3 (Table 1).
The borders of the hypointense areas consisted of intra- They were frequently accompanied by residual tumors (2/
parenchymal straight or slightly curvilinear lines and the edge 3). In the remaining two irradiated rats in Group 3, Type 3
of the left liver lobe. Type 1 hypointense areas were observed hypointense areas were observed. Type 3 hypointense areas
without residual tumors in two rats, whereas those in the other were also frequently observed in sham-irradiated rats (four
four rats were observed with residual tumors at volumes of of four in Group 2 and five of five in Group 4). These
41–303 (average 6 SE, 115 6 63, n 5 4) mm3. Type 3 hypointense areas surrounded enlarged hepatomas.
FIGURE 1: Representative T2*-weighted axial (A), coronal (B), and sagittal (C and D) MR images of the liver, showing Type 1
(severely, well-defined) hypointense areas. These images were acquired 7 days after the administration of SPIO and subsequent
hepatic irradiation in two different rats in Group 1. The borders of the hypointense areas consist of intraparenchymal straight or
slightly curvilinear lines (arrowheads) and the edge of the left liver lobe (arrows). Note that no residual tumor is observed in the
hypointense area in one rat (A-C), whereas a small residual tumor (t) is surrounded by hypointense areas in another rat (D).

FIGURE 2: Representative T2*-weighted, axial MR images of the liver, showing the three types of hypointense areas. These
images were acquired 7 days after the administration of SPIO and subsequent hepatic irradiation in two different rats in Group 1.
A severely, well-defined hypointense area (Type 1) in the left liver lobe (A) is surrounded by an intraparenchymal straight line
(arrowheads) and the edges of the liver lobe (arrows). At the adjacent slice level (B), a mildly, ill-defined hypointense area (Type
2) surrounds a hyperintense residual tumor (t). Note that the Type 2 hypointense area is not clearly demarcated, except for the
edges of the liver lobe (arrows), and does not show an intraparenchymal straight line. This Type 2 hypointense area is likely to be
produced by a partial volume effect from the adjacent Type 1 hypointense area in (A). In another rat, a ring-like hypointense area
(Type 3, open arrows) is observed surrounding an enlarged tumor (t) (C).
On day 3, Type 1 hypointense areas were not detected CD68-positive cells in nonirradiated left and right liver
in Group 1 rats because severely, diffuse hypointense areas lobes were as follows: Group 2, 68 6 27/mm2 and 51 6 9/
in both liver lobes hindered detection (Supporting Fig. S4, mm2; Group 4, 93 6 21/mm2 and 119 6 28/mm2. In the
Supporting Information 4, which shows representative T2*- statistical analysis, the number of CD68-positive cells in the
weighted images of the liver on day 3). Type 2 or 3 hypoin- left liver lobe of Group 3 was significantly larger (P < 0.05)
tense areas were observed in Groups 3 and 4 even on day 3. than that of all the other lobes.
F4 Figure 4 shows temporal changes in normalized rela-
tive intensity on T2*-weighted MR images. The SI of the Discussion
left liver lobe in rats administered SPIOs and subsequently
In the present study, we sequentially acquired T2*-weighted
treated with hepatic irradiation (Group 1) showed a recov-
MR images of the liver after irradiation to the previously
ery delay from the maximum signal drop time point
SPIO-accumulated, hepatoma-bearing liver lobes of rats,
(10 min after the SPIO injection) relative to that of the
and frequently observed severely, well-defined hypointense
right liver lobe. As a result, a difference in signals among
(Type 1) areas 7 days after irradiation (six of eight). The
the left liver lobe, right liver lobe, and tumor was observed
presence of Type 1 hypointense areas was associated with
on day 7 (Fig. 4A). In the statistical analysis, the normalized
relative intensity of the left liver lobe was significantly lower
than that of the tumor (P < 0.05) on day 7, whereas no sig-
nificant differences in signal intensities were noted between
the left and right liver lobes. Pearson product-moment cor-
relation analysis revealed that normalized relative intensity
of the left liver lobe on day 7 correlated with changes in the
hepatoma size in Group 1 (r 5 0.73; P 5 0.03).
Necrotic regions dominated on H&E-stained slides,
and minimal non-necrotic tumor cells were detected in rats
F5 that received irradiation (Groups 1 and 3) (Fig. 5). Of
these, two rats in Group 1 and one rat in Group 3 showed
no residual tumors, which corresponded to MR findings.
On the other hand, a mixture of viable and necrotic tumor
cells was found in sham-irradiated rats (Groups 2 and 4). FIGURE 3: A waterfall plot showing percent increases in tumors
in Group 1 rats 7 days after irradiation. In all rats showing
The number of CD68-positive cells was higher in irra-
Type 1 hypointense areas (solid black bars), tumors decreased
F6 diated liver areas than in nonirradiated liver areas (Fig. 6). in size. In two rats showing Type 3 hypointense areas (dotted
The numbers of CD68-positive cells in irradiated-left and bars), tumors increased in size. One tumor (asterisk) did not
non-irradiated right liver lobes were as follows: Group 1, respond to radiotherapy, which may have been because it was
a rapidly growing tumor; we observed a large mass with exten-
205 6 46/mm2 and 107 6 17/mm2; Group 3, 442 6 101/ sive liquefactive necrosis even at the time of laparotomy just
mm2 and 162 6 93/mm2 (mean 6 SE). The numbers of before irradiation.
January 2017 307

