MR Radar:

Parallel Radiofrequency Transmission in Principle and Practice

by

Cem Murat Deniz

A dissertation submitted in partial fulfillment

of the requirements for the degree of

Doctor of Philosophy

Department of Basic Medical Science

Program in Biomedical Imaging

New York University

May, 2012

___________________________
Daniel K. Sodickson, M.D., Ph.D.

___________________________
Yudong Zhu. Ph.D.
 UMI Number: 3524145

All rights reserved

INFORMATION TO ALL USERS

The quality of this reproduction is dependent on the quality of the copy submitted.

In the unlikely event that the author did not send a complete manuscript
and there are missing pages, these will be noted. Also, if material had to be removed,
a note will indicate the deletion.

UMI 3524145
Copyright 2012 by ProQuest LLC.
All rights reserved. This edition of the work is protected against
unauthorized copying under Title 17, United States Code.

ProQuest LLC.
789 East Eisenhower Parkway
P.O. Box 1346
Ann Arbor, MI 48106 - 1346
 © Cem Murat Deniz

All Rights Reserved 2012

To Nazik
 ACKNOWLEDGEMENTS

I arrived in New York City for my Ph.D. studies five years ago on a rainy,

stormy summer day with Nazik. It is hard to believe that I am writing the

acknowledgement section of my dissertation now. When I look back, I feel deep

gratitude to those people who filled this journey with beautiful memories. I would like

to thank all those people for their help, support, guidance and friendship.

I owe my deepest gratitude to my thesis advisors Daniel K. Sodickson and

Yudong Zhu, who have been teachers, colleagues, and more importantly friends to me.

I feel very fortunate to be part of their research team. Working with Dan was an

amazing learning experience. His enthusiasm and continuous encouragement allowed

me to go beyond my horizons. He always welcomed my questions, critiques, ideas and

provided excellent guidance and feedback whenever I needed them. He taught me how

to become an independent researcher. It was extremely encouraging to feel his support

all the time. Yudong's critical thinking provided a model for me in doing research. I

believe the way I approach a scientific question has developed tremendously over the

years I have worked with Yudong. He was always available whenever I needed his

feedback. I would show up at his door to discuss my questions and he would welcome

me without any hesitation.

My Ph.D. journey was mostly fun thanks to those friends with whom I worked,

hung out, and laughed. I cannot find words to thank Leeor Alon who has been a friend,

colleague, and above all a brother to me. I will always remember our sleepless nights

v
in the 7T room. I don't think any other person could put up with the jokes we make all

the time. Thank you, Leeor!

Gene Cho... You are an amazing friend! Thank you for being my primary

search engine, financial consultant, and personal trainer. You were the best roommate

at the conferences. Thanks for the drinks at the rooftop bar, brother!

Ryan Brown... Ryaaaannn... You are a great researcher and friend. It was

always fun to work with you. Thanks for the detailed edits on all my manuscripts. I

will never forget how we figured out the problem with the parallel transmit system in

the hip study. The golden medals we received for our accomplishment were the best

part of it.

I would like to thank Illiyana Atanasova for being so patient and understanding

for all the jokes and teases coming from us, especially me. Thank you Ili, for creating

a sense of community in the room with all your organizations.

I want to thank everyone in room 420: Vishal Patil, Li Feng, Ding Xia, Elan

Grossmann, Manuska Vaidya, Alicia Yang. Room 420 was a nice work environment

thanks to all of you.

I would like to thank my colleagues at CBI, especially Riccardo Lattanzi,

Kellyanne Mcgorty, Graham Wiggins, Ricardo Otazo, Bei Zhang, Daniel Kim, Pippa

Storey, and Assaf Tal. Ricardo Lattanzi, my post doc... It was a pleasure to work with

you. Your sense of humor makes research fun. Kellyanne, thank you for always being

available whenever I had a question during my experiments. I will never forget our

vi
trip to Alberta on your favorite type of plane. I am grateful to Graham, for his

assistance with SNR analysis and coil design; Ricardo, for his help with all my linear

algebra and numerical optimization questions; Bei, for all her help with the FDTD

simulations; Daniel Kim, for helping me to start on parallel transmit experiments and

sequence design; Pippa for answering all my 'interesting' MR related questions; and

Assaf, for the discussions about adiabatic pulses.

I would like to thank the collaborators from Siemens: Hans-Peter Fautz, Ulrich

Fontius, Bernd Stoeckel, and Niels Oesingmann. I am grateful to Hans-Peter for

providing me with the flip angle sequence and all the support I needed during

experiments. Ulrich was extremely helpful with parallel transmit system hardware and

operations. Niels helped me grasp the details of the sequence design using Siemens

idea environment. Bernd assisted me in solving the problems I had with the MRI

system.

I would like to thank my thesis committee, Elfar Adalsteinsson, Leslie

Greengard, Jens Jensen, and Daniel Turnbull for their precious time and invaluable

feedback. Their questions and comments improved the quality of my work.

Above all, I would like to thank my family, my wife Nazik Dinctopal-Deniz

and my parents Huriye and Duran Deniz. I cannot find the words to thank my

beautiful wife for all her support and encouragement during this journey. She has been

extremely understanding about my work schedule. It was nice to have two Ph.D.

students in the same house. She is the source of joy in my life. I am grateful to my

vii
parents for always being supportive of me. They have always trusted me and my

decisions. I am the person I am today thanks to them.

The last but not the least, I want to thank Joel Oppenheim. I still remember his

warm welcome events at NYU.

viii
 ABSTRACT

Magnetic resonance imaging (MRI) has been driven towards high magnetic

fields in order to benefit from correspondingly high signal-to-noise ratio and spectral

resolution. However, technological challenges associated with high magnetic field

strength, such as increase in radiofrequency (RF) energy deposition and RF excitation

inhomogeneity, limit realization of the full potential of these benefits. Parallel RF

transmission enables decreases in RF energy deposition and in the inhomogeneity of

RF excitations by using multiple-transmit RF coils driven independently and operating

simultaneously. In this work, the behavior of RF excitation and RF energy deposition

is explored from an MRI system perspective. New parallel RF excitation techniques

are introduced to measure subject-specific electric field interactions between transmit

elements. These new techniques are demonstrated in phantom and in vivo studies, and

are shown to enable decreases in RF energy deposition while maintaining RF

excitation fidelity. Since the capacity of MRI systems for RF power delivery and

handling are subject to both technological and regulatory limits, a method was

developed to predict the RF power consequence of transmission on each individual

channel during parallel RF transmission, and this method was used to design parallel

transmission RF pulses obeying strict technical and safety limits. Additionally, MRI

system-subject interactions during parallel RF transmission were studied as a function

of the distance between the subject and the transmit RF coils. Lastly, inner-volume RF

excitations were demonstrated as one of the promising potential applications of

ix
parallel RF transmission. In summary, this work represents a step forward in

overcoming technical challenges to demonstrate potential applications of high field

MRI with parallel RF transmission.

x
TABLE OF CONTENTS

ACKNOWLEDGEMENTS v 

ABSTRACT ix 

LIST OF FIGURES xii 

LIST OF TABLES xiv 

INTRODUCTION 1 

CHAPTER 1: Specific Absorption Rate Benefits of Including Measured Electric Field

Interactions in Parallel Excitation Pulse Design 13 

CHAPTER 2: Maximum Efficiency RF Shimming: Theory and Initial Application for

Hip Imaging at 7 Tesla 48 

CHAPTER 3: Subject-specific Proactive Management of Parallel RF Transmission 81 

CHAPTER 4: RF Energy Deposition and RF Power Requirements in Parallel

Transmission with Increasing Distance from the Coil to the Sample 95 

CHAPTER 5: Sparse Parallel Transmit Excitation Trajectory Design for Rapid

Inner-Volume Excitation 107 

CONCLUSION 136 

APPENDIX 140 

REFERENCES 152 

xi
 LIST OF FIGURES

Figure 1.1 FDTD simulation setup. .............................................................................. 20 

Figure 1.2 Excitation k-space and desired profile. ....................................................... 26 

Figure 1.3 Phantom experiment setup. ......................................................................... 29 

Figure 1.4 Additional inputs used in pulse design for the phantom experiments. ....... 31 

Figure 1.5 Global SAR when the power correlation matrix is incorporated into parallel

RF pulse design. ........................................................................................................... 36 

Figure 1.6 Experimental results. ................................................................................... 39 

Figure 1.7 RF pulse waveforms and RF net power ...................................................... 41 

Figure 2.1 Experimental setup. ..................................................................................... 57 

Figure 2.2 Steps required for the calculation of the maximum efficiency RF shimming

weights. ......................................................................................................................... 63 

Figure 2.3 Representative axial GRE images of one volunteer at 7 T ......................... 68 

Figure 2.4 Adiabatic half passage RF pulse results ...................................................... 70 

Figure 2.5 SNR comparison at 3 T and 7 T .................................................................. 76 

Figure 3.1 Example of calibrated power correlation matrices. .................................... 88 

Figure 3.2 Desired excitation profile and k-space trajectory ....................................... 88 

Figure 3.3 Bloch simulation results and axial GRE images of designed RF pulses .... 90 

Figure 3.4 Comparison of individual channel actual power measurements ................. 91 

Figure 3.5 Measured power for RF pulses designed with different power constraints.94 

Figure 4.1 Transmit array geometries for spherical simulations. ................................. 99 

xii
Figure 4.2 Transmit array geometries for cylindrical simulations. ............................ 100 

Figure 4.3 Optimized global SAR and RF power requirements versus lift-off.......... 102 

Figure 4.4 Local SAR vs lift-off for the sphere.......................................................... 105 

Figure 4.5 Optimized global SAR and RF power requirements versus lift-off for the

cylinder. ...................................................................................................................... 106 

Figure 5.1 Schematic illustration of how selectivity in the image domain depends upon

the dimension of excitation k-space. .......................................................................... 114 

Figure 5.2 An example of 2D spiral RF pulse design. ............................................... 116 

Figure 5.3 Various k-space trajectories which are used for 3D selective RF excitation

using one transmit channel. ........................................................................................ 118 

Figure 5.4 B1+ distribution of the individual elements. .............................................. 127 

Figure 5.5 Distribution of the selected k-space locations for both algorithms. .......... 129 

Figure 5.6 Designed k-space trajectories .................................................................... 131 

Figure 5.7 Experimental flip angle profiles of designed LTA RF pulses .................. 131 

Figure 5.8 Axial and sagittal GRE images acquired .................................................. 132 

Figure A.1 Workflow of an RF shimming experiment. ............................................. 145 

Figure A.2 Screenshot of RF Shimming GUI ............................................................ 146 

Figure A.3 Workflow of a parallel transmission experiment. .................................... 150 

Figure A.4 Screenshot of Parallel Transmit GUI ....................................................... 151 

xiii
 LIST OF TABLES

Table 0.1 Temperature limits in MR experiments.......................................................... 4 

Table 0.2 SAR limits for local transmit coils ................................................................. 4 

Table 1.1 Comparison of STA parallel RF pulse behavior in a simulation ................. 37 

Table 1.2 Experimental parallel RF pulse behavior ..................................................... 43 

Table 2.1 Comparison of four RF shimming methods ................................................. 71 

Table 2.2 Experimentally measured net power deposition and corresponding flip angle

...................................................................................................................................... 72 

Table 2.3 Calculated maximum efficiency RF shimming weights .............................. 73 

Table 2.4 SNR results in the hip articular cartilage of the volunteers at 3 T and 7 T. . 75 

Table 3.1 Power comparison of RF pulses with different power constraints. .............. 93 

xiv
 INTRODUCTION

Among today's large variety of medical imaging techniques, Magnetic

Resonance Imaging (MRI) differentiates itself by its non-invasive nature and its soft

tissue contrast, which facilitates diagnostic imaging of the brain, heart, muscles and

many other organs or tissues. MRI is based on the physical phenomenon called

Nuclear Magnetic Resonance (NMR) which was first detected in solid materials

independently by Purcell et al. (1) and Bloch (2) in 1946 after its discovery in gases by

Rabi et al. (3) in 1938. The significant step from NMR that renders MRI experiments

possible was discovered by Lauterbur (4) in 1973. Lauterbur achieved spatial

encoding of the MR signal by superimposing additional magnetic field gradients on

the main magnetic field, thereby enabling the exact position of the NMR signal in the

sample to be decoded and an image to be formed. Based on Laterbur’s gradient

encoding approach, Kumar et al. (5) proposed Fourier imaging in 1975, which formed

the basis of most variants of MRI that are currently in use. However, Fourier imaging

with gradient encoding implied a fundamental restriction on the speed of MRI

acquisition, since only one position in the gradient encoded spatial frequency space

(which was later defined as k-space by Twieg (6) and Ljunggren (7)) could be sampled

at a time.

The speed of MRI acquisition has increased dramatically with improvements

in gradient technology and the development of new fast imaging acquisition

techniques, such as echo-planar imaging (8), turbo spin echo (9) and spiral imaging

1
(10). However, the sequential nature of Fourier encoding was still one of the main

limitations on the achievable speed. The concept of using multiple receivers for the

purpose of scan time reduction in Fourier imaging was suggested in 1988 (11).

However, successful experiments using parallel receivers for the purpose of scan time

reduction were not demonstrated until the introduction of parallel magnetic resonance

imaging methods in the late 1990s (12,13). The advent of parallel MRI opened a wide

area of research into the acceleration of MR scanning through undersampling of

k-space and subsequent reconstruction of missing image information using

complementary information from the elements of radiofrequency (RF) coil arrays.

Numerous parallel imaging reconstruction techniques and strategies have been

developed since then (14-18) and parallel MRI has become a well established

technique widely used in clinical MRI. Despite the advantage of faster scanning with

parallel MRI, loss of signal-to-noise ratio (SNR) in the reconstructed images (as

compared with fully gradient-encoded images using the same coil array) was observed

due to reduced time averaging of noise using fewer k-space samples as well as to noise

amplification in the image reconstruction process.

Since the SNR of the magnetic resonance signal is known to scale up with

increasing main magnetic field strength (B0) (19), the history of MRI has also seen a

progressive increase in field strength, with the emergence in the past decade or so of

ultra-high-field (UHF, ≥ 7 Tesla (T)) MRI systems for human use. High SNR can be

used to improve spatial / temporal resolution for improved image quality and to

2
decrease image acquisition times. However, the practical SNR increase enabled by

UHF-MRI is substantially limited by constraints on the specific absorption rate (SAR),

a measure of RF energy deposition in tissue. SAR is directly related to electric field

(E) inside the subject. In MRI, E field inside the subject is induced by the RF

magnetic field (B1) which interacts with spins and induces MR signal. This

concomitant E field deposits RF energy in the imaged body and determines SAR,

which is subject to regulatory limits (20,21) aimed at preventing unacceptable

temperature increases within the human body. Allowed values for temperature rise of

the patient caused by MR scanner are defined by the International Electrotechnical

Commission (IEC) as shown in Table 0.1. MR scanners are operated in three various

operating modes as can be seen from Table 0.1. Default operating mode of MR

scanner is the normal mode which guarantees that RF power deposition cannot cause a

physiological stress to patients. Other two operating modes can cause physiological

stress to patients and they have to be controlled by medical supervision. Compliance

to the temperature rise limits for local transmit coils can be achieved by limiting the

SAR (Table 0.2), which is derived so that the spatially localized temperatures are not

expected to result in tissue damage.

3
Table 0.1 Temperature limits in MR experiments (Table 201.104 from Ref. (21))

Operating Maximum Core Maximum Local Rise of Core

Mode Temperature Tissue Temperature Temperature °C

Normal 39 39 0.5

First Level 40 40 1
Controlled

Second Level >40 >40 >1

Controlled

Table 0.2 SAR limits for local transmit coils (Table 201.106 from Ref. (21))

Averaging Time 6 min

Local SAR

Body Region Head Trunk Extremities

Operating Time (W/kg) (W/kg) (W/kg)

Normal 10 a 10 20

First Level >20 a 20 40

Controlled

Second Level >20 a >20 >40

Controlled

Short Duration The SAR limits over any 10 s period shall not exceed two times the
SAR stated values

a
NOTE In cases where the orbit is in the field of a small local RF transmit coil, care should be
taken to ensure that the temperature rise is limited to 1°C

4
 As the B0 field strength increases, the magnitude of E field per unit flip angle

increases (19) and safety limits on allowed SAR limit achievable SNR. Additionally,

at UHF, the interaction of the electromagnetic (EM) field with dielectric tissues tends

to exacerbate inhomogeneities in RF power deposition, which may result in dangerous

local hot-spots. In addition to the SAR limitations, inhomogeneity of the appropriately

polarized transverse magnetic field B1+ hampers clinical use of UHF-MRI systems.

This RF inhomogeneity is related to the reduction in RF wavelength at high field and

causes inhomogeneities of the image contrast and SNR, which can diminish the

quality and diagnostic value of MR images.

In order to overcome patient-induced inhomogeneous RF excitation at UHF,

several RF excitation methods have been proposed using multiple RF transmit coils.

The first method was RF shimming (19,22,23), in which multi-element transmit coil

arrays are driven with a single RF waveform by adjusting phase and amplitude in

individual coils independently. This technique has been successful in improving the

B1+ homogeneity in excited volumes, especially in small local regions-of-interest

(ROIs) (24,25). However, the efficiency of RF shimming diminishes as the ROI

becomes larger. Thus, new approaches have been introduced to mitigate B1+

inhomogeneities in large ROIs. For example, various tailored excitation k-space

trajectories (26,27) have been utilized, and have been shown to reduce B1+

inhomogeneity. Later, parallel RF excitation techniques (28,29) combined and

5
extended the benefits of these two approaches. Parallel excitation methods have been

used to compensate for patient-induced RF inhomogeneities at high B0. In parallel RF

excitation, individual elements of multi-element transmit coils are driven

simultaneously with distinct tailored RF pulses sharing a common gradient waveform.

The additional degrees of freedom available in parallel RF excitation pulses can be

used to shorten multidimensional pulses (30,31), improve spatial definition of the

excitation pattern (32) and decrease RF power deposition (29).

Even though the parallel RF excitation offers a means of overcoming technical

problems associated with UHF, the technical development stage of parallel RF

excitation has been slow compared to that experienced in the field parallel MR

reception. The potential of parallel RF excitation has not been fully explored in human

studies due to the complexity and cost of additional equipment required, the

computational complexity of designing RF pulses, especially for large-tip-angle

(LTA) pulses, and importantly, the need for a real time SAR assessment to ensure

patient safety. The requirement of an additional RF pulse synthesizer and amplifier for

each transmit channel increases the cost of parallel transmit equipment as the number

of transmit channels increases. The cost of parallel RF transmit systems is expected to

decrease in the future as add-on prototype systems are replaced with fully integrated

ones. Since the introduction of the first parallel RF excitation pulse designs (28,29),

various new approaches to the design of parallel excitation RF pulses with reduced

computational complexity have been proposed, first for small-tip-angle (STA) (30,32-

6
39) and later for LTA (40-42). Although the developments in parallel RF excitation

pulse design have been shown to mitigate B1+ inhomogeneity effectively and enable

application specific tailored excitation profiles, the usage of parallel RF excitation is

still limited to the STA regime in subjects (43-45) due to concerns about SAR.

The SAR behavior of parallel excitation RF pulses has been studied

extensively using a variety of excitation k-space trajectories (46-48), coil designs

(49,50), acceleration factors (51), and RF pulse design formalisms (29,36,52-54).

Evaluation and prediction of SAR consequences of designed parallel RF excitation

pulses have commonly relied upon EM simulations using virtual human body models

(55,56) due to a lack of accurate means of measuring and predicting concomitant E

fields inside the human body. Recently, the use of pre-scan-based individualized body

models (57,58) have begun to be used in the EM simulations in order to estimate SAR

closely. However, it remains unclear whether it will be feasible to adapt the details of

simulated coil-subject setup in order to closely track what is happening or what will

happen to a subject during scan. For instance, concerns about using pre-scan-based

virtual body models and EM simulations to estimate actual SAR have been motivated

by the observation of significant SAR changes resulting from minor variations in body

model (59). In order to overcome these concerns, pre-scan-based SAR calibration

methods have been proposed (60-63). These methods enable accurate SAR predictions

specific to coil-subject setup and do not require assumptions about the subject or the

scanner setup. In addition to SAR prediction capability, an additional layer of system

7
monitoring in the parallel RF excitation chain has been implemented to ensure subject

safety, using either pick-up coils (64) or directional couplers (65,66). These additional

monitoring systems have been shown to detect system changes such as hardware

failure, system instability and patient position change which were undetectable with

previous RF monitoring systems.