FIGURE 4: Time course of normalized relative intensity of the left liver lobe, right liver lobe, and tumor on T2*-weighted MR
images among the four groups (A–D). Graph A shows that the intensity in the irradiated, left liver lobe remained lower than that
in the nonirradiated, right liver lobe and irradiated tumor on day 7. Graphs B to D show that the intensities between the left and
right liver lobes are not significantly different when the administration of SPIO or irradiation (RTx) was not performed. Moreover,
graphs C and D show that the intensities of liver lobes and tumors are equivalent when SPIO was not administered. The asterisks
indicate that the intensity of the left liver lobe or the right liver lobe is significantly lower than that of the tumor (P < 0.001,
Tukey-Kramer method). The double asterisk indicates that the intensity of the left liver lobe is significantly lower than that of the
tumor (P 5 0.008, Tukey-Kramer method). Data are the mean 6 standard error of 8, 4, 5, and 5 values in Groups 1, 2, 3, and 4,
respectively, except for 3 values of the tumor on day 7 (2 in Group 1 and 1 in Group 3). These values were not measurable
because of complete tumor regression.
tumor regression because the tumors in all rats with this including a partial volume effect from the adjacent Type 1
type of hypointense areas decreased in size. hypointense area in Group 1. Even in irradiated liver paren-
In contrast, Type 1 hypointense areas were absent, but chyma without SPIO accumulation in Group 3 occasionally
ring-like hypointense areas (Type 3) surrounded enlarged showed Type 2 hypointense areas. Therefore, we did not
hepatomas in the remaining two rats. The Type 3 hypoin- determine the diagnostic value of Type 2 hypointense areas.
tense area was considered to be associated with tumor We contend that the severely, well-defined (Type 1)
enlargement. Indeed, similar ring-like hypointense areas hypointense areas were indicative of irradiated areas because
were frequently observed with enlarged hepatomas in sham- they were in accordance with the square-shaped X-ray ports
irradiated rats (nine of nine in Groups 2 and 4). The ring- we used. These hypointense areas had straight borders adja-
like hypointense areas were likely to be the liver parenchyma cent to isointense liver areas. These straight borders were
compressed by the expanding tumors. Type 2 hypointense reportedly observed between irradiated and nonirradiated
area is likely to be produced by various mechanisms, liver tissues.22,23 Because we delivered 15 3 15 mm2