Apart from SAR considerations, another emerging area of research in parallel

RF excitation involves the use of transmit array to shorten multidimensional RF

pulses, e.g. for tailored regional excitations. For instance, by using accelerated

multidimensional RF pulses, inner-volume excitations with high spatial selectivity

have been experimentally realized (31,67,68). Inner-volume excitation is expected to

reduce total signal acquisition time by reducing the extent of the required receive

FOV. Additionally, smaller inner-volume excitations tend to result in lower SAR for

small acceleration factors (51), which increases the importance of inner-volume

excitations at UHF.

In the light of all these developments, parallel RF excitation, with appropriate

SAR prediction and monitoring, is likely to continue to play an important role in the

future of MRI by the improving diagnostic value of UHF-MRI.

Research Problem Statement

Parallel RF excitation offers the flexibility to tailor both E field and B field

simultaneously. This thesis work centers around the general goal of achieving a

favorable balance between magnetic and electric fields for high-performance MRI.

8
Various approaches to increasing / tailoring B1+ field while decreasing E field in the

body are described.

First, we show how to use measurable subject-specific E field interactions of

individual transmit elements in order to decrease global SAR while achieving a target

B field distribution with high fidelity. The new concept of subject-specific SAR

prediction based on the measurable E field interactions is shown to facilitate the

subject-specific tailoring of B1+ profiles while managing global SAR.

Even though fully functional parallel RF excitation systems have been installed

worldwide, the majority of in vivo research on these systems to date has focused on

RF shimming due to its reduced computational and operational complexity as

compared to full parallel RF excitation. In this arena, we use subject-specific SAR

prediction to develop a maximum efficiency RF shimming method. This new RF

shimming approach aims to obtain the lowest possible net radiofrequency power

deposition into the subject for a given transverse magnetic field strength and

guarantees the global optimality of the resulting RF shimming coefficients.

As parallel excitation relies on simultaneous RF excitation from multiple coil

elements, interactions and coupling between coil elements and between coils and body

structures become more important than for single-element transmit systems. We show

how to extend the subject-specific global SAR prediction and monitoring method to

predict individual channel forward and reflected power for any RF excitation. By

9
using this new prediction capability, we design parallel excitation RF pulses meeting

strict MR scanner power handling limits.

The role of coil array geometry is well studied for parallel MR reception. In

this thesis we investigate the role of coil geometry on SAR and power requirements in

parallel RF excitation. As an example, we change the distance between coil array and

subject and investigate the SAR and power requirements as a function of the distance

between coil array and subject.

Parallel RF excitation offers the flexibility to decrease the RF pulse length by

undersampling excitation k-space trajectories. Shorter RF pulses are especially crucial

for multidimensional RF excitation, where the RF pulses tend to be longer than for

traditional (e.g. slice- or slab-selective) excitation. In this thesis we propose a method

that enables shorter multidimensional RF pulses for inner-volume excitation by sparse

selection of excitation k-space locations to be traversed. The method enables

excitation k-space accelerations beyond the limits of traditional parallel RF excitation

which is limited by the number of transmit coil elements.

Thesis Outline

This thesis consists of the introduction, five chapters describing each of the

research areas outlined above, a final conclusion and an appendix. The next chapter

(Chapter 1) is adapted from a manuscript published in the journal Magnetic Resonance

in Medicine and it explores the effects upon SAR of incorporating experimentally

measurable E field interactions into parallel RF transmission pulse design. Numerical

10
simulations and phantom experiments were used to demonstrate SAR reductions in

new RF pulse design strategy during parallel RF transmission while obtaining similar

excitation fidelity. Additionally, measured E field interactions were used to predict the

net RF power deposition of any parallel RF excitation pulse.

Chapter 2 is an extended version of an abstract presented in 2011 ISMRM

workshop on ultra-high field systems and applications at Lake Louise. It proposes a

new RF shimming algorithm, i.e. maximum efficiency RF shimming, which seeks to

obtain the lowest possible net radiofrequency power deposition into the subject for a

given transverse magnetic field strength. Experiments with volunteers were used to

demonstrate practicability of maximum efficiency RF shimming. Additionally,

quantitative SNR comparison of 3 T and 7 T imaging in the hip articular cartilage was

performed.

Chapter 3 investigates parallel RF excitation from an MR system perspective.

It describes the subject-specific proactive management of parallel RF excitation

aiming to obey RF power requirements of the hardware and SAR requirements for the

subject. Phantom experiments were used to compare RF pulses with and without

subject-specific power supervision.

Chapter 4 is an extended version of an abstract presented in 2009 at the

seventeenth annual meeting of the ISMRM in Hawaii. It investigates the SAR

behavior and the power requirements of parallel RF transmission as the distance

between transmit elements and the surface of the object is altered. Using numerical

11
simulations, various geometrical arrangements of coil elements around a cylindrical

and a spherical object are explored.

Chapter 5 is an extended version of an abstract presented in 2011 at the

nineteenth annual meeting of the ISMRM in Montreal. The chapter summarizes the

excitation k-space concept and the requirements for choice of an excitation k-space

trajectory. Sparse selection of the excitation k-space trajectory is demonstrated,

enabling inner-volume excitations with a UHF whole-body human scanner.

The concluding chapter summarizes the main topics discussed in the thesis and

outlines possible future work.

The appendix demonstrates graphical user interfaces (GUIs) developed in the

course of this work in order to increase the efficiency and the accuracy of parallel

transmit experiments. Workflows are explained with overlying screen captures of two

GUIs: one for RF shimming and one for fully parallel RF transmission pulse design.

12
CHAPTER 1: Specific Absorption Rate Benefits of Including Measured Electric

Field Interactions in Parallel Excitation Pulse Design

Deniz CM, Alon L, Brown R, Sodickson DK, and Zhu Y

Specific Absorption Rate Benefits of Including Measured Electric Field Interactions in

Parallel Excitation Pulse Design

Magnetic Resonance in Medicine 2012 (67): 164-174

Author contributions:

Cem Murat Deniz: Manuscript draft, study design, RF pulse design, sequence design,

data acquisition, data interpretation, literature research

Leeor Alon: FDTD simulations, power calibration system and software, manuscript

editing

Ryan Brown: MR coils and interface, manuscript editing

Daniel K. Sodickson: Study concept, manuscript editing

Yudong Zhu: Study concept, data interpretation, manuscript editing

13
Peer reviewed abstracts from the chapter:

Deniz CM, ALon L, Brown R, Fautz H-P, Sodickson DK, and Zhu Y

Parallel RF Pulse Design with Subject-Specific Global SAR Supervision

In Proceedings of the 19th Scientific Meeting, International Society for Magnetic

Resonance in Medicine, Montreal, Canada. page 210, 2011.

Deniz CM, ALon L, Brown R, Fautz H-P, Sodickson DK, and Zhu Y

Real Time RF Power Prediction of Parallel Transmission RF Pulse Design at 7T

In Proceedings of the 18th Scientific Meeting, International Society for Magnetic

Resonance in Medicine, Stockholm, Sweden. page 1454, 2010

Deniz CM, Alon L, Lattanzi R, Sodickson DK, and Zhu Y

SAR Benefits of Including E-Field Interactions in Parallel RF Pulse Design

In Proceedings of the 18th Scientific Meeting, International Society for Magnetic

Resonance in Medicine, Stockholm, Sweden. page 4930, 2010.

14
 1.1 Abstract

Specific absorption rate management and excitation fidelity are key aspects of

radio frequency pulse design for parallel transmission at ultra high magnetic field

strength. The design of radio frequency pulses for multiple channels is often based on

the solution of regularized least squares optimization problems for which a

regularization term is typically selected to control the integrated or peak pulse

waveform amplitude. Unlike for single channel transmission, the specific absorption

rate of parallel transmission is significantly influenced by interferences between the

electric fields associated with the individual transmission elements, which a

conventional regularization term does not take into account. This work explores the

effects upon specific absorption rate of incorporating experimentally measurable

electric field interactions into parallel transmission pulse design. Results of numerical

simulations and phantom experiments show that the global specific absorption rate

during parallel transmission decreases when electric field interactions are incorporated

into pulse design optimization. The results also show that knowledge of electric field

interactions enables robust prediction of the net power delivered to the sample or

subject by parallel radio frequency pulses before they are played out on a scanner.

1.2 Introduction

Multidimensional spatially selective excitation pulses are used to tailor

volumetric spin excitation (69). Volumetric spin manipulations can serve as a tool to

reduce susceptibility artifacts (70), to improve spatial resolution via inner volume

15
selection (71), and, especially at high magnetic field strength, to compensate for radio

frequency (RF) field inhomogeneity (27). However, RF pulse durations are generally

on the order of transverse magnetization decay ( T 2* ) which limits the practical

application of multidimensional excitations.

Parallel excitation using multiple transmit channels (28,29) has been shown to

allow significant acceleration of multidimensional RF pulses, bringing them within

range of practical applications. There are representative methods for the design of

accelerated multidimensional spatially selective RF pulses in the literature (28,29,34).

However, reduction in RF pulse length often results in increased power deposition in

tissue (72), defined as specific absorption rate (SAR).

SAR behavior in parallel RF transmission was first addressed by Zhu (29),

who proposed a pulse optimization approach in which the degrees of freedom

available in parallel transmission are used not only to achieve a target excitation

profile but also to minimize SAR, expressed as a quadratic function involving the RF

pulse waveforms and a characterization of electric field-induced RF loss. In the

absence of experimental knowledge of transmit coil electric fields, simulations have

been used to analyze / improve the SAR behavior of RF pulses with a variety of

excitation k-space trajectories (46,47,73), coil designs (49,50) acceleration factors

(51) and alternative optimization formalisms (36,74). Recently, an in vivo SAR

calibration method (60) for parallel RF transmission was introduced which enables

direct measurement of the electric field correlations required for accurate SAR

16
prediction and control. With this method, measurements of forward and reflected

power corresponding to a set of predefined parallel RF pulses are used to estimate the

power correlation matrix in a rapid and subject-specific manner. This correlation

matrix may then be used to predict the true RF energy delivered to the imaged sample

or body by any set of parallel pulses, without reliance on simulations or other

assumptions about body geometry, tissue properties, or the coil system.

This work represents an initial effort to incorporate a truly subject-specific

SAR prediction model, made possible by the experimentally calibrated power

correlation matrix, directly into parallel RF transmission pulse design. In this work,

regularization-based methods proposed by Grissom et al. (34) and Xu et al. (40) were

modified to capture global SAR behavior with a new regularization term for small and

large tip angle regimes, respectively. Furthermore, the use of the calibrated power

correlation matrix inside a regularization term extends the SAR-optimal parallel RF

pulse design approach proposed by Zhu (29) for echo-planar excitation k-space

trajectories to arbitrary trajectories and large flip angles.

Simulations of various coil-sample configurations with distinct power

correlation characteristics were used to evaluate the effects on global SAR realized by

employing the power correlation matrix in RF pulse design. These results were

compared to those obtained with RF pulses designed using the conventional integrated

regularization that disregards electric field interactions. Experimental investigations

were further conducted to investigate the feasibility of subject-specific global SAR

17
prediction and proactive management. This involved the use of both the calibration

system (60,61) for measuring the power correlation matrix and the present method that

accounts for electric field interference effects in parallel RF pulse design. Finally,

global SAR predictions were compared to actual power measurements in example

cases of conventional and proposed SAR-optimized RF pulse designs.

1.3 Materials and Methods

1.3.1 Electromagnetic Simulations

Numerical simulations were used to analyze SAR generated from parallel RF

pulses with and without electric field interactions incorporated in the pulse design. The

electric and magnetic fields of four-element transmit array were simulated using the

finite difference time domain method (xFDTD 6.3, REMCOM, State College, PA) at 7

T (297.2 MHz), using mesh data representing both a homogeneous rectangular water

phantom and a human body model.

Three different coil array configurations and relative element positions with

respect to the imaged objects are illustrated in Figure 1.1. Four identical square array

elements with 7 cm length were placed 1 cm above the object or body surface. In two

configurations (Figure 1.1a,b), coils were partially overlapped to imitate a common

practice to reduce inductive coil coupling, although inductive coupling was not present

in these simulations. Overlapping of the elements was achieved without artificial

short-circuits by offsetting the elements slightly with respect to each other in a

direction perpendicular to the plane of the coils. The simulated rectangular water

18
phantom had dimensions 24 x 20 x 24 cm3 and a uniform conductivity, σ, of 0.6 S / m

and relative permittivity, εr, of 80. The simulated HUGO human body model (Figure

1.1d) had heterogeneous dielectric properties and tissue density. Both objects were

defined on a grid of 5 x 5 x 5 mm3 voxel size. Each array element was driven by an

ideal 1-ampere current while keeping the other elements dormant (and therefore

eliminating inductive coupling between transmit elements). Steady-state electric

fields, El(r) (V / m / A), and magnetic fields, Bl(r) (T / A), for each transmit element,

l, were computed for all spatial locations, r, inside the object. The transmit sensitivity

map of the lth element, Sl(r), was calculated as Sl(r) = B1+,l(r) = (Bl,x(r) + i Bl,y(r)) / 2

(75). Figure 1.1e,f shows the phase (e) and amplitude (f) of the B1+ field of all coils in

setup C for the water phantom.

19
Figure 1.1 FDTD simulation setup. Setup A, B and C represent different coil configurations
which are used on a water block phantom (a, b, c) and a human mesh (d). e and f show the
phase and the magnitude of the B1+ map for setup C in the water phantom.

1.3.2 Global Specific Absorption Rate Calculations

RF power deposition into the object can be calculated with knowledge of the

unit-drive steady state electric fields of all transmit elements as well as the electrical

properties of the object and the designed pulse waveforms. Power deposition, P, of

parallel transmit arrays at location r and at each time instant pΔt can be calculated as:

20
  (r ) 2
P (r , pt )  E (r , p  t ) 2
[1.1]
2

L
where σ is the electrical conductivity, E(r, pt )   bl ( pt )el (r ) is the superposition
l 1

of the unit-current electric fields el of the L transmit elements multiplied by the

driving RF pulse waveforms bl, and p is an integer index indicating time in multiples

of the waveform sampling interval t . As shown in Refs. (29,72), total RF power

deposition into the object at any time instant pΔt can be calculated by taking the

following volume integral over the object:

 ( pt )   P(r, pt )dv  b Hpt Φb pt [1.2]

V

T T
where b pt  b1, pt  b L , pt  is the concatenation of the RF pulse waveforms,

H
denotes the matrix transpose, denotes the complex conjugate transpose, and Φ

defining the L x L positive definite Hermitian power correlation matrix with (i, j)-th

element is given by:

1
i , j 
2V  (r )E*i (r, pt )  E j (r, pt )dv [1.3]

where * indicates complex conjugation. Global SAR (W / kg) can be calculated as the

average total RF power deposition into the object divided by the object mass m:

Nt 1
1
 
mT
 t ( pt )
p 0
[1.4]

where Nt is the number of time samples and T is the RF pulse length.

21
 1.3.3 Parallel Excitation RF Pulse Design

The spatial domain parallel RF pulse design method (34) and linear class large-

tip-angle (LCLTA) method (40) were used to design small-tip-angle (STA) and large-

tip-angle (LTA) RF pulses, respectively.

Using the linearization of the Bloch equations within the STA regime (69) and

neglecting relaxation terms, the transverse magnetization M (r, T ) produced by the

parallel RF pulse at position r and time T can be expressed as:

L T
M (r, T )  i M 0 (r ) Sl (r)  bl (t )eiB0 (r )(t T ) eirk (t ) dt [1.5]
l 1 0

where  is the gyromagnetic ratio, M0(r) is the equilibrium magnetization at spatial

position r, L is the number of transmit coils with sensitivity patterns Sl(r), ΔB0(r) is the

local off-resonance field map, bl(t) is the RF pulse waveform of coil l, and

T
k (t )    G ( )d is the excitation k-space trajectory, defined as the time reversed
t

integration of the gradient waveforms G( ) (69). By discretizing in time to Nt samples

and in space to Ns positions, and concatenating matrices and vectors in Eq. [1.5] along

the coil dimension, as in Grissom, et al. (34), RF pulses for parallel excitation can be

calculated to produce a desired transverse magnetization profile vector mdes using a

selected k-space trajectory by solving:

2
bˆ full  arg min{ A fullbfull  m des 2  R(bfull )} [1.6]
bfull

22
where Afull   A1  AL  in which Al is a Ns x Nt system matrix with elements

aij  itM 0  ri  Sl  ri  e , b full  b1  b L  is a concatenation of the

iB0 ( ri )( t j T ) iri k ( t j ) T
e

RF pulse waveforms of L coils and R(bfull) is a general parameter for regularization

term which will be explained in detail at the end of this section.

When "linear class" assumptions (76) about excitation k-space are satisfied and

relaxation effects are neglected, the flip angle distribution,  (r) , of parallel RF pulses

in the LTA regime may be expressed, following Xu et al. (40), as:

L T
 (r )    S (r )  bl* (t )eiB (r )(t T ) eirk (t ) dt
*
l
0
[1.7]
l 1 0

By discretizing in time and space and concatenating matrices and vectors,

Eq. [1.7] can be expressed in matrix form as θ  Cfullb*full , where Cfull  C1  CL  ,

 iB0 ( ri )( t j T )  iri k ( t j )
and Cl is a Ns x Nt system matrix with elements cij  tSl* (r j )e e .

Given a desired flip angle distribution, θdes, and a chosen k-space trajectory, parallel

pulse waveforms can be calculated by solving the minimization problem:

2
bˆ full  arg min{ Cfullb*full  θdes  R(b*full )} [1.8]
2
b*full

Regularization terms R in both STA (Eq. [1.6]) and LTA (Eq. [1.8]) designs

can be used to protect against an ill-conditioned matrix inversion and to control the

integrated or peak RF pulse waveform. One widely used approach in RF pulse design

is Tikhonov regularization R (b full )   b full

H
b full , where β is used to tradeoff excitation

23
profile error against the integrated RF pulse waveform amplitude square (34,40,77).

Unlike in single-channel transmission systems, however, controlling the integrated RF

pulse waveform amplitude may not be an effective way to minimize SAR in parallel

pulse design, where SAR may be significantly influenced by electric field interactions

inside the object (29). We propose to use the following regularization term in STA RF

pulse design which incorporates the full constructive and destructive electric field

interferences into RF pulse design that are ignored by conventional regularization

terms:

RSTA (bfull )   bfull

H
Φfullbfull [1.9]

In Eq. [1.9],

Φ 0
Φfull 
   [1.10]

 0 Φ  N LxN L
t t

where Φ is the power correlation matrix defined in Eq. [1.3], and the β parameter is

now used to trade off excitation error against true global SAR. Similarly, the

regularization term for LCLTA RF pulse design can be defined as:

RLCLTA (bfull )   bfull

H
Φ*fullbfull [1.11]

Eqs. [1.6] and [1.8], now with the revised regularization terms, can be efficiently

solved with conjugate gradient methods.

24
 1.3.4 RF Shimming

As a special case of parallel RF transmission, the RF shimming method (19,23)

can be used to correct B1+ inhomogeneities by time-independent control of relative

amplitude and phase of individual transmit elements which share a common RF

waveform. The desired B1+ distribution Sdes, is generated by applying a set of complex

weights, wl, to the individual transmit elements such that the following equality holds

at every spatial location r inside the selected shim volume:

L
Sdes (r)   wl Sl (r) [1.12]
l 1

By discretizing spatial locations once again into Ns samples, Eq. [1.12] can be

written in the matrix form:

 Sdes (r1 )   S1 (r1 ) S 2 (r1 )  S L (r1 )   w1 

 S (r )   S (r ) S (r )  S (r )   w 
 des 2    1 2 2 2 L 2  2
[1.13]
          
    
 S
 des (rs ) 
  S1 (rs ) S2 
 
(rs )  S L (rs )   wL 
 
S des S w

Regularized least-squares solution for the desired RF shim coefficients can be

efficiently obtained by solving:

2
ˆ  arg min{ Sw  S des 2  R(w )}
w [1.14]
w

where R(w) is the regularization term which can be defined either as R (w )   w H w

(penalizing waveform amplification) or as R (w )   w H Φw (penalizing true global

SAR directly) following the discussion in the previous section.

25
 1.3.5 Simulated RF Pulse Designs

A constant density inward spiral trajectory (Figure 1.2a) was used to cover

excitation k-space for simulated pulse designs, using the following gradient design

parameters: maximum amplitude 40 mT / m, maximum slew rate = 150 mT / m / s,

and sampling period = 10 μs. The k-space trajectory was chosen to achieve a spatial

resolution of 10 mm. 12 spiral turns were used for the unaccelerated pulse trajectory,

and acceleration by a factor of R was achieved by undersampling radially, yielding

12/R turns. Based on the parameters listed above, RF pulse lengths of 5 ms for R = 1

and 2.5 ms for R = 2 were in effect.

Figure 1.2 Excitation k-space and desired profile. a: Constant density spiral-in excitation
k-space trajectory used in simulations (acceleration factor R = 1). b: Desired excitation profile.
c, d: Bloch equation simulation results for RF pulses designed with conventional (c) and
proposed (d) methods for setup C.

26
 2D transmit sensitivity maps of 5 mm resolution for all transmit elements were

extracted from 3D FDTD simulations after defining the slice of interest (1.5 cm and 4

cm below the surface for water phantom and virtual human mesh, respectively). For

RF shimming simulations, a desired coronal B1+ distribution, Sdes, with uniform

magnitude and zero phase inside the water phantom was selected. For parallel RF

pulse design, the desired excitation profile for the water phantom (Figure 1.2b) was

defined as a centrally located coronal disk of uniform flip angle and zero phase with

diameter equaling 15 cm. For the human mesh dataset, a 7.5 x 15 cm2 rectangle of

uniform flip angle and zero phase at the center of FOV was defined as the desired

excitation profile. Flip angles of 10° and 90° were chosen for STA and LTA RF pulse

designs, respectively.

RF pulses and RF shim weights were calculated from Eqs. [1.6], [1.8] and

[1.14] with conventional and proposed regularization terms, using custom code

developed in Matlab (version 7.9, MathWorks, Inc., Natick, MA, USA). Using the

transmit coil sensitivity profiles combined with the computed pulse waveforms / RF

shim weights, the net magnetic fields resulting from the parallel RF pulses / RF shim

weights were calculated. Subsequently, spinor domain-based Bloch equation

simulations described in Ref. (78) and developed by Hargreaves

(http://mrsrl.stanford.edu/~brian/blochsim/) were used to generate the excitation

profile of parallel RF pulses over a selected slice with 2.5 x 2.5 mm2 resolution.

Relaxation effects were ignored in Bloch simulations.

27
 To quantify excitation fidelity of a pulse design, the normalized

root-mean-square error (NRMSE) between the desired magnetization and the

magnetization profile obtained from Bloch simulation, mbl, was calculated as

mbl  mdes 2 / mdes 2 . For fair comparison of various pulse designs’ SAR

performance, the NRMSE’s of the designs were equalized using distinct heuristically

chosen regularization parameters. Excitation fidelity can be similarly quantified and

aligned for the RF shimming cases. At comparable fidelity, global SAR performance

of RF pulses or RF shim weights designed with different regularization terms were

then compared.

1.3.6 Experimental RF Pulse Designs

To evaluate the benefits of incorporating global SAR information into the RF

pulse design, experiments were performed on a Siemens whole body 7 T Magnetom

scanner (Erlangen, Germany) equipped with an eight-channel parallel transmit system.

An eight-channel stripline coil array was used for RF excitation and reception (Figure

1.3a). The striplines were mounted on an acrylic former with 27.9 cm diameter and

azimuthally separated by 45. The striplines were built on 15 x 4 x 1.3 cm3 teflon bars

with 14 x 2 cm2 conductive strips, 15 x 4 cm2 ground planes, and sidewalls with 1.3

cm height to reduce inter-element coupling. Two tuning capacitors of approximately

6.8 pF and 8.2 pF were inserted on opposing ends of each stripline to achieve

resonance at 297.2 MHz. The striplines were matched to 50 through a series

capacitor of approximately 2.2 pF while loaded with a 7.3-L cylindrical water

28
phantom with 15 cm diameter containing 1.25 g / L NiSO4.6H2O and 4 g / L NaCl

(σ = 0.7 S/m, εr = 80.6) (Figure 1.3b). Forward and reflected power readings of eight

channels at a sampling rate of 10 μs were obtained with a power sensor (NRP-Z11,

Rhode&Schwarz, Munich, Germany) connected to directional couplers at the output

of each RF amplifier via an RF switch (Dual 16 x 1 MUX, National Instruments,

Austin, TX, USA).

Figure 1.3 Phantom experiment setup. a: 8 channel transmit-receive coil array. b: Cylindrical
water phantom. c: B1+ amplitude map for each element of the array. d: B1+ phase map for each
element of the array.

B1+ calibration was performed following the method described in Ref. (79). In

order to obtain individual transmit channel B1+ profiles, non- or selective saturation

pulses on one channel at a time were used to produce a spatial-dependent flip angle

map. A reference image was obtained from selective excitation of all channels without

magnetization preparation. The reference image was used to obtain the cosine of the

29
flip angle map by dividing the saturated image. RF shimming was used to have

enough SNR throughout the reference image. B1+ magnitude maps (Figure 1.3c) in the

axial plane through the isocenter were obtained by processing data from 1.5 ms

rectangular saturation pulses followed by a multishot segmented spoiled turbo fast

low-angle shot (FLASH) imaging acquisition with 2 segments (segment repetition

time = 5 s). Relative B1+ phase distributions for different coils (Figure 1.3d) were

calculated from additional turbo FLASH scans using only one coil for excitation at a

time. The following imaging parameters were used: FOV = 240 x 240 mm2, echo time

(TE) = 1.97 ms, acquisition matrix = 128 x 128 and slice thickness = 8 mm. Total

acquisition time for B1+ profiles in all eight channels was 53 s. ΔB0 was measured

using the phase information from two gradient echo (GRE) images with different TE

values (TE1 / TE2 = 7.14 / 5.1 ms) and was incorporated into RF pulse design to

compensate for the phase accrual due to main magnetic field inhomogeneity (Figure

1.4b).

30
Figure 1.4 Additional inputs used in pulse design for the phantom experiments. a: Desired
excitation profile. b: Measured off-resonance map. c: Calibrated power correlation matrix of
the phantom-coil setup. d: Variable density spiral-in excitation k-space trajectory used in
experiments.

The subject-specific power correlation matrix Φ used for SAR prediction and

optimization was estimated using the automated Power Prediction and Monitoring

(PPM) technique described by Zhu and coworkers. (61). To accurately characterize the

field interference effects on SAR this technique estimates  by measuring in situ

individual channel forward and reflected power that correspond to the application of a

set of calibration RF pulses. By the law of conservation of energy, pfwd - prfl, gives

the net RF power delivered, which allows the assembling and solving of a set of Eq.

[1.2]-type linear equations but with the b’s as the coefficients and the entries of Φ as

the unknowns. The calibrated Φ matrix (Figure 1.4c) was used in parallel RF pulse

31
design via a regularization term (Eq. [1.9]) in order to trade off global SAR against

excitation fidelity.

The linear class LTA method (40) was used to design parallel RF pulses with a

90° target flip angle. The target excitation flip angle distribution θdes (Figure 1.4a) was

a homogenous 4 x 2 cm2 rectangular 2D profile blurred by convolving it with a

Gaussian kernel of full-width half-maximum (FWHM) = 1.2 cm to reduce ringing

artifacts in the resulting magnetization distribution. A variable density (80) inward

spiral trajectory (Figure 1.4d) was used to cover excitation k-space with the following

parameters: α = 2 (defines the amount of oversampling near the origin of the k-space),

sampling interval = 10 µs and duration = 7 ms (corresponding to 3-fold acceleration

with respect to a 21 ms non-accelerated RF pulse using constant rate spirals), in-plane

resolution = 3.78 mm, maximum gradient slew rate = 150 mT / m / s, and maximum

gradient amplitude = 40 mT / m.

Two different regularization terms, conventional and proposed, were used to

design parallel RF pulses. As in the simulations, in order to provide a fair comparison

of the global SAR effects of different regularization schemes, both NRMSE and

nominal flip angle were aligned between different parallel pulse designs. For fully

parallel transmission, this was achieved using different heuristically chosen

regularization parameters, β, and NRMSE was computed only inside the region where

the desired rectangular magnetization profile (Figure 1.4a) has flip angle values

greater than 0°, since accuracy of the B1 mapping algorithm diminishes for low flip

32
angles. Flip angle profiles of the designed RF pulses were measured using the

technique described earlier for B1+ map acquisition (specifically, designed parallel RF

pulses were played as saturation pulses followed by a multishot segmented turbo

FLASH acquisition with 4 segments). Imaging parameters were: FOV = 240 x 240

mm2 TE = 1.97 ms, acquisition matrix = 128 x 128, acquisition time = 40 s. Linearity

of the designed RF pulses and the parallel transmission system was assessed by

measuring the average flip angle over a range of transmit voltages. Actual global SAR

was experimentally measured using forward and reflected power readings during the

RF excitation period. In addition, expected power deposition into the phantom was

predicted by Eq. [1.2] and compared with the actual net power measurements.

1.4 Results

1.4.1 Water Phantom Simulations

For each of the experimental setups RF shimming with coefficients calculated

using the two different regularization terms were performed. Comparison of global

SAR was conducted at aligned NRMSE level of 0.99 for all experimental setups

(Figure 1.1a-c). For experimental setup A at the same NRMSE of 0.99 for example,

global SAR for RF shim weights were 0.15 W / kg with conventional and 0.1 W / kg

with proposed regularization terms. Including the global SAR knowledge into RF

shim weight design via proposed Φ-based regularization resulted in 34.9%, 9.7% and

27.5% decrease compared to conventional I-based regularization in average net power

deposition into the water phantom for experimental setups A, B and C, respectively.

33
 Bloch simulation results of parallel RF transmission pulses designed for

experimental setup C are shown at Figure 1.2c,d for conventional and proposed

regularization terms, respectively. The two RF design approaches achieved

comparable fidelity producing the desired flip angle distributions with the desired flip

angle profile (Figure 1.2b).

Figure 1.5a shows the percentage global SAR benefit (at fixed NRMSE) of

using the proposed SAR-minimizing method versus the conventional method for

different acceleration factors (R = 1 and 2) and RF pulse design methods. Both linear

class LTA and STA pulse designs that accounted for electric field interactions resulted

in lower global SAR than those that ignored electric field interactions. Calculated

power correlation matrices of three different simulations are shown in Figure 1.5c.

The greatest SAR benefits were observed for setup B. For setup C, global SAR

differences between RF design schemes were minor due to a highly-diagonal power

correlation matrix that resembles a scaled identity matrix. Table 1.1 shows that

incorporating the Φ in STA RF pulse calculations results in lower global SAR for

every coil arrangement without sacrificing excitation fidelity. Global SAR decrease

was more significant for the arrays in setup A and B, for which stronger electric field

interactions resulted in large variations among the diagonal and off-diagonal elements

of the Φ. Although the use of the Φ instead of I in pulse design increased the sum of

all channel RF current amplitude squared for most of the experiments, as indicated by

34
the values in the third row of Table 1.1, it did not result in higher global SAR. This

result indicates that electric field interferences play an important role in global SAR.

35
Figure 1.5 Global SAR when the power correlation matrix is incorporated into parallel RF
pulse design. Values are reported as the percent improvement in global SAR with respect to
using a conventional regularization term for the same flip angle and excitation fidelity. Results
of different RF pulse design methods and acceleration factors for three different transmit array
setups are shown for the water phantom (a) and the human model (b). Calculated power
correlation matrices for three different array setups are shown for the water phantom (c) and
the human model (d).

36
 Table 1.1 Comparison of STA parallel RF pulse behavior in a simulation involving imaging of a water phantom with various transmit
array configurations and acceleration factors.

Setup A Setup B Setup C

R =1 R =2 R =1 R =2 R =1 R =2

Φ I Φ I Φ I Φ I Φ I Φ I

SAR (W/kg) 0.097 0.109 0.619 0.661 0.100 0.123 0.667 0.723 0.115 0.116 0.854 0.866

NRMSE 0.021 0.021 0.020 0.020 0.020 0.020 0.020 0.020 0.022 0.022 0.021 0.021
37

Total RF (A2) 1584 1313 5028 4319 2421 1533 8033 5429 1066 1048 3959 3954

% SAR decrease 10.7 6.4 18.2 7.8 1.4 1.3

The conventional RF pulse design approach is denoted by I (indicating that the power correlation matrix was replaced by the identity
matrix in the regularization term), and the proposed approach is denoted by Φ.
SAR = specific absorption rate in watts per kilogram. NRMSE = normalized root mean square error between desired and achieved
magnetization profile. Total RF = Integrated RF current amplitude (in amperes) square over all channels.
 1.4.2 Human Mesh Simulations

In simulations using the human mesh model, larger global SAR benefits were

observed compared to water phantom simulations in all transmit array configurations

(Figure 1.5b). Both LCLTA and STA pulse design resulted in lower global SAR when

electric field interactions were incorporated. As was the case for water phantom

simulations, the greatest global SAR benefits were observed for setup B in which Φ

deviates significantly from a scaled identity matrix (Figure 1.5d). For setup C, the

global SAR benefit was more accentuated in the human body model than in the water

phantom, which can be explained in part by an increased variation amongst the

diagonal elements of Φ for the inhomogeneous human mesh compared with the

homogeneous water phantom.

1.4.3 Experiments

Prior to applying calculated RF pulses on the scanner, Bloch simulation results

of RF pulses with different regularization terms were used to align the NRMSE of the

magnetization distributions. Figure 1.6a,b represent the Bloch-simulated results of flip

angle profiles calculated with conventional and proposed regularization terms,

respectively. Both approaches resulted in NRMSE of 0.0319. The amplitude of the

designed RF pulse waveform in one of the channels is shown in Figure 1.7a. Notice

that incorporating Φ into RF pulse design resulted in local changes in the RF pulse

waveform to improve the SAR management of the pulse as a whole while preserving

the excitation fidelity.

38
Figure 1.6 Experimental results. a, b: Bloch equation simulation results for NRMSE-aligned
RF pulses calculated with conventional (a) and proposed (b) regularization terms. c: Shimmed
reference image used for transmit sensitivity mapping. d, e: MR images obtained using
parallel RF pulses designed with conventional (d) and proposed (e) approaches as saturation
pulses. f,g,h,i: Flip angle (f,g) and phase (h,i) maps of the conventional RF design with
transmit voltage 135V (f,h) and the proposed RF design with transmit voltage 130V (g,i).

39
 To verify the flip angle profile in actual experiment, the calculated RF pulses

were used as a saturation pulse. An RF shimmed reference image was obtained from

selective excitation of all channels without magnetization preparation (Figure 1.6c).

Table 2 shows that the mean flip angle and NRMSE for the conventional and proposed

regularization terms were aligned over a range of transmit voltages (110 V - 150 V).

For the given input transmit voltage range, the LCLTA parallel transmit RF pulse

design resulted in a linear response of the system (Figure 1.7b). This validated the

linear class assumption used in the pulse calculation and also confirmed the linearity

of the system within the given input voltage range. Saturation effects resulting from

the designed 90° RF pulses can be seen in Figure 1.6d,e for 135V and 130V transmit

voltages of RF designs with conventional and proposed regularization terms,

respectively. Rectangular black regions within the phantom correspond to positions

where 90° flip angle was produced by the saturation pulse (i.e., the designed 90°

excitation pulse). Figure 1.6f,g show the flip angle maps of the designed RF pulses

which were extracted from the ratio between reference and saturation images. It is

clear from the figure that there is a good agreement between Bloch simulations and

experimental results. For the measured flip angle maps, calculated mean flip

angle / NRMSE ratios were 88.67 / 0.116 and 88.14 / 0.116 for RF pulses designed

with conventional (transmit voltage 135 V) and proposed (transmit voltage 130 V)

regularization terms, respectively.

40
Figure 1.7 RF pulse waveforms and RF net power a: Amplitude of the designed RF pulse
waveforms in one of the transmit channels. b: Linearity of the system and RF pulse design
process with respect to transmit voltage. c,d: Measured and predicted net power (in kW) of the
transmit array with a 7 ms LCLTA RF pulse. c: Conventional RF design method with transmit
voltage 135V. d: Proposed RF design method with transmit voltage 130V.

The forward and the reflected power of the designed RF pulses were measured

using the power meter. The average net power measurements from the RF power

amplifiers and predicted average net power deposition calculated according to Eq.

[1.2] with Φ and calculated pulse waveforms are shown in Table 1.2 for various

transmit voltages. Measured / predicted average net power for RF designs were 212.16

/ 237.81 W with conventional and 197.01 / 212.07 W with proposed regularization

41
terms. Including global SAR model into RF pulse design via regularization resulted in

~7.4% decrease in average net power dissipation for comparable average flip angle

and NRMSE. Figure 1.7c,d shows the net power measurements and predictions of

designed RF pulses. The net power measurements tend to be somewhat lower than the

predicted power values. This was attributed, in part, to the timing offset between

power measurement and the scanner's RF pulse update (every 10μs) and the temporal

averaging involved in power measurement.

42
 Table 1.2 Experimental parallel RF pulse behavior

Proposed Parallel RF Pulse Design Conventional Parallel RF Pulse Design

Transmit Voltage (V) 110 115 120 125 130 135 140 120 125 130 135 140 145 150

Mean Flip Angle 75.1 78.3 81.7 84.9 88.1 91.4 94.4 79.3 82.4 85.5 88.7 91.8 94.8 97.8

NRMSE 0.18 0.15 0.13 0.12 0.12 0.13 0.14 0.15 0.13 0.12 0.12 0.13 0.15 0.17
43

Measured Power (W) 142 155 168 182 197 213 228 168 183 197 212 228 246 261

Estimated Power (W) 152 166 181 196 212 229 246 188 204 221 238 256 274 294

Comparison of conventional and proposed parallel RF pulse design methods in terms of mean flip angle inside the region where the
desired magnetization profile has greater than 0° flip angle, normalized root mean square error NRMSE between desired and actual
magnetization distributions, measured power and predicted power.
 1.5 Discussion

In this work, we have demonstrated the reduction in RF power deposition by

incorporating measured electric field interactions into pulse design for parallel

excitation. This work represents an original effort to incorporate a truly

subject-specific SAR prediction model into parallel RF transmission pulse designs and

to further validate the designs in MR experiments. The use of an

experimentally-calibrated global SAR prediction model as an explicit regularization

term gives the user the flexibility to tradeoff SAR and excitation fidelity. By contrast,

strict constraint-based optimization approaches (29,36) (realized, for example, via

Lagrange multipliers) guarantee that the specified constraints, e.g. on excitation

fidelity (29) or on SAR (36), are always met. Within the parameter space allowed by

the constraints, the solution that minimizes the remaining optimization goals is then

selected. A strict-constraint-based approach is in fact possible using our experimental

global-SAR calibration method, and this could be valuable in ensuring that patient-

specific global SAR is always maintained below a target value. Our current

regularization-based approach, however, is applicable to a broad range of pulse design

and optimization problems. It is fully applicable to the design of LTA parallel RF

pulses when the linear class assumption for the k-space trajectory holds. Linear class

assumption restrictions on LTA parallel RF pulse design can be overcome by using

regularization terms in design procedures which accept arbitrary excitation k-space

44
trajectories, such as the additive angle method (42) and the optimal control approach

(41).