C
O
L
O
R
FIGURE 5: Representative hematoxylin and eosin-stained specimens of irradiated or non-irradiated, tumor-bearing liver lobes on
day 7. Figure A shows regressed tumor tissue (arrowheads) containing a necrotic tumor area (arrows) after radiotherapy (RTx1) in
Group 1, whereas Figure B shows enlarged tumor tissue (dotted line) after sham irradiation (RTx2) in Group 2. The tumor consists
of a mixture of viable and necrotic (open arrows) tumor areas. Figures C and D show similar results to those in Figures A and B,
respectively. These results suggest that tumor responses to radiotherapy are equivalent irrespective of the administration of
SPIO. Bars represent 300 lm.
collimated X-rays to the liver lobes, which typically had a are highly unlikely in rats treated with hepatic irradiation
ventro–dorsal thickness of 5 mm, as well as to the hepatoma and subsequently administered SPIO (i.e., the reverse of our
bulging from the liver lobe (Supporting Fig. S1, Supporting method), because previous studies have indicated that irradi-
Information 1, which explains the relationships among the ated liver areas may show higher signals under this condi-
irradiated field, liver tumor, and surrounding liver parenchy- tion.9,10 They attributed the T2*-high signals to radiation-
ma), we consequently observed smaller volumes for the Type induced impairments in the phagocytotic functions of KCs
1 hypointense area than the maximum possible irradiated vol- to SPIO particles. The high signals in irradiated liver paren-
ume (i.e., 58–593 mm3 versus 15 3 15 3 5 5 1125 mm3). chyma presumably hinder the detection of liver tumors with
As described above, irradiated areas were observed at equivalent high signals on T2*-weighted images.
different slice levels from those with tumors in two rats. We presumed that the cause of the severely, well-
The target lesions may have been missed despite the X-ray defined hypointense areas was SPIO-derived iron deposition
port being set just above the lesions. Body motions, includ- in the irradiated liver. SPIO-derived iron deposition is
ing respiratory motions, during the 10-min irradiation peri- known to produce hypointense areas in the liver on T2*-
od may have resulted in malposition. A respiratory gating weighted images.24,25 In the healthy liver, SPIO-derived
technique overcomes this issue, but was not available in the iron deposition disappears within approximately 7 days.26
present study. However, iron deposition persisted for more than 7
Severely, well-defined hypointense areas were only days in the liver parenchyma treated with high-dose irradia-
observed in rats administered SPIO and subsequently tion, and produced T2*-weighted hypointense areas at 9.4
treated with hepatic irradiation; the administration of SPIO tesla, as previously reported in an animal study by Furuta
or hepatic irradiation alone did not produce them. Further- et al.14 This study attributed persistent hepatic iron deposi-
more, the visualization of T2*-hypointense irradiated areas tion to the delayed clearance of SPIO particles from the
January 2017 309

C
O
L
O
R
FIGURE 6: Representative immunohistochemical staining for CD68 of irradiated or non-irradiated liver lobes on day 7. Figure A
was acquired after the administration of SPIO and subsequent radiotherapy (RTx1) in Group1. Figure A shows a higher number of
CD68-positive cells in the sinusoid than Figure B, which was acquired after the administration of SPIO and sham irradiation (RTx2)
in Group 2. Figure C similarly shows higher numbers of CD68-positive cells after irradiation than Figure D, which was acquired
after sham irradiation, even when SPIO was not administered. These results suggest that irradiation to the liver increased CD68-
positive cell numbers irrespective of the administration of SPIO. Bars represent 100 lm.
irradiated liver, which have been due to radiation-induced The present preclinical study demonstrated that this SPIO
impairments in the exocytotic functions of KCs to SPIO dosing technique is applicable for visualizing the radiation
particles. The results of the present study suggest that a sim- safety margin and hepatic tumors, and, thus, warrants further
ilar mechanism gave rise to the T2*-weighted hypointense study to demonstrate its clinical usefulness.
areas in irradiated liver areas, and these areas were observ- While some tumors did not respond to irradiation,
able even at clinically-relevant 3 tesla. our results suggest that irradiation causes tumor regression
We contend that the administration of SPIO before because of the negative values for % increases in tumor vol-
irradiation may reduce the total dose of SPIO in longitudi- umes in irradiated rats. Furthermore, irradiation effectively
nal MR examinations for radiation-treated hepatic tumors. induced tumor necrosis because the proportion of necrosis
The present study demonstrated that once SPIOs were admin- in tumor tissue was greater in irradiated rats than in sham-
istered before the start of radiotherapy, additional SPIO dos- irradiated rats.
ing was not necessary to visualize hepatic tumors on T2*- We assessed the number of CD68 positive cells
weighted MR images for at least one week after irradiation. In because CD68 is a well-known antigen that is highly
addition, these irradiated tumors and hypointense-irradiated expressed on the surface of Kupffer cells.30 Kupffer cells
areas were highly contrasted. The high contrast between the phagocyte SPIO particles, and are attributed to signal reduc-
tumor and irradiated liver parenchyma may allow for the safe- tion on T2*-weighted images. CD68-positive cells increased
ty margin status to be evaluated and the risk of local recur- in number in irradiated liver lobes. A possible reason for
rence after radiotherapy to be predicted. A similar SPIO this increase is a subacute hepatic inflammatory reaction
dosing technique has already been applied to the MR imaging induced by irradiation.31 In addition to CD68-positive
of liver cancer lesions treated with radiofrequency ablation27 KCs, CD68-positive macrophages and monocytes may be
and clinically used to evaluate the ablative margin status.28,29 present in the irradiated liver.32 Although a higher number