It is possible to integrate explicit global SAR management into many of the

existing parallel excitation pulse design methods. The power correlation matrix-based

SAR tracking metric is simple in form (quadratic) and behave nicely (convex

functions), making integration straightforward. In the present work for example, the

SAR-tracking quadratic term introduced simply replaces a regularization term existing

already in the design method, resulting in minimum additional numerical burden. We

expect the impact of introducing the term on complexity to be small even in the more

involved optimal control design case (41), as the new composite metric to be

minimized remains convex (sum of quadratic functions).

Global SAR benefits of incorporating electric field interactions into RF pulse

design were validated in phantom and human body model simulations and in phantom

experiments. As was reflected in high NRMSE's for RF shimming simulations,

magnetization profiles for our RF shimming simulations were substantially different

from the desired magnetization profile. This deficiency could be reduced to some

extent by increasing the number of transmit elements or by using a target profile

phase-relaxed RF shimming algorithm. However, calculation of phase-relaxed RF

shimming coefficients (32) may stall in local optimal solutions, which would

complicate a fair comparison of the effects of different regularization schemes.

45
 A power calibration system (61) was used to obtain subject-specific

information about electric field interactions and to predict RF power deposition for

arbitrary parallel RF transmission pulses. While results showed notable agreement

between power deposition predictions and experimental measurements, a significant

limitation of our current experimental setup can be identified. Given that the RF power

sensor in the present system is located at the output of the power amplifiers, the

system overestimates SAR in the imaged subject. At this location, the sensor’s power

readings include a significant cable loss component - a separate measurement

indicated that RF loss in the coaxial cables connecting the power amplifiers to the

coils accounts for over 50% of total RF power delivered by the RF power amplifiers.

This significantly impacts the structure of the calibrated power correlation matrix Φ,

making the entries on the diagonal dominate (and making Φ resemble a scaled identity

matrix). This setup can be improved and more significant SAR reduction can be

observed (81) by moving the power sensing location close to the transmit coils. The

improvement however, can be a challenge to implement as it requires a significant

portion of the power measurement instrumentation to be compatible with the 7 T

magnetic field.

1.6 Acknowledgements for Chapter 1

I would like to thank Dr. Hans-Peter Fautz from Siemens Medical Solutions in

Erlangen, Germany for collaboration on the flip angle mapping sequence. Dr. Graham

Wiggins is acknowledged for discussions on development of the coil array and

46
interface used for parallel transmission. I would also thank to Dr. Riccardo Lattanzi

for helpful discussions on simulations.

47
 CHAPTER 2: Maximum Efficiency RF Shimming: Theory and Initial

Application for Hip Imaging at 7 Tesla

Deniz CM, Alon L, Brown R, Sodickson DK, and Zhu Y

Maximum Efficiency RF Shimming: Theory and Initial Application for Hip Imaging at

7T

Manuscript in progress

Author contributions:

Cem Murat Deniz: Manuscript draft, study design, RF shimming software, data

acquisition, data analysis, data interpretation, literature research

Ryan Brown: MR coils and interface, data acquisition, manuscript editing

Riccardo Lattanzi: Study concept, SNR analysis software, manuscript editing

Leeor Alon: Power calibration system and software, manuscript editing

Daniel K. Sodickson: Study concept, manuscript editing

Yudong Zhu: Study concept, data interpretation, manuscript editing

48
Peer reviewed abstracts from the chapter:

Deniz CM, Brown R, Lattanzi R, Alon L, Sodickson DK, and Zhu Y

Maximum Efficiency RF Shimming

In Proceedings of the 20th Scientific Meeting, International Society for Magnetic

Resonance in Medicine, Melbourne, Australia, page 3479, 2012.

Deniz CM, Brown R, Alon L, Sodickson DK, Zhu Y, and Lattanzi R

MRI of the Hip at 7T Using RF Shimming with 4-Channel Excitation

In Proceedings of the ISMRM Workshop on Ultra-High Field Systems &

Applications: 7T & Beyond: Progress, Pitfalls & Potential, Lake Louise, Alberta,

Canada, 2011.

49
 2.1 Abstract

Radiofrequency shimming with multiple channel excitation has been proposed

to increase the transverse magnetic field uniformity and reduce specific absorption rate

at high magnetic field strengths (≥ 7 Tesla) where high-frequency effects can make

traditional single channel volume coils unsuitable for transmission. In the case of deep

anatomic regions and power-demanding pulse sequences, optimization of transmit

efficiency may be a more critical requirement than homogeneity per se. This work

introduces a novel method to maximize transmit efficiency using multiple channel

excitation and radiofrequency shimming. Shimming weights are calculated in order to

obtain the lowest possible net radiofrequency power deposition into the subject for a

given transverse magnetic field strength. The method was demonstrated in imaging

studies of articular cartilage of the hip joint at 7 Tesla. We show that the new

radiofrequency shimming method can enable reduction in power deposition while

maintaining an average flip angle or adiabatic condition in the hip cartilage. Building

upon the improved shimming, we further show the signal-to-noise ratio in hip cartilage

at 7 Tesla can be substantially greater than that at 3 Tesla, illustrating the potential

benefits of high field hip imaging.

2.2 Introduction

The promise of improved morphological and functional imaging due to higher

signal-to-noise ratio has motivated the pursuit of ultra-high field MRI (≥ 7 T).

However, ultra-high field MRI is challenging, due to inhomogeneities of the

50
transverse radiofrequency magnetic field (B1+), which compromise image quality, and

specific absorption rate constraints, which limits the strength of MR excitation at

depth.

RF shimming (19,23) and parallel excitation (28,29) techniques using multiple

transmit channels have been shown to allow significant reductions in B1+ field

inhomogeneities. While RF shimming has limited capability to provide homogeneous

B1+ over large regions, local phase-only RF shimming (24) which aims for B1+ phase

coherence / constructive interference in small target regions, such as the prostate, have

been shown to provide reasonable homogeneity and increase in B1+ field for a given

transmit RF power. Depending on the particular application in question, different RF

shimming methods have been proposed which, for example target B1+ magnitude and

phase homogeneity (19), target B1+ magnitude homogeneity without regard for B1+

phase (82), tradeoff B1+ homogeneity for SAR minimization (72,83,84), pursue B1+

homogeneity by sequential application of two different RF shim weights (85), or

facilitate the adiabatic condition of the RF pulse (86,87).

In this paper a maximum efficiency RF shimming approach is presented,

which, given any coil-subject setup, calculates a set of RF shim weights that

maximizes B1+ strength for any given level of RF power deposition into the subject.

Field interference is the fundamental principle underpinning multi-port or array coil

transmission. While desirable B1+ interference patterns are sought after for managing

the excitation profile (as is the case in shimming for B1+ homogeneity or in

51
full-fledged parallel RF transmission), concomitant electric field interference impacts

RF power deposition. The present approach, which not only accounts for transmit

sensitivity patterns but also for subject-specific electric field interference effects on

global SAR and net RF power, has a unique advantage over existing methods that aim

at increasing power efficiency by performing guided manipulation of the B1+ field

only.

The feasibility of the new RF shimming method was demonstrated in imaging

studies of hip articular cartilage, an application that is clinically significant yet

technically challenging. Accurate assessment of hip anatomy and function has become

a critical concern in recent years, after it was shown that the success of surgical

procedures aimed at delaying or preventing hip osteoarthritis by correcting the bony

abnormalities associated with femoroacetabular impingement (88) depends on the

absence of irreversible degenerative changes in the hip cartilage (89). However the

evaluation of the hip cartilage is currently a challenge even at 3 T and normally

requires administration of exogenous contrast agent to capture areas of abnormalities

with sufficient contrast and SNR (90).

Maximum efficiency RF shimming can be an enabling technical solution to

advance high field hip imaging and benefit clinical practice. Hip imaging represents an

application that targets a deep anatomical region and demands good SNR.

Conventional RF shimming at ultra high field tends to face difficulties managing RF

52
power. Furthermore, as the thickness of the hip articular cartilage, which covers the

femoral head and the acetabulum, ranges approximately from 1.5 to 5 mm (91)

decreased power deposition could be leveraged to increase spatial resolution and/or

SNR by allowing increases in the achievable flip angle and / or the number of

refocusing pulses.

Following a description of the maximum efficiency RF shimming method,

experimental calibration of the inputs for the proposed RF shimming method is

explained. Using experimental power and B1+ map measurements, maximum

efficiency RF shimming was compared with non-targeted unit RF shimming using

different RF pulse types (sinc and adiabatic). In addition, simulations incorporating

subject-specific calibration data, including individual channel B1+ maps and transmit

power correlation matrix, were used to compare the proposed method with three

existing RF shimming methods. Finally, in order to illustrate the gains at high field

strength in anatomical regions of significant clinical interest, quantitative SNR

comparisons in hip articular cartilage between 3 T and 7 T were conducted.

2.3 Materials and Methods

2.3.1 Maximum Efficiency RF Shimming

RF shimming (19,23) was proposed to correct B1+ inhomogeneities by

optimizing the relative amplitudes and phases of multiple transmit elements driven

53
with a common RF waveform. Flexibility to control the relative amplitude and phase

of the individual transmit elements can also be exploited to increase transmit

efficiency. This section describes the maximum efficiency RF shimming method

which aims to maximize B1+ strength for any given level of RF power deposition into

the subject.

In order to obtain the complex-valued RF shim weights that correspond to the

amplitude and phase modulation associated with maximum transmit efficiency, we use

a transmit efficiency metric defined as B1+ magnitude squared per unit dissipated

power, following earlier work by Zhu et al. (92). Using the superposition principle of

linear systems, the net B1+ and electric field at each spatial location r and at each time

t can be defined as:

N N
B1 (r)   w( n)b( n) (r) and E(r)   w( n )e( n ) (r) [2.1]
n 1 n 1

where N is the number of transmit elements, and the weights w(n) specify the amplitude

and phase modulation of the driving RF current waveform in the nth channel of a

transmit array. The complex-valued b ( n ) (r ) and e( n ) (r ) represent, respectively, the

B1+ and electric fields corresponding to unit weighting on the nth channel and zero

weights on the others. Choosing an ROI, the values of B1+ at all M spatial locations

included in the ROI can be combined in matrix form as B1  Cw where C is an M x N

matrix with C mn  b ( n ) (rm ) . The average B1+ squared in the ROI can be expressed as:

2
average B1  w H Γw [2.2]

54
where the B1+ correlation matrix Γ  M 1C H C , and H denotes the conjugate transpose.

Here, Γ is an N x N positive-definite complex Hermitian matrix.

The total RF power deposited by the parallel transmit array into the object at

time t can be calculated by taking the following volume integral over the object and

substituting the linear superposition of the electric fields from Eq. [2.1] :

 (r)
P
2
E(r ) 2 dv  w H Φw [2.3]
v
2

where σ is the electrical conductivity, and Φ defines the N x N positive-definite

Hermitian power correlation matrix whose (i, j)-th element is given by:

1
2 v
i, j   (r )e(i ) (r )*  e( j ) (r )dv [2.4]

and * indicates complex conjugate.

A rapid calibration scheme to measure experimentally the elements of the

power correlation matrix in Eq. [2.4] has recently been described (60,61,66). Once Φ

is known, RF power dissipation (Eq. [2.3]) (66) can be determined for any possible set

of RF shimming weights w, allowing prediction of the global SAR consequences of

any imaging sequence (84). For cases in which radiative losses and coil losses are

significant, the resulting predicted power dissipation is an upper bound on overall RF

power deposition in the subject; for the more common situation in which body losses

are the dominant contribution, the predicted power dissipation more closely tracks

global SAR in body tissues. Ref. (66) addresses these considerations in detail, as well

as identifying potential improvements to the calibration process. The entries of the

55
matrix C can be measured with any B1+ mapping technique, allowing the evaluation of


Γ. Using the derived expressions for the average B1 squared and the total RF power

deposition for any RF shim weights w, the transmit efficiency metric can be defined as

(92):

w H Γw
 H [2.5]
w Φw

By streamlining the power calibration and B1+ mapping, the efficiency metric,

η, can be practically evaluated in vivo. In practice, it is convenient to use units of μT

squared per Watt. In the conventional single channel case, Γ and Φ reduce to scalars,

and the metric captures B1+ squared per unit power, compatible with existing practice.

In the multi-channel transmission case, different w's correspond to different efficiency

in general. In addition, given the bilinear form in both numerator and denominator, η

is independent of any overall scale factor in the RF shimming weights (and therefore

independent of any overall changes in transmit voltage).

Depending on the RF shimming coefficients, a given transmit array loaded

with a given subject operates over a range of efficiencies. Searching for the RF shim

weights that maximize η can be accomplished using various numerical optimization

algorithms. However, it can be shown that calculating the maximum and minimum of

η can be treated as a generalized eigenvalue problem which does not require a

nonlinear search and guarantees the calculation of the global optimum. From the

solution obtained with numerical calculations (for example with the Matlab function

56
eig(Γ,Φ)), the largest eigenvalue and its corresponding eigenvector represent the

maximum transmit efficiency and the maximum efficiency RF shim weights, w,

respectively. Calculated maximum efficiency RF shim weights can be used in

experiments to obtain the highest possible transmit efficiency for the given coil-patient

configuration.

Figure 2.1 Experimental setup. a: Cross-sectional schematic of the coil and phantom setup. b:
Photograph of a loop/stripline module. The conductor layout of the active element of the
stripline and loop can be seen. c: Photograph of the 7 T experimental setup with a molded
human phantom. Loop/stripline coil modules indicated by numbers 1-5 were used in hip
experiments and the remaining modules were removed.

2.3.2 System Hardware and RF Coil Array

We evaluated the benefits of maximum efficiency RF shimming at 7 T,

targeting hip imaging as an exemplary application. Experiments were performed on a

57
whole body 7 T scanner (Magnetom, Siemens Medical Solutions, Erlangen, Germany)

equipped with an eight-channel parallel transmit system (1kW peak power per

transmit channel) and a gradient system capable of achieving peak gradient strength of

40 mT / m and a slew rate of 150 T / m / s. A 10-channel transmit / receive modular

array (93) (Figure 2.1a and c) consisting of five loop / stripline modules was used for

RF excitation and reception. Loop coils were 8 x 20 cm2 with a solid copper shield 2

cm above the loop conductors to reduce radiation loss and coupling to neighboring

coils and undesired anatomy such as the arms. The loops were tuned to 297.2 MHz

using 16 distributed capacitors of approximately 16 pF. Striplines with 15 cm length

and 2 x 3 x 15 cm3 Teflon dielectric were tuned using two capacitors of approximately

4.3 pF at opposing ends of the stripline. The striplines were centered with respect to

the loop coils (Figure 2.1b) such that their arrangement provided a naturally decoupled

loop/stripline module similar to that described in Ref. (94). Both loops and striplines

were capacitively matched to 50 Ω while loaded with a body-size agar phantom with

uniform electrical properties of average human muscle at 297.2 MHz (εr ≈ 58, σ ≈ 0.77

S / m.). The array of loop/stripline modules was placed around the phantom or human

torso with two posterior, one anterior and two lateral modules. The four loops closest

to the targeted ROI (inside modules 1-4 in Figure 2.1c) were used for RF transmission

and reception, while the remaining 6 elements (loop inside module 5 and striplines

inside modules 1-5) were used for RF reception only.

58
 In order to evaluate the maximum efficiency RF shimming approach in terms

of RF power reduction per B1+ squared, forward and reflected power readings were

obtained from four channels via an RF switch (Dual 16 x 1 MUX, National

Instruments, Austin, TX, USA) with a power sensor (NRP-Z11, Rhode & Schwarz,

Munich, Germany) connected to directional couplers (C8705, Werlatone, New York)

located at the penetration panel.

A 3T MRI scanner (Verio, Siemens Medical Solutions, Erlangen, Germany)

equipped with a gradient system capable of achieving peak gradient strength of 40

mT / m and a slew rate of 150 T / m / s was used to image the hip of the same

volunteers, and the SNR in the hip region was compared with that achieved using

maximum efficiency RF shimming at 7 T. The body coil was used for RF excitation

and a 32-element cardiac coil array (Invivo, Orlando, FL) was placed around the

pelvis for signal reception at 3 T.

2.3.3 RF Shimming Experiments

The use of local transmit coils called for RF power limits to restrict possible

tissue heating caused by the induced electric fields. To predict the spatial positions

with the greatest electric fields, a finite difference time domain (FDTD) (Computer

Simulation Technology, CST, Darmstadt, Germany), simulation (2 mm3 spatial

resolution) was performed in which a loop coil representative of that used in the

experiments was positioned adjacent to a uniform elliptical cylindrical phantom whose

size and electrical properties were similar to the human torso and muscle, respectively

59
(major diameter = 47.2 cm, minor diameter = 24.7 cm, electrical conductivity = 0.77

S/m, and dielectric constant = 58). To experimentally determine the safe operating

limit of a single coil, the temperature of a 3.6 kg lamb slab was recorded using

fluoroptic temperature probes (Luxtron M3300, Lumasense Technologies, Santa

Clara, CA, USA) during RF irradiation. The fluoroptic probes were inserted

approximately 5 mm into the lamb at four locations, including those with maximal

electric fields according to the FDTD simulation; 1) coil drivepoint, 2) capacitor

opposite the drivepoint, 3) capacitor midway along the side conductor, and 4) center of

coil. Distance from the coil conductor to the lamb was approximately 2 cm. RF power

was delivered to the coil for 10 min while the time-averaged (10 s) power was

monitored by vendor-provided hardware. Following RF irradiation, temperature

monitoring was continued for 2 mins to assess heat diffusion from locations adjacent

to the temperature probes. No temperature increase was observed during the post-RF

period. Assuming a linear relationship between RF irradiation and the rate of

temperature change, the safe operating limit was defined as the 10 s time-averaged

power input necessary to produce a 1°C temperature increase during a 10 min RF

irradiation period. For a single coil, the total power limit was hence determined to be

10W. In the parallel transmit experiments, power limits were applied using

conservative criteria that assumed the worst-case scenario in which electric-fields due

to individual transmit elements add constructively. Since four transmit loops were

used in the present study, this approach limited the individual input power to 16 times

60
less than the limit for a single loop (0.625W). In addition, 10 s and 6 min average RF

power was monitored for each channel in real time.

Calculation of the maximum efficiency RF weights requires B1+ profiles for

each transmitter along with the power correlation matrix. B1+ mapping was performed

following the method described in Ref. (79) by performing two separate

measurements using selective excitation of all channels without magnetization

preparation and with a saturation pulse on one channel at a time to produce spatial-

dependent B1+ map of the channel. B1+ magnitude maps, Figure 2.2c, in an axial plane

were obtained with sinc saturation pulses followed by a spoiled turbo fast low-angle

shot (FLASH) imaging acquisition with selective excitation from all channels.

Additional turbo FLASH imaging, using one coil for excitation at a time, were used to

calculate relative B1+ phase distribution for different coils (Figure 2.2d). Relevant

imaging parameters used for B1+ mapping were: field of view (FOV) = 360 x 360

mm2, echo time (TE) = 1.97 ms, acquisition matrix = 128 x 128, TR = 3 s, saturation

thickness = 10 mm, and slice thickness = 8 mm. Total acquisition time for B1+ maps in

all four channels was 27 s. An ROI over the hip articular cartilage was defined on one

of the images (Figure 2.2b) and the corresponding Γ-matrix (Figure 2.2e) was

calculated using the individual coil B1+ profiles.

The subject-specific power correlation matrix Φ (Figure 2.2f) was estimated

from measurements of the individual channel forward and reflected power, using the

power sensors connected to the directional couplers, associated with a set of

61
calibration RF pulses (60,61,66). The net power measurements (forward minus

reflected) during each predefined calibration pulse allows a set of linear equations

resembling Eq. [2.3] to be assembled and solved, using the predefined values of w as

known coefficients and assigning entries of Φ as unknowns. The calibrated Φ-matrix,

the Γ-matrix and the ROI were used to calculate the maximum efficiency RF shim

weights for the hip articular cartilage, following the procedure described earlier in the

text.