of CD68-positive cells does not directly connect with the Japan Society for the Promotion of Science to M.Y. The
functional KC population, it suggests that an increase in authors thank Dr. Michiko Nagai and Mr. Keisuke Iijima
these inflammatory cells in response to irradiation may be for their support in the animal experiments. We also thank
related to a regional alteration in iron metabolism by KCs, Mr. Akira Nabetani and Mr. Atsushi Nozaki of GE Health-
resulting in the deposition of SPIO-derived iron particles. care Japan for constructing the 3T multiarray coil as well as
Further studies are needed to elucidate this. Mr. Kenji Yamazaki of the Association for Nuclear Technol-
This study had several limitations. First, we delivered a ogy in Medicine for constructing the animal beds used in
single dose of 70 Gy of irradiation to the liver. This irradia- this study. The authors also thank Dr. Shusei Hamamichi
tion protocol is not clinically relevant. Our study protocol for English proofreading.
did not allow for dose fractionation experiments because we
delivered X-rays to liver lobes that had been pulled out
from the abdominal cavity; therefore, the delivery of X-rays
for several days may result in further complications to lapa- References
rotomy procedures. Even if laparotomy procedures are 1. Torre LA, Bray F, Siegel RL, Ferlay J, Lortet-Tieulent J, Jemal A. Glob-
al cancer statistics, 2012. CA Cancer J Clin 2015;65:87–108.
avoided, it is difficult to produce reproducible irradiated
2. Kirkpatrick JP, Kelsey CR, Palta M, et al. Stereotactic body radiothera-
areas in each radiation fraction with our equipment. Fur- py: a critical review for nonradiation oncologists. Cancer 2014;120:
thermore, we delivered a clinically unjustifiable single radia- 942–954.
tion dose to the liver from the view point of hepatic 3. Van De Voorde L, Vanneste B, Houben R, et al. Image-guided stereo-
tactic ablative radiotherapy for the liver: a safe and effective treat-
toxicity.33,34
ment. Eur J Surg Oncol 2015;41:249–256.
However, because we contend that a low dose of irra-
4. Poon RT, Fan ST, Ng IO, Wong J. Significance of resection margin in
diation will not control N1-S1 hepatomas, the high dose of hepatectomy for hepatocellular carcinoma: a critical reappraisal. Ann
irradiation at 50–70 Gy was experimentally acceptable. As Surg 2000;231:544–551.
was shown above, some hepatomas became enlarged even 5. Akino Y, Oh RJ, Masai N, Shiomi H, Inoue T. Evaluation of potential
internal target volume of liver tumors using cine-MRI. Med Phys 2014;
after the delivery of 70 Gy of irradiation; therefore, they 41:111704.
may have been resistant to radiation. Second, we used ROI
6. Xu Q, Hanna G, Grimm J, et al. Quantifying rigid and nonrigid motion
analysis. Although manual positioning of ROIs was done by of liver tumors during stereotactic body radiation therapy. Int J Radiat
a single observer to minimize variability in our study, the Oncol Biol Phys 2014;90:94–101.
method is strongly operator dependent. Quantitative T2*- 7. Clement O, Muhler A, Vexler VS, et al. Evaluation of radiation-
induced liver injury with MR imaging: comparison of hepatocellular
mapping will be more precise methods to compare T2* val- and reticuloendothelial contrast agents. Radiology 1992;185:163–
ues, although calculation of accurate quantitative value is 168.
beyond the scope of our study. Finally, only one single sub- 8. Masui T, Yan H, Kosugi I, et al. Assessment of early radiation effects
jective observer performed semiquantitative ROI analysis as on the liver. Comparison of SPECT and MR. Acta Radiol 1996;37:665–
671.
mentioned above.
9. Morimoto N, Ebara M, Kato H, et al. Early detection of radiation-
In conclusion, SPIO-enhanced MR imaging visualized induced liver injury in rat by superparamagnetic iron oxide-enhanced
irradiated liver areas as severely, well-defined hypointense MR imaging. J Magn Reson Imaging 1999;9:573–578.
areas and liver tumors as hyperintense lesions 7 days after a 10. Yoshioka H, Itai Y, Saida Y, Mori K, Mori H, Okumura T. Superpara-
magnetic iron oxide-enhanced MR imaging for early and late
high radiation dose was delivered to the previously SPIO-
radiation-induced hepatic injuries. Magn Reson Imaging 2000;18:
accumulated, hepatoma-bearing liver lobes of rats. The pres- 1079–1088.
ence of Type 1 hypointense areas was associated with tumor 11. Sanuki N, Takeda A, Oku Y, et al. Threshold doses for focal liver reac-
regression. In contrast, the observation of hypointense rings tion after stereotactic ablative body radiation therapy for small hepa-
tocellular carcinoma depend on liver function: evaluation on magnetic
was related to tumor growth. Collectively, these results indi- resonance imaging with Gd-EOB-DTPA. Int J Radiat Oncol Biol Phys
cate the potential of SPIO-enhanced MR imaging as a 2014;88:306–311.
modality to monitor irradiated liver parenchyma and liver 12. Jarraya H, Borde P, Mirabel X, et al. Lobulated enhancement evalua-
tion in the follow-up of liver metastases treated by stereotactic body
tumor responses to radiotherapy. This MR technique may
radiation therapy. Int J Radiat Oncol Biol Phys 2015;92:292–298.
also contribute to determining the radiation safety margin
13. Lall C, Bhargava P, Sandrasegaran K, et al. Three-dimensional confor-
adjacent to liver cancer lesions. mal radiation therapy in the liver: MRI findings along a time continu-
um. J Comput Assist Tomogr 2015;39:356–364.
14. Furuta T, Yamaguchi M, Nakagami R, Akahane M, Minami M, Ohtomo