62
Figure 2.2 Steps required for the calculation of the maximum efficiency RF shimming
weights. a: Axial GRE image of one volunteer with the approximate stripline/loop coil
module locations overlaid in red (transmit/receive loops and receive-only striplines) and green
(receive-only loops and striplines). b: Zoomed GRE image with the target hip cartilage ROI
(red) for maximum efficiency optimization. c: B1+ amplitude maps for each transmit loop. d:
B1+ phase maps for each transmit loop. e: Γ-matrix calculated using the B1+ maps in the ROI.
f: Calibrated power correlation matrix, Φ, measured using the forward and reflected power
measurements of the system.

63
 Because experimental evaluation of several shim methods in the same

volunteer would require excessive examination time, we used experimentally acquired

B1+ maps and the calibrated Φ-matrix as simulation inputs for the offline comparison

of four RF shimming strategies: the proposed maximum efficiency RF shimming,

non-targeted unit RF shimming, local phase matching RF shimming, and

uniformity-targeted RF shimming. The non-targeted unit RF shimming delivers RF

with unit amplitude and zero phase offset to all transmit channels. Local phase

matching (24) aims to increase the constructive B1+ interference in the target ROI by

adjusting the transmit phase offset between each transmit channel.

Uniformity-targeted RF shimming (19) aims to increase B1+ homogeneity inside the

ROI by adjusting the relative transmit amplitude and phase to each channel through a

least squares solution.

In addition to the comparison of different RF shimming strategies in

simulations, net average power deposition and flip angle maps were measured to

compare the transmit efficiency achieved with the maximum efficiency RF shimming

to that achieved with non-targeted unit RF shimming. The flip angle distribution

resulting from both RF shimming methods, with acquisition matrix 256 x 256, was

measured using the method detailed above with saturation pulses played

simultaneously on all transmit channels. High resolution axial spoiled GRE images of

the hip region were acquired with maximum efficiency and non-targeted unit RF

shimming using the following parameters: acquisition matrix = 512 x 512, spatial

64
resolution 0.7 x 0.7 x 2 mm3, TE/TR = 4.73/400 ms, FOV = 360 x 360 mm2,

bandwidth (BW) = 300 Hz / pixel, and acquisition time 210 s. Additionally, net

average power deposition was measured using power sensors during GRE image

acquisition.

Four volunteers (three men and one woman; age = 37.5 ± 9.2 years) were

imaged in an axial plane through the left hip articular cartilage. Volunteer imaging

was performed with protocols approved by the New York University School of

Medicine Institutional Review Board, and written informed consent was obtained from

volunteers.

Maximum efficiency RF shimming does not inherently increase B1+

homogeneity within the selected ROI. However as shown before, adequate B1+

homogeneity can be achieved in a small ROI, such as the prostate (24). On the other

hand, special classes of RF pulses, such as the adiabatic pulse, inherently improve flip

angle uniformity. A drawback of adiabatic pulses is the high RF power deposition

required to satisfy the adiabatic condition at every voxel within the ROI. Since

maximum efficiency RF shimming aims to maximize B1+ field while minimizing the

power deposition, adiabatic pulses could benefit from the proposed RF shimming

method. We tested this hypothesis both in simulation and in experiments for an

adiabatic half passage (AHP) RF pulse (95). Experimentally measured individual

channel B1+ profiles (Figure 2.2c and d) and an AHP RF pulse of length 10.24 ms

were used in spinor-domain Bloch simulations (78) to calculate the flip angle

65
distribution. For the maximum efficiency shim and the non-targeted unit shim, the

adiabatic condition at the position in the ROI with the weakest B1+ was determined by

calculating the frequency response of the AHP RF pulse over a range of transmit

voltages using Bloch simulations; the adiabatic condition at this position was satisfied

when the z-component of magnetization at zero frequency was approximately zero and

not affected by further transmit voltage increase. Net power deposition of AHP RF

pulses with maximum efficiency shim and non-targeted unit shim was measured with

the transmit voltage for each shim set such that the AHP RF pulse satisfied the

adiabatic condition at all locations within the ROI. Flip angle distributions of the AHP

RF pulses with maximum efficiency shim and non-targeted unit shim were measured

using the technique described earlier for B1+ map acquisition (specifically, AHP RF

pulses were played as saturation pulses followed by a turbo FLASH acquisition). In

order to examine the off-resonance effect on AHP RF pulse, off-resonance maps were

calculated from individual receive coils using the three point "Dixon method" that

decomposes fat, water, and off-resonance through a least-squares calculation with

complex gradient echo images at TE = 4.08, 4.42, and 4.76 ms (96). A combined

off-resonance map was formed by weighting the contribution of each coil by the

square of its signal intensity. The combined off resonance map was smoothed using a

median filter with 5 x 5 kernel size.

66
 2.3.4 SNR Comparison

The SNR in the hip articular cartilage achieved with maximum efficiency RF

shimming at 7 T was compared with that achieved at 3 T. High resolution spoiled

axial GRE images of the hip region were acquired with low flip angles at both 3 T and

7 T, using the parameters given in the previous subsection. Due to different sample T1

in each magnet, low flip angle excitation was utilized to avoid magnetization

saturation effects that would complicate SNR analysis. Noise data were acquired with

zero transmit voltage and used to compute the noise covariance matrix of the receive

coils. The GRE images and the noise covariance matrix were used to generate SNR

maps following a method by Kellman and McVeigh (97).

Flip angle maps of the GRE acquisitions, with acquisition matrix 256 x 256,

were obtained using the flip angle mapping protocol explained in the previous section.

As the flip angle mapping algorithm is more prone to error at low flip angles, the flip

angle maps were acquired with higher transmit voltages than those used in the GRE

acquisitions. The acquired flip angle map was then scaled by the ratio of GRE transmit

voltage to flip angle mapping transmit voltage. The scaled flip angle maps were

interpolated to a matrix size of 512 x 512 to match the matrix size of the GRE

acquisition. For fair comparison between 3 T and 7 T, we removed the effect of spatial

flip angle variations by normalizing the SNR maps with the sine of the flip angle at

each voxel. Three of the four volunteers were imaged at both 3 T and 7 T for SNR

comparison.

67
Figure 2.3 Representative axial GRE images of one volunteer at 7 T (volunteer 1), acquired
with non-targeted unit RF shimming (a) and maximum efficiency RF shimming (b). Zoomed
images of the hip articular cartilage show that low signal caused by destructive RF
interference with non-targeted unit RF shimming (arrow in c) is restored using maximum
efficiency RF shimming (d).

2.4 Results

Representative 7 T axial GRE images with non-targeted unit RF shimming and

maximum efficiency RF shimming for volunteer 1 are shown in Figure 2.3a and

Figure 2.3b, respectively. Non-targeted unit RF shimming resulted in B1+

inhomogeneity and large signal intensity variations in the hip region and a local signal

68
drop indicated by the arrow in Figure 2.3c. The maximum efficiency RF shimming,

with a targeted ROI covering the left hip articular cartilage (Figure 2.3b), resulted in

improved homogeneity in the hip region (Figure 2.3d) and ~2.4 times increase in

transmit efficiency, as calculated with Eq. [2.5]. In achieving similar average flip

angles over the ROI in the cases of unit RF shimming (flip angle 27.6° ± 12.4°) and

maximum efficiency RF shimming (flip angle 25.3° ± 12.1°), respectively, the net

average energy deposition were measured to be 155 W and 58.8 W.

Transmit efficiency comparisons between different RF shimming methods are

summarized in Table 2.1. For all volunteers, maximum efficiency RF shimming

provided the highest transmit efficiency. Among the RF shimming methods, local

phase matching provided the second highest transmit efficiency (on average 22%

lower than that of the maximum efficiency method). In all volunteers, uniformity RF

shimming resulted in lowest transmit efficiency. This could be attributed to the

method's priority of increasing B1+ uniformity over increasing average B1+. The RF

power deposition benefit using maximum efficiency RF shimming compared to

non-targeted unit RF shimming for all volunteers is shown on the last row of Table

2.1. Table 2.2 shows that in imaging experiments the net RF power deposition with

maximum efficiency shim weights was on average 39% lower than that required to

achieve similar flip angles with the non-targeted unit RF shim. The experimentally

quantified RF power deposition benefit of using maximum efficiency RF shimming

was in good agreement with the RF power deposition benefit quantified with

69
simulations (last rows of Table 2.1 and Table 2.2). Calculated maximum efficiency

RF shim weights varied substantially among the volunteers due to differences in body

composition and size (Table 2.3).

Figure 2.4 Adiabatic half passage RF pulse results : B1+ maps with (top row) and without
(bottom row) maximum efficiency RF shimming. AHP pulses provided improved B1+
uniformity (columns two and three) over standard sinc pulses (column one) in the hip articular
cartilage. h is the measured off resonance map.

70
 Table 2.1 Comparison of four RF shimming methods in simulations based on experimentally acquired transmit sensitivity and power
correlation data: A: Non-targeted Unit RF shimming, B: Maximum Efficiency RF Shimming, C: Local phase matching RF shimming
from Ref. (24), and D: Uniformity RF Shimming from Ref. (19). The RF power deposition benefit of using maximum efficiency RF
shimming versus non-targeted unit RF shimming was calculated by comparing estimated power depositions per average unit squared
flip angles.

Volunteer 1 Volunteer 2 Volunteer 3 Volunteer 4

Shim
A B C D A B C D A B C D A B C D
Method

45.9° 40.0° 44.1° 17.8° 27.0° 23.4° 30.7° 9.0° 28.0° 23.9° 30.8° 7.6° 41.8° 34.5° 46.5° 13.2°
71

Flip Angle ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±
14.3° 11.2° 11.2° 4.1° 7.6° 5.2° 6.4° 2.6° 19.4° 15.1° 20.5° 4.8° 12.0° 9.4° 13.9° 3.3°

Estimated
231.3 73.9 218.5 50.5 148.8 76.3 151.5 1.8 147.7 80.2 147.7 44.1 283.5 138.2 274.7 134.5
Power (W)

Efficiency
9.97 23.34 9.46 6.58 5.27 7.51 6.51 1.78 7.83 9.96 9.23 1.83 6.62 9.22 8.57 2.08
(η)

RF Power
Deposition 58% 32% 26% 29%

Benefit
 Table 2.2 Experimentally measured net power deposition and corresponding flip angle with non-targeted unit RF shimming and
maximum efficiency RF shimming. Experiments using non-targeted unit RF shimming are denoted by A, and experiments using
maximum efficiency RF shim weights are denoted by B. The RF power deposition benefit of using maximum efficiency RF shimming
versus non-targeted unit RF shimming was calculated by comparing average net power depositions per average unit squared flip
angles.

Volunteer 1* Volunteer 2 Volunteer 3 Volunteer 4

Shim Method A B A B A B A B

Flip angle 27.6° ± 12.4° 25.3° ± 12.1° 6.3° ± 2.3° 5.8° ± 2.2° 4.3° ± 2.6° 3.8° ± 2.8° 6.9° ± 3.2° 5.9° ± 2.6°
72

Net Power
155.33 58.85 5.28 2.74 6.80 3.69 18.10 8.76
Deposition (W)

RF Power
Deposition 55% 39% 30% 33%
Benefit

*: GRE images and power measurements were obtained with higher flip angles compared to other volunteers
 Table 2.3 Calculated maximum efficiency RF shimming weights and measured individual anatomical dimensions for all volunteers.

Transmit Coil Number

Body Dimensions (cm)*
1 2 3 4
‖ ‖ ∡ ‖ ‖ ∡ ‖ ‖ ∡ ‖ ‖ ∡ A to C P to C L to C R

Volunteer 1 1 216.3° 0.44 31.3° 0.45 0.6° 0.22 0° 7.1 10.8 7.6 2.1

Volunteer 2 0.76 273.4° 0.71 17.7° 1 342.7° 0.36 0° 7.6 9.6 8.9 2.5

Volunteer 3 1 316.0° 0.5 45.2° 0.77 345.6° 0.57 0° 8.9 12.1 12.6 2.0
73

Volunteer 4 1 1.59° 0.45 92.1° 0.67 348.1° 0.61 0° 7.7 12.1 12.3 2.2

* Anterior (A), posterior (P), left (L), center of femoral head (C), and radius of femoral head (R).
 The power deposition and B1+ distribution of an AHP RF pulse with maximum

efficiency RF shim and non-targeted unit RF shim weights were assessed on volunteer

2. The maximum power required to meet the adiabatic condition at points inside the

ROI with the weakest B1+ distribution was 411 W for maximum efficiency shim and

789 W for non-targeted unit RF shim. Both shims resulted in a maximum of 2.31 μT

instantaneous B1+ field in the weakest B1+ location. Bloch simulations of AHP RF

pulses with the specified voltages resulted in a mean of 1% ± 3% z-magnetization,

which corresponded to remarkably uniform flip angle distribution of 89.4° ± 1.7°. Flip

angle distributions for maximum efficiency RF shim and non-targeted unit shim in the

ROI are shown for Bloch simulations (Figure 2.4c and Figure 2.4d) and experiments

(Figure 2.4d and Figure 2.4e). Whereas the Bloch simulation resulted in

approximately uniform 90° flip angle, experimental flip angle maps show increased

deviation in some locations which appear to correspond to locations with high main

magnetic field gradients (Figure 2.4h). This was confirmed in additional Bloch

simulations that incorporated the off-resonance maps. The AHP pulse provided a clear

improvement in B1+ uniformity over standard sinc excitation pulses (Figure 2.4a and

Figure 2.4b).

Figure 5 shows GRE images of volunteer 4 at 7 T and 3 T (Figure 2.5a and

Figure 2.5b) with flip angles (Figure 2.5c and Figure 2.5d) of 5.7° ± 2.6° and 9.7° ±

1.2° in the hip cartilage, respectively. Quantitative SNR maps of acquired GRE

images are shown in Figure 2.5e and Figure 2.5f. Figure 2.5e and Figure 2.5f clearly

74
show the actual SNR benefits of moving to higher field strength. Despite the lower flip

angles at 7 T (Figure 2.5c) compared to 3T (Figure 2.5d), the average SNR was

greater: 8.3 ± 4.9 at 7 T versus 6.9 ± 2.8 at 3 T (Figure 2.5e and Figure 2.5f). After

dividing the SNR maps by the sine of the flip angle at each voxel, the normalized SNR

was 43.4 ± 18.2 at 3 T and 83.5 ± 38.7 at 7 T. Averaged over all volunteers, 7 T

normalized SNR was 133% greater than that at 3 T (Table 2.4).

Table 2.4 SNR results in the hip articular cartilage of the volunteers at 3 T and 7 T.

Volunteer 2 Volunteer 3 Volunteer 4

3T 7T 3T 7T 3T 7T

Flip Angle 9.6° ± 0.9° 4.9° ± 1.9° 10.2° ± 1.4° 5.1° ± 1.9° 9.7° ± 1.2° 5.7° ± 2.6°

SNR 8.8 ± 3 10.7 ± 5.9 6.0 ± 2.2 7.7 ± 3.3 6.9 ± 2.8 8.3 ± 4.9

Normalized SNR 53.4±20.1 124.5±52.8 34.6 ± 13.8 96.9 ± 51.9 43.4 ± 18.2 83.5 ± 38.7

Normalized SNR
2.3 2.8 1.9
gain at 7T

75
Figure 2.5 SNR comparison at 3 T and 7 T Axial GRE images (top row) and zoomed flip
angle (middle row) and SNR maps (third row) from volunteer 4 at 7T (left column) and 3T
(right column).

76
 2.5 Discussion

In this work, we have demonstrated a maximum efficiency RF shimming

method that finds the lowest possible net RF power deposition into the subject for a

given flip angle inside the ROI. The proposed RF shimming method calculates optimal

shim weights which increase transmit efficiency by utilizing in vivo calibrated

predictions of the net RF power deposition along with B1+ field maps. Previous RF

shimming methods (24,87) only utilize B1+ constructive interference without

accounting for electrical field effects or power deposition. In addition, the transmit

efficiency metric defined in Eq. [2.5] enabled the global optimum RF shimming

weights to be efficiently calculated, without the need for computationally intensive

nonlinear search algorithms (86,98).

The proposed RF shimming method was compared in simulations using

experimental B1+ maps and power correlation matrices with three different RF

shimming methods: a) non-targeted unit RF shimming; b) uniformity-targeted RF

shimming (19), which has been used to address the challenge of signal inhomogeneity

that has hindered ultra-high-field imaging; and c) local phase matching RF shimming

(24), which has been found to perform effectively in FDTD simulations (98). The

simulations showed maximum efficiency RF shimming increased the transmit

efficiency compared to other methods by including calibrated subject-specific RF

power deposition predictions in RF shimming calculations. Some of the simulations

were further corroborated by imaging experiments, which confirmed the validness of

77
the comparison (last rows of Tables 1 and 2). In addition to the global RF power

deposition behavior documented here, local SAR properties of the proposed method

should be analyzed and compared with other RF shimming methods. However, for

such a comparison, full knowledge of actual electric field information inside the

subject is required. In future work, FDTD simulations can be used for such

comparisons, since determining the actual electric field inside the subject is not yet

feasible.

Hip imaging at 7 T was chosen as a representative application to demonstrate

maximum efficiency RF shimming. In fact, imaging the hip joint is challenging due to

its deep anatomical location, which requires large transmit voltages and results in

severe B1+ inhomogeneities at high field strength. In our volunteer experiments, power

measurements for sinc and AHP RF pulses in axial GRE acquisitions confirmed up to

50% decreases in RF power deposition while maintaining average flip angle

distributions. This suggests that SAR-intensive pulse sequences, such as turbo spin

echo (commonly used at lower magnetic field strengths for clinical hip imaging due to

high SNR and contrast-to-noise ratio), may become feasible at 7T using multiple-coil

transmission with maximum efficiency RF shimming. Our results show that the flip

angle-normalized SNR in the hip articular cartilage was on average 2.3 times greater

at 7 T than at 3 T. Furthermore, we showed that optimizing RF power deposition in a

ROI tends to reduce B1+ inhomogeneities within it. One limitation of our SNR

comparison study is the difference in receive coil sensitivities at 7 T and 3 T due to

78
variation in coil size and structure. However, the hip lies at a similar depth (5 to 12 cm

from the body surface) as the heart, suggesting that the cardiac array used at 3 T may

serve as a reasonable SNR benchmark.

In this study, GRE pulse sequences were used for the SNR comparison because

they are less SAR-intensive and therefore facilitated a comparison between 3 T and

7 T with our existing transmit hardware setup and safety limits. While GRE images

are not widely used for clinical morphological assessment of the hip articular cartilage,

they are employed for biochemical assessment in which T1 and T2* are measured

(99,100). In these applications, which could be facilitated at 7 T using maximum

efficiency RF shimming, improved SNR would result in more reliable T1 or T2*

quantification, or could be traded off for increased spatial resolution, which is

essential to resolve the thin layer of articular cartilage in the hip joint (91).

In summary, the maximum efficiency RF shimming method utilizes both

electric and magnetic field measurements corresponding to the in situ transmit array

and is subject to promptly calculate transmit shim weights that minimize the power

required for a given flip angle. An accompanying benefit of the proposed shim method

was that it provided reasonable flip angle uniformity in a clinically relevant ROI. The

shim method was successfully demonstrated in experimental 7 T MRI of the hip

articular cartilage, confirming the present method’s potential to outperform other shim

methods in terms of efficiency.

79
 2.6 Acknowledgements for Chapter 2

I would like to thank Dr. Hans-Peter Fautz from Siemens Medical Solutions in

Erlangen, Germany for collaboration on the flip angle mapping sequence. Dr. Graham

Wiggins is acknowledged for discussions on SNR comparison. I would like to thank

Dr. Bei Zhang for her help in FDTD simulations, Kellyanne Mcgorty for her help on

sequence protocols and Dr. Assaf Tal for useful discussions on adiabatic RF pulses.