Acknowledgments K, Fujii H. Delayed hepatic signal recovery on ferucarbotran-enhanced
Contract grant sponsor: Japan Society for the Promotion of magnetic resonance images in a rat model with regional liver irradia-
tion. Magn Reson Mater Phy 2014;27:501–508.
Science; contract grant number: 25461864
15. Yamaguchi M, Mitsuda M, Ezawa K, et al. Artifact-reduced simulta-
The authors have no conflict of interest to disclose. This neous MRI of multiple rats with liver cancer using PROPELLER.
study was partially supported by JSPS KAKENHI from the J Magn Reson Imaging 2013;38:225–230.
January 2017 311

16. Guo Y, Zhang Y, Klein R, et al. Irreversible electroporation therapy in 26. Weissleder R, Stark DD, Engelstad BL, et al. Superparamagnetic iron oxide:
the liver: longitudinal efficacy studies in a rat model of hepatocellular pharmacokinetics and toxicity. AJR Am J Roentgenol 1989;152:167–173.
carcinoma. Cancer Res 2010;70:1555–1563.
27. Nagai M, Yamaguchi M, Mori K, et al. Magnetic resonance-based
17. Babsky AM, Ju S, George B, et al. Predicting response to benzamide visualization of thermal ablative margins around hepatic tumors by
riboside chemotherapy in hepatocellular carcinoma using apparent means of systemic ferucarbotran administration before radiofrequency
diffusion coefficient of water. Anticancer Res 2011;31:2045–2051. ablation: animal study to reveal the connection between excess iron
deposition and T2*-weighted hypointensity in ablative margins. Invest
18. Babsky AM, Ju S, Bennett S, George B, McLennan G, Bansal N. Effect
Radiol 2015;50:376–383.
of implantation site and growth of hepatocellular carcinoma on appar-
ent diffusion coefficient of water and sodium MRI. NMR Biomed 28. Mori K, Fukuda K, Asaoka H, et al. Radiofrequency ablation of the liver:
2012;25:312–321. determination of ablative margin at MR imaging with impaired clear-
19. Guo Y, Jin N, Klein R, et al. Gas challenge-blood oxygen level- ance of ferucarbotran--feasibility study. Radiology 2009;251:557–565.
dependent (GC-BOLD) MRI in the rat Novikoff hepatoma model. 29. Koda M, Tokunaga S, Miyoshi K, et al. Ablative margin states by mag-
Magn Reson Imaging 2012;30:133–138. netic resonance imaging with ferucarbotran in radiofrequency ablation
20. Thompson SM, Callstrom MR, Knudsen B, et al. Development and for hepatocellular carcinoma can predict local tumor progression.
preliminary testing of a translational model of hepatocellular carcino- J Gastroenterol 2013;48:1283–1292.
ma for MR imaging and interventional oncologic investigations. 30. Dijkstra CD, Dopp EA, Joling P, Kraal G. The heterogeneity of mono-
J Vasc Interv Radiol 2012;23:385–395. nuclear phagocytes in lymphoid organs: distinct macrophage subpo-
21. Mitsuda M, Yamaguchi M, Nakagami R, et al. Intensity correction pulations in the rat recognized by monoclonal antibodies ED1, ED2
method customized for multi-animal abdominal MR imaging with 3T and ED3. Immunology 1985;54:589–599.
clinical scanner and multi-array coil. Magn Reson Med Sci 2013;12: 31. Hebard DW, Jackson KL, Christensen GM. The chronological devel-
95–103. opment of late radiation injury in the liver of the rat. Radiat Res 1980;
22. Itai Y, Murata S, Kurosaki Y. Straight border sign of the liver: spectrum 81:441–454.
of CT appearances and causes. Radiographics 1995;15:1089–1102. 32. Takeda A, Sanuki N, Eriguchi T, et al. Stereotactic ablative body
23. Park HC, Seong J, Han KH, Chon CY, Moon YM, Suh CO. Dose- radiotherapy for previously untreated solitary hepatocellular carcino-
response relationship in local radiotherapy for hepatocellular carcinoma. J Gastroenterol Hepatol 2014;29:372–379.
ma. Int J Radiat Oncol Biol Phys 2002;54:150–155.
33. Lawrence TS, Robertson JM, Anscher MS, Jirtle RL, Ensminger WD,
24. Wang YX, Hussain SM, Krestin GP. Superparamagnetic iron oxide Fajardo LF. Hepatic toxicity resulting from cancer treatment. Int J
contrast agents: physicochemical characteristics and applications in Radiat Oncol Biol Phys 1995;31:1237–1248.
MR imaging. Eur Radiol 2001;11:2319–2331.
34. Christiansen H, Sheikh N, Saile B, et al. x-Irradiation in rat liver:
25. Reimer P, Balzer T. Ferucarbotran (Resovist): a new clinically approved RES- consequent upregulation of hepcidin and downregulation of hemo-
specific contrast agent for contrast-enhanced MRI of the liver: properties, juvelin and ferroportin-1 gene expression. Radiology 2007;242:
clinical development, and applications. Eur Radiol 2003;13:1266–1276. 189–197.

01 Jmri - 2017 PDF

Uploaded by

Copyright:

Available Formats

You might also like

01 Jmri - 2017 PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

01 Jmri - 2017 PDF

Uploaded by

Copyright:

Available Formats

ri

Officers of the Society ISMRM Advertising Policy

Bronze Corporate Members

Editor-in-Chief Mark E. Schweitzer, MD

Editors Emeritus C. Leon Partain, MD, PhD

Abdomen Imaging Evidence-Based Practice Physics

Ronald Arildsen, MD Robert Edelman, MD Michael E. Moseley, PhD Neil M. Rofsky, MD