80
 CHAPTER 3: Subject-specific Proactive Management of Parallel RF

Transmission

Author contributions:

Cem Murat Deniz: Chapter writing, study design, RF Pulse design, Matlab software,

data acquisition and analysis

Leeor Alon: Power calibration system and software, chapter editing

Ryan Brown: MR coils and interface

Daniel K. Sodickson: Study concept, chapter editing

Yudong Zhu: Study concept and design, data interpretation, chapter editing

81
 3.1 Abstract

MR scanners have predefined power delivery and reflection handling

capabilities. Any practical RF pulse used on a scanner must be designed with those

capabilities in mind. In parallel transmission, the interactions between individual

channels, and between these channels and the imaged subject, play an important role

in power delivery in determining the demands placed upon the power amplifiers. By

using pre-scan based individual channel forward and reflected power calibration, we

designed parallel RF excitation pulses obeying the forward / reflected peak and

average power limits of the RF power amplifier. Additionally, global SAR limits were

incorporated in the RF pulse design. Results showed that the prediction capability of

this new calibration method enables the design of parallel RF excitation pulses

respecting strict and multifaceted power limits.

3.2 Introduction

When applying parallel RF transmission in practice, coupling and interaction

taking place in the multi-port coil structure as well as in the subject can significantly

affect individual channel RF power transmission towards and away from the subject,

posing challenges to transmit channel instrumentation and safety monitoring. Tracking

and predicting these effects and proactively managing power transmission is important

for ensuring a smooth scan. In this chapter, PPM technique (60,61,66) described in

Section 1.3.6 for global SAR is further extended to individual channel forward and

reflected power for any RF excitation. The forward and reflected power predictions

82
were used proactively to design constrained parallel RF excitation pulses to meet the

RF power requirements. The constrained parallel RF excitation pulses designed in this

way were played out on the MR scanner, and resulting forward and reflected power

measurements as well as excitation fidelity were compared with unconstrained pulse

designs or designs constrained by global SAR only.

3.3 Materials and Methods

3.3.1 Individual Channel Power Prediction

Our PPM technique (60,61,66) uses in situ individual channel forward and

reflected power measurements that correspond to the application of a set of calibration

RF pulses to estimate the global power correlation matrix Φ. This scheme can be

extended to model and predict individual channel forward or reflected power by using

the notation of Section 1.3.2 and the following equations:

l
Pfwd ( pt )  b Hpt Φlfwd b pt and Prfll ( pt )  bHpt Φlrfl b pt [3.1]

l
where Pfwd ( pt ) and Prfll ( pt ) are the lth channel's measured forward and reflected

T
power at time instant pΔt, respectively. b pt  b1, pt  bL , pt  defines the

T
predefined input calibration weights from L transmit channels similar to Eq. [1.2],
H
denotes the transpose and denotes the complex conjugate transpose. By using the

predefined calibration weights and measuring the associated power, forward, Φ lfwd ,

and reflected, Φlrfl , power correlation matrices of all channels can be estimated.

83
 There are various possibilities for leveraging channel-by-channel power

prediction capability: 1) Given the peak power rating of the power amplifiers assigned

to drive the parallel transmission channels, knowing in advance the peak power

requirements for the individual channels allows the user to proactively adapt the

excitation pulse design and / or reconfigure the transmit hardware (e.g., by updating

the power combination scheme applied to the component amplifier units). 2) Given the

reflected power handling capacity of the amplifiers / circulators on the parallel

transmission channels, knowing in advance large peak reflected power for the

individual channels similarly allows the user to implement software- and / or

hardware-based mitigation strategies. 3) Checking the individual channel power

predictions against actual measurements, or comparing the matrices determined at

baseline and those updated periodically afterwards, provides diagnostics that can

detect in real-time system changes caused by, for example, hardware failure, system

instability or patient position change. These diagnostics can be used as triggers to

suspend scanning as needed. In other words, using the power prediction models both

in planning and in monitoring may avert amplifier peak power or voltage standing

wave ratio faults, protection hardware breakdown, and excessive SAR due to system

failure.

Potential use of individual channel power prediction models in detecting /

diagnosing system changes in real-time (option #3 above) was explored in detail in

Ref. (66). In this chapter, the first two options described above will be explored by

84
using constrained parallel RF transmission pulse design. Forward and reflected power

correlation matrices will be used to guide parallel RF transmission pulse design with

strict peak and average forward and reflected power constraints.

3.3.2 Constrained RF Pulse Design

The calibration of forward and reflected power correlation matrices enables the

prediction of an individual channel's forward and reflected power given an arbitrary

RF pulse at any time instant. This property enables proactive power transmission /

resource management through RF pulse calculation. One way of integrating power

prediction capability into RF pulse design is to use the following convex inequalities:

b Hpt Φlfwd b pt  Pfwd

l
, peak , l  1, , L [3.2]

b Hpt Φlrfl b pt  Prfll , peak , l  1, , L [3.3]

  b Hpt Φlfwd b pt  Pfwd

l
, ave , l  1,  , L [3.4]
p

  b Hpt Φlrfl b pt  Prfll ,ave , l  1,, L [3.5]

p

l l
where Pfwd , peak represents the lth channel's peak power delivery capacity, Prfl , peak

l
represents the lth channel's tolerance to reflected peak power, Pfwd , ave represents the lth

channel's average power delivery capacity, Prfll ,ave represents the lth channel's average

power reflection capacity and α = 1 / RF pulse width. In addition to the constraints

involving individual power predictions, predefined maximum global SAR limits

allowed by FDA guidelines (20) can be incorporated using the global power

85
correlation matrix of Eq. [1.10]. Using STA approximation and variables defined in

Section 1.3.3, RF pulses for parallel excitation can be calculated by solving the

following optimization problem:

2
bˆ full  arg min Afullbfull  mdes 2
bfull

such that b H Φ b  globalSARLimit

full full full

b Hpt Φlfwd b pt  Pfwd

l
, peak l , p
[3.6]
b Hpt Φlrfl b pt  Prfll , peak l , p
  b Hpt Φlfwd b pt  Pfwd
l
, ave l
p

  b Hpt Φlrfl b pt  Prfll ,ave l

p

T
where b full  b1Tt , b T2 t  b TPt  is the concatenation of the coil RF pulse waveforms

for each time point pΔt, and Φfull is the matrix containing global power correlation

information to be used in conjunction with b full as explained in Eq. [1.10].

This optimization problem can be solved by using a range of efficient

strategies for convex optimization since the power correlation matrices are positive

definite and the constraints are quadratic convex functions. Convex optimization

guarantees that a global optimum, if it exists, will be found within a defined error

bound. The complexity of the optimization problem increases with the RF pulse

length, the number of channels, and the desired magnetization resolution. The

complexity of a similar optimization problem (36) was reduced using least-squares

projections (101) in order to find a small number of basis vectors which still contains a

good approximation to the original problem but reduces the optimization search space

86
drastically. We followed steps described in Ref. (36) to reduce the complexity of the

optimization problem, specifically using Lanczos algorithm with Gram-Schmidt re-

orthogonalization steps (102). New formulation of the convex optimization problem

using reduced-basis vectors still includes the exact power constraints as defined in Eq.

[3.6] and can be solved efficiently by using a variety of well established solvers. In

this work, the SeDuMi (103) v1.2.1 solver, interfaced with YALMIP (104), was used

to solve the reduced basis convex optimization problem.

3.3.3 Experimental RF Pulse Design

Experiments were performed on a Siemens whole body 7 T Magnetom scanner

(Erlangen, Germany) equipped with an eight-channel parallel transmit system (1kW

peak power per transmit channel) in order to demonstrate the subject-specific

proactive management of parallel transmission RF pulse design by using the calibrated

power correlation matrices. The eight channel coil array (Figure 1.3a), phantom

(Figure 1.3b) and power measurement setup described in Section 1.3.6 were used in

this study as well. B1+ calibration was performed on an axial slice at the isocenter

following the method and the parameters described in Ref. (79) and Section 1.3.6,

respectively.

Global Φ and channel-by-channel forward Φ lfwd and reflected Φlrfl power

correlation matrices were calibrated by measuring in situ individual channel forward

and reflected power for a set of calibration RF pulses. Power sensors were connected

to directional couplers at the output of each RF amplifier, The calibrated power

87
correlation matrices were used in constrained parallel transmission RF pulse design in

order to limit the global SAR (by using Φ, Figure 3.1a), peak and average forward

power (by using, Φ lfwd , Figure 3.1b), and peak and average reflected power (by using

Φlrfl , Figure 3.1c).

Figure 3.1 Example of calibrated power correlation matrices. a: global power correlation
matrix, b: forward power correlation matrix of transmit channel 4, and c: reflected power
correlation matrix of transmit channel 4.

Figure 3.2 Desired excitation profile and k-space trajectory a: Desired 2D axial excitation
profile, and b: spiral-in excitation k-space trajectory

88
 Unconstrained, global SAR constrained, and fully constrained (global SAR,

peak forward and reflected power, average forward and reflected power) RF pulses

were designed using the target excitation profile shown in Figure 3.2a, using custom

code and a custom-built GUI (see Appendix) developed in Matlab (version 7.13,

MathWorks, Inc., Natick, MA, USA). A constant rate spiral-in excitation k-space

trajectory (Figure 3.2b) was used with duration = 4.5 ms (corresponding to 4.3-fold

acceleration with respect to unaccelerated k-space), excitation resolution = 3.8 mm,

sampling interval = 10μs, maximum gradient slew rate =150 mT / m / s and maximum

gradient amplitude = 40 mT / m. In the present feasibility study, the following power

l
limits were used in constrained RF pulse design: global SAR = 3.2 W / kg, Pfwd , peak =

700 W, Prfll , peak = 50 W, Pfwd

l l
, ave = 50 W, Prfl ,ave = 25 W. Forward and reflected power

in eight channels were measured with a sampling rate of 5μs while calculated RF

pulses were used in a 3D GRE acquisition with the following parameters: FOV = 240

x 240 mm2, TR = 80 ms, TE = 5 ms, matrix size = 64 x 64, number of slices = 48, and

slice thickness = 5mm. Measured powers were compared to channel-by-channel

forward and reflected power predictions based on calibrated power correlation

matrices.

3.4 Results and Discussion

Axial GRE images and Bloch simulation results for the excitation profiles of

RF pulses designed with different constraints are shown in Figure 3.3. The NRMSEs

of the desired and obtained magnetization from Bloch simulations were 0.0220 /

89
0.0224 / 0.0258 for unconstrained / global SAR constrained / fully constrained RF

pulse designs. All designs resulted in similar acceptable excitation fidelity. The

increase in NRMSE in constrained RF pulse design shows that in order to meet the

strict constraint requirements, some compromise in excitation fidelity was required.

However, this 0.3% increase in excitation error is hardly noticeable on the GRE

images (Figure 3.3d, f).

Figure 3.3 Bloch simulation results and axial GRE images of designed RF pulses are shown in
a and d for unconstrained design, b and e for global SAR constrained design, c and f for fully
constrained design. Red circle in f represents the phantom boundary.

Figure 3.4 shows the individual channel forward power predictions and actual

power measurements for Channel 4. There is notable agreement between power

predictions and experimental power measurements (similar agreement was observed

90
for net power measurements and predictions as illustrated in Figure 1.7).

Approximately 10% lower measured peak power was observed compared to

predictions. This could be explained by mismatch between the RF and the power

meter raster times, and by temporal averaging involved in power measurement. In

order to further emphasize the necessity of power correlation matrix calibration for

proactive management of parallel RF transmission, the reflected power in Channel 4

was estimated by neglecting the contributions of other channels to the reflected power

in that channel, i.e. by using single-element reflected power correlation matrix shown

in Figure 3.4e. This resulted in significant deviations from measured reflected power,

as indicated by red arrows in Figure 3.4a, and inaccurate maximum power estimation.

Figure 3.4 Comparison of individual channel actual power measurements (a) and power
prediction (b) using calibrated reflected power correlation matrix (d) for Channel 4. Assuming

91
the reflected power correlation matrix as shown in e (i.e. neglecting reflected power
contributions from other coils) resulted in reflected power predictions (c) which deviate
significantly from measured power, as indicated by red arrows in a.

92
Table 3.1 Power comparison of RF pulses with different power constraints. Italic channel
numbers indicate which transmit channel had the peak power displayed in the table for the
particular measurement. Measured global SAR, forward and reflected peak and average power
values for the fully constrained RF pulse design can be compared with the estimated values
using the last two columns.

Measured Estimated

Global SAR Fully Fully

Unconstrained
Constrained Constrained Constrained

Global SAR
5.14 2.93 2.3 2.4
(W/kg)

FWD Peak 1103.2 916.2 683.9 700

(W) (ch3) (ch1) (ch4)

RFL Peak 86.5 67.5 46.4 50

(W) (ch3) (ch3) (ch3)

FWD Average 65.3 40.3 30.7 33.5

(W) (ch1) (ch1) (ch1)

RFL Average 3.7 2.2 1.59 1.96

(W) (ch2) (ch3) (ch3)

Table 3.1 summarizes the benefits of RF pulse design with the guidance of

calibrated power correlation matrices. RF pulse design without any constraints

violated some of the limits in various channels (column Unconstrained). Designing RF

pulses with only a global SAR constraint successfully enforced the global SAR limit,

but violated peak and reflected power limits in some channels (column Global SAR

Constrained). All violations were removed by designing the RF pulse with all

93
constraints active (column Fully Constrained). Proper guidance can also be verified by

the last column of the Table 3.1, in which the power prediction matches well with

experimental measurements in the indicated channels. In Figure 3.5, the measured

forward power of Channel 5 and reflected power of Channel 3 show that violations of

peak power limits (indicated by red lines) are removed by guiding constrained RF

pulse design with calibrated power correlation matrices.

Figure 3.5 Measured power for RF pulses designed with different power constraints. Red
horizontal lines indicate the power limits used in constrained RF pulse design. Left column:
forward power measurements in a representative channel, Channel 5. Right column: reflected
power measurements in a representative channel, Channel 3.

In this work, we demonstrated the subject-specific proactive management of

parallel transmission using calibrated power correlation matrices and RF pulse design

with convex optimization Strict power limits on patient safety and on MR scanner

hardware are guaranteed during RF pulse design.

94
 CHAPTER 4: RF Energy Deposition and RF Power Requirements in Parallel

Transmission with Increasing Distance from the Coil to the Sample

Deniz CM, Lattanzi R, Zhu Y, Wiggins G, and Sodickson DK

RF Energy Deposition and RF Power Requirements in Parallel Transmission with

Increasing Distance from the Coil to the Sample

In Proceedings of the 17th Scientific Meeting, International Society for Magnetic

Resonance in Medicine, Honolulu, page 4802, 2009.

Author contributions:

Cem Murat Deniz: Abstract draft, study design, simulations, data analysis

Riccardo Lattanzi: Simulation software, abstract editing, data interpretation

Graham Wiggins: Study concept and design

Yudong Zhu: Study concept

Daniel K. Sodickson: Study concept and design, data interpretation, abstract editing

95
 4.1 Abstract

Minimizing SAR while maintaining a homogenous excitation is one of the

principal challenges associated with the use of ultra high magnetic field strengths. We

investigated the SAR behavior and the power requirements for parallel transmission as

the gap between transmit elements and the surface of the object is increased. Various

simulated geometrical arrangements of coil elements around a sphere and a cylinder

were explored: one in which an increasing number of coils of fixed size were placed

around the object, and another in which a fixed number of coils with increasing radius

were arranged at increasing distance from the object. We found that global SAR and

peak SAR during parallel excitation decreases with lift-off for spherical object-coil

setup, approaching the lowest SAR allowed by electrodynamics (i.e. the ultimate

intrinsic SAR) while the input power requirements to achieve the desired excitation

increases rapidly with lift-off. On the other hand, optimal coil lift-off that minimizes

the global SAR and input power requirements were found for the cylindrical object-

coil setup. Thus, for parallel transmission there are SAR benefits in moving coils away

from the object, but RF power requirements may represent a practical limiting factor.

4.2 Introduction

Parallel transmission with multiple RF coils (28,29) enables homogeneous

excitations at ultra high magnetic field strengths, while minimizing the specific

absorption rate over the entire volume of the sample (29,72). However, electric fields

generated by transmit coils placed close to the body may cause dangerous hotspots,

96
even though average global SAR remains small over the duration of excitation. On the

other hand, if the coils are placed at a distance from the body, the RF power required

to achieve a given flip angle distribution may be high, and it may be feared that this

will result in increased global SAR. Furthermore, increasing the distance between the

transmit array and the sample widens the area of overlap between individual coil

sensitivities, which may compromise the performance of parallel transmission

techniques. SAR dependence on array geometry in parallel transmission was studied

by Katscher et al. (49), by changing the relative orientation between two transmit coils

placed at a fixed distance from the center of a spherical object. For a two-coil

experimental setup, Katscher et al. (49) found that the angular tolerance of the coil

positions was typically ~20° - 30° with a tolerance of 10% increase in SAR compared

to the optimal coil SAR deposition.

In this work we investigated global, peak and local SAR behavior and the

corresponding RF power requirements with respect to the separation between the

transmit elements and the surface of the object, in the case of a dielectric sphere and

cylinder at 3 T and 7 T main magnetic field strengths. Ultimate intrinsic global SAR

(72), which is defined as the lowest possible global SAR consistent with

electrodynamics for a particular excitation profile but independent of transmit coil

design, was used to compare how closely different transmit array designs were able to

approach the best possible configuration. Surface-contoured rectangular coils and

97
circular coils of different sizes and numbers were used to identify the optimal lift-off

distance between the transmit array and the object surface.

4.3 Materials and Methods

A dyadic Green’s function (DGF) formulation (105) was used to derive the

full-wave electromagnetic fields inside a dielectric sphere / cylinder from a complete

basis of current modes, which were defined on a spherical / cylindrical surface

concentric with the object. The calculated complete basis set of current modes was

employed to calculate ultimate intrinsic global SAR (72). In order to calculate the

appropriate weighting of the current modes, uniform target excitation profiles were

chosen in the transverse plane for spherical, and coronal and transverse planes for

cylindrical simulation setups. Minimum global SAR for finite arrays of transmit loop

coils were calculated using current mode weights for parallel transmission aimed at

simultaneous global SAR minimization and B1+ homogeneity (106). The

corresponding input RF power requirements for calculated coil weights were

estimated by adding the RF power dissipated in coil conductors to the RF power

deposited in the sphere. In addition to sample and coil losses accounted for in the

spherical geometry, cylindrical simulations incorporated losses due to eddy currents

into the input RF power requirements by modeling the conductive magnetic shield

around the cylinder.

98
Figure 4.1 Transmit array geometries for spherical simulations. a: Belt-like design in which
coil size is kept constant and the number of coils is increased during lift-off. b: Symmetric
design in which the number of coils is kept constant and coil radius is increased during lift-off.

Increasing the distance between the object surface and the center of transmit

elements requires an increase in either number or size of the transmit coil elements. In

this work, both lift-off strategies were simulated for spherical and cylindrical

simulation setups. Figure 4.1 illustrates an example of transmit array geometries for

spherical simulations using both lift-off strategies. For the increased coil number

strategy, an increasing number of loop coils were arranged like a belt around the

sphere equator, fixing coil radius to 5 cm ("belt-like design," Figure 4.1a). For the

increased coil size strategy, a fixed number of coils were symmetrically packed around

the sphere, with individual coil radii scaling up with increasing lift-off ("symmetric

99
design," Figure 4.1b). In both simulation strategies, a 15 cm radius dielectric sphere

was used with the following average brain tissue properties (107): dielectric constant

εr = 63.1 / 52, conductivity σ (S/m) = 0.46 / 0.55 for 3 T / 7 T.

Figure 4.2 Transmit array geometries for cylindrical simulations. a: Array design in which
coil size is kept constant and the number of coils is increased during lift-off. b: Array design
in which the number of coils is kept constant and coil size is increased during lift-off.

Figure 4.2 illustrates transmit array geometries used for cylindrical simulations

using two different lift-off strategies: 1) increasing the number of coils (Figure 4.2a)

and 2) increasing the coil size (Figure 4.2b). A cylindrical object of radius 15cm and

length 40 cm was used in both simulations. Since the cylindrical simulation setup

100
resembles application typical body more than a typical head, dielectric properties of

dog skeletal muscle were used.

The excitation of a uniform target profile on a transverse plane through the

center of the sphere was simulated for the case of a 32 x 32 EPI excitation trajectory,

using a SAR minimization algorithm for parallel transmission (29,72). Similarly,

transverse and coronal planes through the center of the cylinder were simulated using

24 x 24 and 18 x 24 EPI excitation trajectory, respectively. The complete current basis

set was defined on the spherical and cylindrical surfaces where the individual coils are

located in order to calculate the ultimate intrinsic global SAR. Calculations were

performed in MATLAB (Mathworks, Natick, USA) for different lift-offs, coil

numbers and field strengths (3 T and 7 T). Convergence tests of the ultimate global

SAR optimization, by changing the maximum order of the basis function expansion,

resulted in 13122 / 18281 current modes in the spherical / cylindrical simulation basis

sets for full convergence.

4.4 Results and Discussion

The target excitation profile was achieved in all cases. Figure 4.3 shows

minimum global SAR and RF power requirements as a function of the distance of the

finite arrays to the surface of the sphere. Results are presented for 3 T and 7 T main

field strengths and for different coil designs (belt-like and symmetric). Each plot is

normalized to the ultimate intrinsic SAR of the corresponding main magnetic field

strength, which notably remains constant for different lift-offs. In the ultimate case,

101
local SAR also does not change with lift-off, suggesting that there is a single optimal

electromagnetic field distribution that minimizes SAR while maintaining profile

fidelity, and it can be always achieved by choosing the appropriate combination of

modes in the basis set.

Figure 4.3 Optimized global SAR and RF power requirements versus lift-off , for the belt-like
(left column, a-c) and symmetric (right column, b-d) array design in spherical phantom. Each
plot is normalized to the ultimate intrinsic SAR at the corresponding magnetic field strength.

It is apparent from Figure 4.3a,b that global SAR is reduced, approaching

more closely the theoretical smallest value, ultimate intrinsic global SAR, as coils are

moved further from the object in both belt-like and symmetric designs at both 3 T and

102
7 T. However, when the same B1+ field distribution is used as a target, the

corresponding RF power requirements increase dramatically with increasing radius

(Figure 4.3c, d). Both global SAR and RF power requirements are higher at 7 T than at

3 T in all cases. RF power requirements for the 24-element symmetric array are lower

than for the 12-element array when the coils are close to the object, but power

requirements grow more rapidly with lift-off, since individual coil dimensions increase

and lead to larger dissipation.

Spatial distributions of local SAR in the center of excitation k-space and peak

SAR during the entire excitation are shown in Figure 4.4 for different lift-offs, in the

case of the belt-like array design for spherical object. It appears that when the coils are

near the surface of the sphere, electric fields generated by the coils are larger and may

cause higher RF energy deposition. Additionally, the increase in the number of coils

during lift-off can also be a contributing factor in the decreasing peak local SAR.

Figure 4.5 illustrates the optimal global SAR and RF power requirements for

the cylindrical object simulations using coronal and transverse target excitation FOV.

In all the simulations uniform target excitation profile was achieved while the

remaining degrees of freedom were used to decrease the global SAR of the calculated

RF pulses. As coils moved further from the object global SAR approached the

ultimate intrinsic global SAR (Figure 4.5a). Figure 4.5b indicates that the global SAR

minimization is more effective using larger numbers of coils: lower global SAR is

achieved in 32 coils as compared to 16 coils for the same lift-off distance. Both global

103
SAR and RF power requirements in both designs were higher at 7 T than at 3 T main

magnetic field strength. In contrast to the spherical simulations, an optimal coil lift-off

that minimizes RF power requirements was found for cylindrical object simulations.

In this work, we found that for parallel transmission there are SAR benefits in

moving the transmit coils away from the object, especially at higher field strengths.

Global SAR was found to decrease monotonically for the spherical case. On the other

hand, an optimum lift-off distance with minimum RF power requirements was

observed for particular coil geometries for the cylindrical case. Peak local SAR

decreases with lift-off in all cases. However, the increase in corresponding RF power

requirements may constitute a practical limitation to these benefits. There were a few

notable differences between the spherical and cylindrical simulations. The conductive

shield was not modeled for spherical simulations and losses due to eddy currents were

not included in the RF power requirements. Results for the cylindrical case could be

affected by the particular choice of placing the coils, i.e. rectangular coil size and

number were adjusted only in one dimension. Ultimate intrinsic global SAR is

independent of coil lift-off and can be used in this case as an absolute reference.

Ultimate intrinsic global SAR was approached with finite coils as lift-off distance and

number of transmit coils were increased.

In summary, we showed that there will be SAR benefits of moving RF coils

away from the subject when power requirements are well compensated. These

104
findings will serve as important guide for improving existing RF transmit coil designs

for practical parallel transmission.

Figure 4.4 Local SAR vs lift-off for the sphere . Peak SAR (top) and local SAR (bottom)
versus lift-off for the belt-like array design, at 3T and 7T. Normalized spatial SAR distribution
(base-10 log scale) within the FOV, during excitation of the center of k-space is shown for the
smallest (1 cm), and intermediate (10.9 cm) and the maximum (20.9 cm) lift-off value.

105
Figure 4.5 Optimized global SAR and RF power requirements versus lift-off for the cylinder.
Two different cylindrical phantom designs A (a) and B (b) are used as shown in Figure 4.2.
Results from both coronal (left column) and transverse (right column) FOVs are illustrated.
An optimal coil lift-off that minimizes the RF power requirements was observed. Each plot is
normalized to the ultimate intrinsic SAR at the corresponding magnetic field strength.

106
 CHAPTER 5: Sparse Parallel Transmit Excitation Trajectory Design for Rapid

Inner-Volume Excitation

Deniz CM, Chen D, Alon L, Brown R, Fautz H-P, Sodickson DK, and Zhu Y

Sparse Parallel Transmit Excitation Trajectory Design for Rapid Inner-Volume

Excitation

In Proceedings of the 19th Scientific Meeting, International Society for Magnetic

Resonance in Medicine, Montreal, Canada. page 4434, 2011.

Author contributions:

Cem Murat Deniz: Abstract draft, study design, RF pulse design, data acquisition, data

analysis, literature survey

Dong Chen: subspace OMP method and software

Leeor Alon: Power measurement software

Ryan Brown: MR coils and interface

Hans-Peter Fautz: Flip angle mapping sequence

Daniel K. Sodickson: Study concept, abstract editing

Yudong Zhu: Study concept, abstract editing

107
 5.1 Abstract

Tailored inner-volume excitation on whole-body scanners is often limited by

long 3D RF pulses. Effective pulse length reduction with parallel transmission

requires careful selection of the excitation k-space trajectory. In this work,

twomethods of determining sparse excitation trajectories were compared for parallel

transmit pulse design in the small-tip angle and large-tip-angle regimes: a) a subspace

Orthogonal Matching Pursuit algorithm, and b) a single-step thresholding algorithm.

Reasonable inner-volume excitations with a pulse length of less than 9 ms were

achieved using an eight-channel transmitter on a whole-body human 7T scanner.

5.2 Introduction

Tailored inner-volume excitation presents many challenges on whole-body

MRI systems, such as maximum gradient strength and slew rate limitations, the

selection of robust excitation k-space trajectories, and transmit field inhomogeneity for

high-field systems. Selection of the excitation k-space locations is one of the most

crucial decisions in RF pulse design as it directly impacts image quality and scan time.

This section briefly describes the excitation k-space concept and clarifies the

requirements of selecting a trajectory for a given MRI application.

5.2.1 Excitation k-space

The excitation k-space formalism is closely related to the concept of k-space

used more familiarly in MR imaging (6,7). In order to traverse (imaging) k-space, the

gradient magnetic fields, generated by x, y and z gradient coils, are superimposed

108
upon the main magnetic field. During MR measurements, imaging k-space is filled

with MR signal S(t) while driving the gradient coils simultaneously. Using the MR

signal equation neglecting decay terms, this process at any time instant t can be

described by:

S (t )   q(r)eik (t ) r dr [5.1]
R

where q(r) is a factor which is mainly proportional to local magnetization density ρ(r)

at location r and R is the imaging volume. Imaging k-space locations, k(t), are defined

by

t
k (t )    G ( )d [5.2]
0

where G(τ) represents the gradient waveforms and γ is the gyromagnetic ratio. As can

be seen from Eq. [5.1], encoded signal in imaging k-space is the Fourier representation

of the magnetization density distribution, while position r and spatial-frequency k are

Fourier transform pairs. This formalism helped the MRI community to better

understand and visualize the signal acquisition mechanism, which then formed a basis

for new acquisition approaches as parallel MRI (12,13).

A k-space approach can also be applied to the design of RF excitation pulses

(69). Using the STA approximation and neglecting the relaxation terms T1 and T2, the

transverse magnetization, Mxy, at time T can be described as a function of the applied

RF, B1(t), and gradient fields:

109
 T T

M xy (r )  i M 0  B1 (t )e
 i r  t G ( ) d dt [5.3]
0

where M0 is the initial magnetization before applying RF. Defining a spatial frequency

variable k(t) as

T
k (t )    G ( )d [5.4]
t

the excitation k-space is generated by applied gradient fields during RF transmission.

Both imaging and excitation k-spaces defined by Eqs. [5.2] and [5.4] are

specified by applied gradient fields. In excitation k-space, the spatial frequency

variable k(t) is defined as the integral of the remaining gradient field compared to the

imaging k-space which is defined as the integral of the elapsed gradient field. This

difference is a natural result of the applied gradient field's effect on the transverse

magnetization phase distribution. As time passes, phase distribution of the transverse

magnetization that is excited by a STA RF pulse at time instant t evolves with the

gradient field applied until time T. On the other hand, in imaging k-space, until MR

signal is acquired, the phase distribution of the transverse magnetization after RF

excitation evolves while the gradient fields are applied for imaging. For that reason,

the imaging k-space is defined by the integral of the elapsed gradient field. The

excitation k-space formalism can be used to design STA RF excitation pulses (69) and

some special classes of LTA RF excitation pulses (76). One such special class of pulse

is the linear class, for which the k-space trajectory satisfies the following linear class

assumptions: 1) k(t) starts and ends at the center of excitation k-space, 2) k(t) can be

110
decomposed into a sequence of inherently refocused subtrajectories, in which STA

subpulses can be maintained. However, for LTA excitations which do not satisfy the

linear class assumption, the excitation k-space trajectories defined in Eq. [5.3] and

[5.4] may not be optimal, and our intuition must be adjusted.

Selection of the excitation k-space excursion and sampling density is one of the

major decisions for RF excitation pulse design. In the image domain, the chosen

excitation resolution specifies the smallest possible excitation volume and the

minimum transition width of the sharp edges required in RF pulse excitation. The

three dimensional excitation resolution in the image domain, Δr, is inversely

proportional to the three dimensional excitation k-space extent, kmax, with

 r  1 / 2k max , indicating that increased excitation resolution requires wider excitation

k-space coverage. Once a target excitation k-space extent has been chosen, the RF

excitation pulse design problem still requires definition of the density with which the

continuous excitation k-space will be discretely sampled. As is the case for MR

imaging k-space, the Nyquist-Shannon sampling theorem (108,109) defines the

excitation k-space sampling density requirements to enable excitation of the spins

without aliasing in the defined excitation field-of-view (xFOV). Aliasing of the

excitation / image occurs when a continuous RF / free induction decay (FID) signal is

digitized at a rate that is insufficient to capture the changes in the signal. This

sampling requirement in excitation k-space can be achieved by defining the xFOV and

obtaining the excitation k-space sampling interval, Δk, via k  1/ xFOV .

111
 In MRI, imaging k-space data is acquired sequentially using phase and

frequency encoding generated by gradient fields. In every TR, lines in k-space are

filled with the acquired signal obtained after RF pulse excitation. This flexibility of

acquiring different k-space lines within different TRs enables the acquisition of high

resolution images. However, excitation k-space must be covered for every TR. This

requirement imposes a maximum possible RF length and excitation resolution

depending on the local properties of the body, e.g. T1 and T2*, which is the main

reason that most MRI applications use short (<< T2*) slice selective or nonselective RF

pulses.

The advent of parallel MRI (12,13) enabled faster MRI signal acquisitions by

reworking the imaging k-space sampling density requirements with spatially distinct

sensitivities from multiple receiver channels supplementing the spatial encoding that is

typically performed by gradients. In parallel imaging, undersampled imaging k-space

is acquired during MR signal acquisition and the full image is reconstructed using

receive sensitivity profiles after MR signal acquisition. Similarly, in parallel

transmission (28,29), the excitation k-space sampling density requirements can be

reduced by using multiple RF excitation coils and their sensitivities. Now, shorter RF

pulses can be achieved without aliasing in the volume of interest. The selection of

excitation k-space locations to be traversed during RF pulse excitation must be defined

while RF pulse length, excitation fidelity and gradient specifications are kept in mind.

The next section briefly surveys different excitation k-space location selections for

112
selective RF excitation pulse design both in one channel and in multiple channel

systems. From now on, the term k-space will be used to refer to excitation k-space.

5.2.2 Selection of Excitation k-space Locations

The dimension of the traversed k-space specifies the selectivity of the

excitation RF pulse in the image domain. For example, unidirectional slice selection

gradients with sinc RF pulses are widely used to excite spins in a specified slice. Since

slice selection gradients traverse a line in the k-space  0,0, kz  | kz  (kmax , kmax ) ,

their selectivity in the image domain is restricted to one dimension only. Assuming a

uniform transmit coil sensitivity over space, all spins within the selected transverse

slice will be excited. Figure 5.1 demonstrates the excitation k-space traversal of the

slice selection gradient and the location of the excited spins. The k-space extent, kmax,

specifies the resolution and the transition width of the excitation. The sampling

density, Δk, defines where aliased excited spins are localized.

113
Figure 5.1 Schematic illustration of how selectivity in the image domain depends upon the
dimension of excitation k-space. a: A sinc RF excitation pulse with a unidirectional slice
selection gradient along z. b: k-space trajectory traversed by the slice selection gradients while
playing the RF pulse. c: Excited region of the phantom (slice), shown in blue. Using only a
slice selection gradient results in selectivity only in the z dimension in the image domain.

The selective excitation can be extended into two dimensions by extending the

k-space locations to be traversed into two dimensions. In this case, the search space for

k-space locations increases dramatically from a line to a plane. For selectivity in x and

y dimensions, the k-space can be defined as

 k , k , 0 | k  (k
x y x x max 
, k x max ), k y  (k y max , k y max ) . Selective RF pulse design

requires the selection of k-space locations to be traversed with predefined sampling

density, k-space extent, maximum gradient strength, gradient slew rate and RF pulse

length requirements. Just as for the case of traversing 2D imaging k-space in one TR,

114
echo planar, constant angular rate spirals and variable density spirals can be used for

traversing 2D excitation k-space for selective excitation. Constant angular rate spirals

were first demonstrated for 2D selective excitations using the k-space formalism (69).

By traversing the center of the k-space at the end of the RF pulse and weighting the

k-space symmetrically, the excited volume is automatically refocused. Figure 5.2

demonstrates one example of a selective 2D spiral-in k-space trajectory, its

corresponding gradients, and the designed RF pulse (77) intended for selective

excitation of a rectangular region of interest in the middle of a phantom. Since

excitation k-space covers only two dimensions, there is no z-axis selectivity as can be

seen in Figure 5.2d.

The addition of a kz dimension to two dimensional k-space enables three

dimensional selective RF excitation. However, selecting and traversing k-space

locations within three dimensional k-space increases the RF pulse length extensively,

due to the gradient limits imposed by the system as well as peripheral nerve

stimulation limits. Depending on the application requirements, various types of three

dimensional k-space trajectories have been proposed. A stack of spirals (Figure 5.3a)

approach was used for selective inversion recovery pulses for coronary artery imaging

(110) and reduction of susceptibility artifacts in functional MRI (70). Long RF pulse

lengths (in the order of ~20ms) as well as inadequate resolution along the z-direction

hamper the practical application of the abovementioned k-space trajectories.

115
Figure 5.2 An example of 2D spiral RF pulse design. Gradient waveforms (a) are used to
traverse excitation k-space (c). The calculated RF pulse (b) results in selective rectangular
excitation (d) within the x-y plane. However, there is no selectivity in the z-dimension.

The first method to decrease pulse length and increase excitation resolution in

2D RF pulses used multiple shot RF pulse excitations (111,112), and summed

complex images from individual RF excitations yield a final image with full effective

selectivity. This approach takes advantage of the linearity of the STA approximation

(69). Following a similar approach, multi-shot 3D stack of spiral RF pulses were

implemented, resulting in high resolution selective RF excitations with 4 shots of 40

ms RF pulses (113) with an extension to variable density stack of spirals (114) (Figure

116
5.3b). Although the multi shot approach decreases the RF pulse duration, it increases

the acquisition time and is not applicable to large flip angle excitation RF pulses

because of different steady state response of the spins for different shots. With the

development of parallel transmission technology (28,29), it was shown that the

duration of spatially-selective multidimensional RF pulses could potentially be

reduced using multiple RF transmission elements. Similar to parallel imaging, parallel

transmission benefits from the individual coil sensitivities and enables the use of

undersampled k-space without compromising excitation fidelity. By choosing

predefined undersampled spiral (28) and echoplanar (29) k-space trajectories, design

and verification of 2D selective RF pulses was successfully demonstrated. Similarly, a

3D shells trajectory was demonstrated for 3D parallel spatially selective RF pulse

aimed at exciting an arbitrarily shaped region of interest in small animal MR-scanners

(31).

Unlike MR signal acquisition, in which underlying image is a priori unknown,

desired excitation pattern in RF excitation design is one of the parameters used in RF

pulse design. This prior knowledge was first used to determine the kx and ky locations

of a echo-volumar k-space trajectory (Figure 5.3c) via collapsing the 3D power

spectrum, calculated via Fourier transform, of desired pattern along the kz dimension

(115). Incorporating the desired excitation profile information into selection of the

k-space trajectory resulted in significantly reduced RF pulse lengths.

117
Figure 5.3 Various k-space trajectories which are used for 3D selective RF excitation using
one transmit channel. a: Stack of spirals k-space trajectory from Ref. (110), Fig. 2. b: Skip-kz
stack of spirals k-space trajectory from Ref. (113), Fig. 1. c: Echo-volumar k-space trajectory
from Ref. (115), Fig. 1. d: Fast kz-(spokes) k-space trajectory from Ref. (27), Fig. 2.

In addition to reduction of susceptibility artifacts and proper selection of the

excitation ROI, selective excitation RF pulses have been used to reduce B1+

inhomogeneity at high magnetic field strengths. B1+ inhomogeneity correction at 3 T

was achieved by adjusting quadratic in-plane spatial variations of the desired

excitation profile (27). Defining B1+ inhomogeneity as quadratic, a few kx-ky locations

were chosen and shown to be enough for B1+ inhomogeneity correction for the brain

at 3 T field strength (27). This approach resulted in a new type of k-space trajectory

118
using a series of amplitude and phase modulated slice-select subpulses along kz and

phase encoding blips along kx-ky (fast-kz or spokes, Figure 5.3d). When B1+

inhomogeneity is more severe and cannot be well estimated by quadratic terms, e.g. at

7 T, more kx-ky locations are required to obtain homogeneous excitation. However,

additional kx-ky locations increase the RF pulse length. Therefore, both the number of

excitation points and kx-ky locations for slice-selective spokes must be carefully

selected such that B1+ inhomogeneity is mitigated while k-space is traversed within an

acceptable duration.