11 Practical Applications of Balanced Steady-State Free-Precession (bSSFP) Imaging

21 Whole-Body PET/MRI for Colorectal Cancer Staging: Is It the Way Forward?

36 Guidelines for Documentation and Consent for Nonclinical, Nonresearch MRI in

51 Quantifying Metal-Induced Susceptibility Artifacts of the Instrumented Spine at

59 MRI of the Knees in Asymptomatic Adolescent Soccer Players:

66 Comparison of Chemical Shift-Encoded Water Fat MRI and MR Spectroscopy in

Breast 74 Background Parenchymal Enhancement Over Exam Time in Patients With

84 Stimulated Echo Diffusion Tensor Imaging (STEAM-DTI) with Varying Diffusion

94 Dynamic Contrast-Enhanced and Diffusion-Weighted MRI of Estrogen

118 Incremental Value of MRI for Preoperative Penile Cancer Staging

VOLUME 45, NUMBER 1, JANUARY 2017

139 MRI-Based Computational Hemodynamics in Patients With Aortic Coarctation

147 Effects of Cortisol on the Heart: Characterization of Myocardial Involvement in

167 Benign and Malignant Orbital Lymphoproliferative Disorders: Differentiating

177 Intrinsic Brain Network Abnormalities in Codeine-Containing Cough

187 Reproducibility Measurement of Glutathione, GABA, and Glutamate: Towards

199 Differential Diagnosis of Mitochondrial Encephalopathy With Lactic Acidosis and

Neuro 207 Magnetic Susceptibility-Induced Echo-Time Shifts: Is There a Bias in Age-Related

229 Thrombus-Mimicking Artifacts in Two-Point Dixon MRI: Prevalence, Appearance,

237 Optimization of Two-Compartment-Exchange-Model Analysis for Dynamic

260 Intravoxel Incoherent Motion Diffusion-Weighted Imaging of the Pancreas:

281 Gadoxetic Acid-Enhanced MRI for the Characterization of Hepatocellular

303 Persistent T2*-Hypointensity of the Liver Parenchyma After Irradiation to the

Note to Contributors on Deposit of Accepted

Practical Applications of Balanced

B alanced Steady-State Free-Precession (bSSFP; True FISP

View this article online at wileyonlinelibrary.com. DOI: 10.1002/jmri.25336

C 2016 International Society for Magnetic Resonance in Medicine

ID: srinivasanv Time: 15:48 I Path: //chenas03/Cenpro/ApplicationFiles/Journals/Wiley/JMRI/Vol04501/160160/Comp/APPFile/JW-JMRI160160

Journal of Magnetic Resonance Imaging

tion will be added to the residual transverse magnetization.

12 Volume 45, No. 1

ID: srinivasanv Time: 15:48 I Path: //chenas03/Cenpro/ApplicationFiles/Journals/Wiley/JMRI/Vol04501/160160/Comp/APPFile/JW-JMRI160160

Schieda et al.: bSSFP Imaging in the Abdomen and Pelvis

Applications of bSSFP in Abdominal and

ID: srinivasanv Time: 15:48 I Path: //chenas03/Cenpro/ApplicationFiles/Journals/Wiley/JMRI/Vol04501/160160/Comp/APPFile/JW-JMRI160160

Journal of Magnetic Resonance Imaging

14 Volume 45, No. 1

ID: srinivasanv Time: 15:49 I Path: //chenas03/Cenpro/ApplicationFiles/Journals/Wiley/JMRI/Vol04501/160160/Comp/APPFile/JW-JMRI160160

Schieda et al.: bSSFP Imaging in the Abdomen and Pelvis

ID: srinivasanv Time: 15:49 I Path: //chenas03/Cenpro/ApplicationFiles/Journals/Wiley/JMRI/Vol04501/160160/Comp/APPFile/JW-JMRI160160

ID: srinivasanv Time: 15:49 I Path: //chenas03/Cenpro/ApplicationFiles/Journals/Wiley/JMRI/Vol04501/160160/Comp/APPFile/JW-JMRI160160

Schieda et al.: bSSFP Imaging in the Abdomen and Pelvis

ID: srinivasanv Time: 15:49 I Path: //chenas03/Cenpro/ApplicationFiles/Journals/Wiley/JMRI/Vol04501/160160/Comp/APPFile/JW-JMRI160160

Journal of Magnetic Resonance Imaging

18 Volume 45, No. 1

ID: srinivasanv Time: 15:49 I Path: //chenas03/Cenpro/ApplicationFiles/Journals/Wiley/JMRI/Vol04501/160160/Comp/APPFile/JW-JMRI160160

Schieda et al.: bSSFP Imaging in the Abdomen and Pelvis

ID: srinivasanv Time: 15:49 I Path: //chenas03/Cenpro/ApplicationFiles/Journals/Wiley/JMRI/Vol04501/160160/Comp/APPFile/JW-JMRI160160

Journal of Magnetic Resonance Imaging

20 Volume 45, No. 1

Whole-body PET/MR (25-30 min) Dedicated MR* (30 min)