The addition of parallel transmission into k-space trajectory selection enabled

short RF pulse lengths while mitigating the B1+ inhomogeneity at 7 T. An algorithm

enforcing sparsity in the number of kx-ky locations provided an adequate B1+

inhomogeneity mitigation within a specified excitation FOV in the human brain at 7 T

(116). In addition to B1+ inhomogeneity mitigation, complex target excitation patterns

within a slice were achieved using a sparsity-enforced kx-ky location selection method

(38). This method was shown to outperform Fourier-based (115) and inversion-based

(38) k-space location selection methods. Later, sparse k-space location selection was

employed on 2D parallel transmit RF pulse design (73), and joint methods of

designing k-space trajectories and STA RF pulses simultaneously have emerged with

improved excitation accuracy (117,118).

In this work, the sparse selection of k-space locations for 2D parallel RF pulse

design (73) was extended into 3D selective parallel excitation RF pulse design in order

119
to achieve tailored inner-volume excitations on whole body MRI system at 7 T. The

k-space locations of great importance for achieving the desired excitation profile were

selected by the subspace Orthogonal Matching Pursuit (OMP) method (73) and used

to design STA and LTA RF pulses. The subspace OMP method was compared to a

single-step thresholding method in simulations and phantom experiments.

5.3 Material and Methods

5.3.1 Subspace Orthogonal Matching Pursuit Method

This section revisits the subspace OMP method developed and used for 2D

parallel transmit RF pulse design by Chen et al. (73). In this work, the subspace OMP

method (119) was used to select the most important k-space locations for sparse

parallel transmit 3D RF pulse design.

Theoretical results in sparse signal approximation (120-122) and the

Compressed Sensing method (123) for faster imaging by sparsifying imaging k-space

inspired sparsifying excitation k-space with the addition of parallel transmission

(38,73). By following the notation in Chapter 2, STA parallel transmission RF pulse

design in Cartesian Nyquist sampled k-space can be described in matrix notation,

neglecting the local off-resonance by:

L
m des  c  Dl Ab l [5.5]
l 1

where c  itM 0 is a constant term assuming initial magnetization, M0, is constant

over the desired excitation profile mdes, Dl  diag{Sl (rs )} is a diagonal matrix

120
containing samples of the sensitivity pattern of coil l, bl is the RF pulse waveform of
iri k j
coil l, and A is the Fourier Transform matrix, where aij  e is the Fourier basis

function for the jth Nyquist k-space location. The goal of sparse RF / k-space joint

design is to find the fewest k-space locations that can represent the desired excitation

profile mdes within the specified excitation error tolerance.

A sparse approximation of the design problem can be approached by solving

the L1-regularized least squares problem (38,122) or by greedy style algorithm

(120,121). Depending on the sparsity of desired excitation profile and coil sensitivity

patterns in the Fourier domain, the number of k-space locations required for a given

excitation error tolerance could be drastically smaller than that required by the Nyquist

theorem. By overcoming the Nyquist limit, tailored inner-volume excitations can be

implemented with reasonable RF pulse lengths (for example, shorter than the sample

T2 decay time).

The subspace OMP method uses a greedy type approach to include the next

k-space location that provides the maximum error reduction between the desired and

actual profiles in every iteration. Details of the subspace OMP method can be found in

Ref. (119). The subspace OMP algorithm for sparse 3D k-space locations selection is

briefly outlined below. In Step 2, the subspace OMP algorithm finds the k-space

location which maximizes the projection of A(k)b to the residual. Until convergence

criteria (Step 6) are met, OMP algorithm continues to add the next k-space location

into the k-space location subset, K.

121
 The single-step thresholding method selects all k-space locations in the first

iteration step of subspace OMP algorithm. In Step 2 of the subspace OMP algorithm,

single-step thresholding algorithm chooses nmax k-space locations, from the 3D

Nyquist grid, which have the highest contribution to the desired excitation profile,

mdes, by ordering projections of A(k)b to the residual.

After selecting the k-space locations with subspace OMP and single-step

thresholding methods, k-space trajectories were designed.

122
Algorithm for Subspace OMP algorithm

1. Initialize: n = 1, residual0 = mdes, nmax (maximum allowed k-space locations, to

limit the length of the RF pulse in case defined tolerance "TOL" hasn't met)

and TOL are given

2. Find the next best k-space location which minimizes the norm between residual

and RF weights b of size1 x Nc :

kn  arg min
k 3D Nyquist Grid
 residual n 1  A(k )b 2  for any complex b. Here A(k) is the
Ns x L transmit sensitivity weighted Fourier harmonics generated by k :

A (k )  [ S1  e ik r ,..., S L  e ik r ]

3. Add new k-space location to previously chosen k-space location subset

Κ n 1  k1 , k 2 , k n 1

4. Design the n x L size complex RF weights bn using least squares RF pulse

design with Kn and mdes: b n  arg min m des  A(K n )b

b
 2 
5. Calculate the residual: residualn  m des  A(K n )b n

6. if residualn 2
 TOL or n  nmax then

7. n = n+1; go to Step 2

8. else

9. return Kn with success

123
 5.3.2 k-space Trajectory Design

After the k-space locations for the desired excitation and coil transmit

sensitivity profiles in 3D k-space is determined, a time ordered k-space trajectory

needs to be defined in order to design 3D selective parallel excitation RF pulse. There

are many options on the selection of k-location order. The goal is to minimize the time

needed to traverse the k-space locations. In the presence of local field inhomogeneity,

the ordering of the selected k-space locations critically affects excitation accuracy. For

example, visiting k-space center last results in automatically refocused excitations in

the case of symmetrically weighted k-space (69). Even though the selected k-locations

do not impose symmetry, center of the k-space should be traversed last to ensure

minimum effect due to dephasing and decay of the central k-space components.

Determination of time-ordered k-space locations was achieved by connecting

selected k-space locations either in a suboptimal EPI-like manner or using a genetic

algorithm (124) framed as a modified traveling salesman problem. Specifically, the

traveling salesman problem was modified such that central k-space locations were

sampled last and the distance between two locations was defined as the Euclidean

distance in 3D space. Both ordering methods were restricted such that each k-space

location was visited exactly once. The genetic algorithm was initialized with 60

population size and limited to1000 iterations.

Gradients have inherent maximum amplitude and slew rate limitations. Using

the gradient constraints of maximum amplitude of 40 mT / m and slew rate of

124
120 mT / m / s, gradient waveforms were designed for the chosen k-space locations

ordering based on the method for designing time-optimal gradient waveforms (125).

After designing the gradient waveforms for both k-space location ordering methods,

the shortest length trajectory was chosen for 3D selective excitation RF pulse design.

For fair comparison of subspace OMP and single-step thresholding k-space location

selection methods, the number of selected k-space locations, nmax, is altered for

matching the k-space trajectory lengths of both methods.

5.3.3 Selective Excitation RF Pulse Design

Using the calculated 3D k-space trajectories and desired excitation profile,

parallel excitation RF pulses were designed in STA and LTA regimes. In order to

design in STA regime, the spatial domain parallel RF design method (34), as

explained in Section 1.3.3, was used. Since calculated k-space trajectories do not obey

the linear class assumptions, direct calculation of the RF pulses using LCLTA method

(76) is not feasible. Therefore, the additive angle method (42) was employed for

calculating LTA RF pulses.

Following the notation of Section 1.3.3 and neglecting the local off-resonance,

excited flip angle pattern, θ(r), of the STA RF pulse can be written similar to Eq. [1.7]

as:

L Nt
 it  Sl (r ) bl (t j )e
iM xy ( r ) irk ( t j )
 (r )e [5.6]
l 1 j 1

125
where M xy (r ) is the phase of the transverse magnetization at spatial location r. The

additive angle method includes iterative updates to designed RF pulse and is

initialized by the STA pulse b1 , b 2 ,, b L  . Differences between the desired flip

angle, θdes, and the flip angle pattern resulting from Bloch simulation (78) of RF pulse

b1 , b2 ,, b L  , θ, are used to design a new STA RF pulse b , b ,, b  using the
1 2 L

following cost function:

L 2 L 2
 (b 1 ,..., b R )   Dl Ab l  d new   bl  b l
l 1 l 1
2
[5.7]
2


iM xy ( r )
where d new (r ) [ des (r )   (r )]  e . Adding the phase term into dnew ensures that

the flip angle produced by the calculated pulses b 1 , b 2 ,, b L   will add with the


proper sign. Next iteration will start with the pulses b1  b 1 , b2  b 2 ,, b L  b L  and
the process continues until excitation accuracy stops improving.

5.3.4 Experimental Setup

Experiments were performed on a Siemens whole body 7 T Magnetom scanner

(Erlangen, Germany) equipped with an eight-channel parallel transmit system. An

eight-channel stripline coil array and 7.3-L cylindrical water phantom shown in Figure

1.3 was used in experiments.

126
Figure 5.4 B1+ distribution of the individual elements. a: Axial B1+ amplitude map for each
element of the array. b: Sagittal B1+ amplitude map of transmit channel 2. c: Axial B1+ phase
map for each element of the array. d: Axial B1+ phase map of transmit channel 2.

Multi-slice acquisition for B1+ calibration was performed following the method

described in Ref. (79) and explained in Section 1.3.6. In Figure 5.4a,c, measured

individual channel B1+ magnitude and phase maps are shown in the axial plane

through the isocenter. One representative sagittal B1+ magnitude and phase map of

transmit coil 2 is shown in Figure 5.4b,d. The following imaging parameters were

used in B1+ calibration: FOV = 260 x 260 mm2, echo time (TE) = 1.99 ms, acquisition

matrix = 96 x 96, number of slices = 21, and slice thickness = 8 mm. Total acquisition

time for B1+ profiles in all eight channels was 357 s. ΔB0 was measured using the

phase information from two multi slice GRE images with different TE values

127
(TE1 / TE2 = 5.1 / 4.08 ms) and was incorporated into RF pulse design to compensate

for the phase accrual due to main magnetic field inhomogeneity.

The spatial domain parallel RF design (34) and additive angle method (42)

were used to design parallel RF pulses with a 20° and 90° target flip angle,

respectively. The target excitation flip angle distribution θdes was a homogenous 4 x 2

x 2 cm3 rectangular box profile with axial distribution blurred by convolving with a

Gaussian kernel of FWHM = 1.2 cm to reduce ringing artifacts in the resulting

magnetization distribution. 3D k-space was undersampled by a factor of two to help

both algorithms extend coverage of the outer regions of excitation k-space. In addition

to undersampling k-space, slice resolution of the B1+ maps was reduced from 96 x 96

to 33 x 33 for managing computational cost of k-space locations selection step

efficiently.

Designed RF pulses were simulated using Bloch simulator. Excitation profiles

of RF pulses designed with k-space trajectories calculated from subspace OMP and

single-step thresholding method were compared using NRMSE of the magnetization

and RMSE of the flip angle for STA and LTA designs, respectively. Multi-slice flip

angle profiles of the designed RF pulses were measured using the B1+ calibration

technique (specifically, designed parallel RF pulses were played as saturation pulses

followed by a multishot segmented turbo FLASH acquisition with 2 segments).

Imaging parameters were: FOV = 260 x 260 mm2 TE = 1.97 ms, acquisition matrix =

128 x 128, acquisition time = 168 s, number of slices = 21, and slice thickness = 8

128
mm. In addition to flip angle maps, 3D spoiled GRE using calculated RF pulses as

excitation pulse were acquired for comparing excitation fidelity of both k-space

location selection methods. Longer TRs were used for designed LTA RF pulses in

order to decrease the saturation effects on the excitation profile. The following GRE

imaging parameters were used: FOV = 260 x 260 mm2, TR = 50 ms for STA / 300 ms

for LTA, TE = 7.9 ms, acquisition matrix = 256 x 256 for STA / 128 x 128 for LTA,

number of slices = 48, and slice thickness = 5 mm. During GRE image acquisition, the

net power deposition of designed RF pulses was measured using the power monitoring

setup described in Section 1.3.6.

Figure 5.5 Distribution of the selected k-space locations for both algorithms.

129
 5.4 Results

5.4.1 k-space Trajectory

Selected k-space locations for OMP and single-step thresholding methods are

shown in Figure 5.5 by projecting the 3D k-space along the axis dimensions. Use of

subspace OMP method for selection of the k-space locations resulted in larger k-space

coverage compared to single-step thresholding method. By calculating the

time-optimal gradient waveforms, k-space trajectories that obeyed the system

maximum gradient and slew rates were defined. In Figure 5.6, blue dots represent the

selected k-space locations and red lines represent the designed k-space trajectory. 120

and 200 k-space locations were selected to approximately match RF pulse lengths of

the subspace OMP (8.78ms) and single-step thresholding (8.65ms) methods. These

pulses correspond to ~35 times reduction of the fully sampled 3D Cartesian k-space

trajectory length.

130
Figure 5.6 Designed k-space trajectories for subspace OMP method and single-step
thresholding method

Figure 5.7 Experimental flip angle profiles of designed LTA RF pulses using k-space
trajectories designed with subspace OMP and single-step thresholding method.

131
 5.4.2 Experiments

Using designed k-space trajectories (Figure 5.6), selective excitation parallel

RF pulses were designed for STA and LTA regimes. Prior to experimental application

of the calculated RF pulses, Bloch simulation results of both k-space location selection

methods were compared. According to Bloch simulations, the extension of k-space

coverage achieved with the subspace OMP method resulted in reduced error:

NRMSE / RMSE = 0.011 / 0.26 (subspace OMP), 0.013 / 0.34 (single-step

thresholding).

Figure 5.8 Axial and sagittal GRE images acquired using designed STA (a) and LTA (b)
selective excitation parallel RF pulses. The red circle and rectangle indicates the boundaries of
the phantom.

The flip angle profiles of designed RF pulses were verified in experiments.

Figure 5.7 shows the axial / sagittal flip angle maps of LTA RF pulses designed using

k-space trajectories obtained from subspace OMP and single-step thresholding

132
methods. Experimental flip angle profiles verified that the subspace OMP method

resulted in higher fidelity flip angle distributions compared to the single-step

thresholding method, especially in the axial flip angle distributions.

The axial and sagittal experimental MR signal profiles obtained from subspace

OMP versus the single-step thresholding method are shown in Figure 5.8 a and b for

STA and LTA, respectively. As the excitation flip angle increases, excitations on

undesired locations (where desired excitation flip angle is 0) becomes more

pronounced. Improved STA excitation fidelity of the RF pulse associated with k-

locations selection with subspace OMP method was also associated with a slight

increase in net power deposition (~0.9W compared to ~0.8W for single-step

thresholding). However, the power deposition behavior was reversed for the LTA

case: ~34W for OMP and ~47W for thresholding.

5.5 Discussion

Feasibility of inner-volume excitations with good selectivity was demonstrated

on a whole-body 7T scanner using an eight channel parallel transmit system.

Reasonable RF excitation pulse lengths (~8.7 ms) were realized using multiple

transmit elements and sparse subselection of k-space locations by subspace OMP in

one case and single-step thresholding method in another case. These sparse k-space

trajectories represent ~35 times reduction in RF pulse lengths compared to fully

sampled 3D Cartesian k-space trajectories. This enabled acceleration beyond the limits

of conventional parallel transmission with eight elements, while preserving acceptable

133
excitation profiles. Calculation time to determine k-space locations was greater for the

subspace OMP algorithm since the duration of each sparsifying iteration is

approximately equivalent to the overall duration of the single-step thresholding

approach.

By using the calculated k-space trajectories, flip angle maps and GRE images

using LTA parallel RF pulses were demonstrated, even though the k-space formalism

is only valid in the STA regime as explained in the section 5.2.1. In other words, the

applied k-space trajectories are not necessarily optimal for LTA RF pulse design, but

can still be used to design LTA RF pulses with reasonable inner-volume excitations.

Imperfections in the excitation profile were more pronounced at the locations where

the desired flip angle is zero especially for the LTA RF pulse design. These

imperfections could stem from gradient imperfections, eddy current effects and local

main field inhomogeneities. Some of these effects can be measured, e.g. with field

monitoring (126), and corrected for in RF pulse design in the STA regime (39). The

inability to perfectly null outer volumes with inner-volume excitations is one of the

main limitations of reduced FOV imaging, since excited regions outside of the reduced

FOV will fold into the region of interest. It was shown that parallel imaging

techniques in addition to 2D parallel RF excitation can be used to overcome this

difficulty (127). Future work will incorporate parallel imaging methods into 3D inner

volume excitations in order to overcome unwanted aliasing from excited locations

outside the region of interest.

134
 5.6 Acknowledgements for Chapter 5

I would like to thank Dr. Dong Chen for his help and collaboration on the

subspace OMP method. Dr. Hans-Peter Fautz from Siemens Medical Solutions in

Erlangen, Germany is acknowledged for collaboration on the flip angle mapping

sequence.

135
 CONCLUSION

The need for higher SNR and higher acquisition speeds will continue to drive

demand for UHF-MRI in the future. Nevertheless, many technical challenges remain

to be overcome. The inhomogeneity of the traditionally generated B1+ field and, more

importantly, the increase in SAR per unit flip angle are significant challenges which

continue to obstruct or at least complicate the diagnostic usage of UHF-MRI. These

challenges have forced the MR community to go beyond traditional low-field

approaches and to research new possibilities. Parallel RF excitation techniques offer

significant relieve of UHF challenges by enabling decreases in B1+ inhomogeneity and

SAR. However, parallel RF excitation is a developing technique and continued

progress will be required in order to fully explore the potential of UHF-MRI for

clinical diagnosis.

In this thesis we studied B1+ field behavior and global SAR interactions in the

parallel RF excitation from a systematic perspective. We developed methods to

incorporate measured subject-specific E field interactions into parallel RF excitation

pulse design and RF shimming in order to reduce SAR while maintaining excitation

fidelity. We showed that including E field interactions results in lower global SAR in

phantom and in vivo studies while maintaining / improving the B1+

fidelity / homogeneity. Additionally, we demonstrated the quantitative SNR benefits

of UHF-MRI systems in vivo using developed RF excitation methods. For MR system

monitoring, we proposed a pre-scan-based power calibration technique to estimate

136
subject-specific individual channel power properties of a parallel transmission MRI

system. The proposed technique was used successfully to design parallel RF excitation

pulses obeying strict power limits of the MR system, such as peak and average power.

The importance of coil-subject setup increases at UHF due to SAR concerns. We

analyzed the RF power requirements and SAR of parallel RF excitation systems as a

function of the distance between the transmit coil array and the subject in simulations.

It was found that there are SAR benefits in moving transmit coils away from the

subject. In the last chapter, we utilized the sparse selection of k-space trajectories in

order to design parallel RF excitation pulses for inner-volume excitation. We

demonstrated the feasibility of inner-volume excitations with reasonable RF pulse

lengths in phantom studies.

Recommendations for future work

In this work we demonstrated the benefits of including measurable E field

interactions in parallel RF excitation. The pre-scan power calibration step accounting

for E field interactions uses a power measurement system which includes RF power

sensors and directional couplers. The accuracy of calibration and tracking is expected

to improve as the measurement system is moved closer to the subject. In our current

power measurement setup calibrated E field interactions overestimate the SAR in the

subject. An improved power measurement system could be located close to the

transmit coils, but this improvement could be quite challenging given the need to

operate in the presence of high magnetic fields. A power sensing approach with

137
directional couplers fed into MR receivers was shown to be able to detect changes in

the play out of predefined RF pulses (65). Similar MR receiver based power sensing

apparatus could be used to better estimate SAR inside the subject.

The proposed maximum efficiency RF shimming aims to increase

homogeneity while decreasing RF power deposition in small ROIs for local RF

shimming. The method could be extended to enable homogeneous excitations in larger

ROIs by imposing additional constraints on homogeneity in the calculation of the RF

shimming weights. This may require a different algorithm to find the associated

coefficients.

The capability to predict individual channel forward and reflected power in

parallel RF transmission systems can be used further in parallel RF excitation pulse

design in order to minimize reflected power. Decrease in the reflected power is

desirable from a system perspective since it allows power amplifiers to deliver power

to the transmit coils more efficiently. For the tracking of global SAR as well as

forward and reflected power, separate power measurement systems - one close to the

coils and another at the output of the power amplifiers - may be desirable.

Inner-volume excitations are valuable at UHF as they offer the potential to

reduce image acquisition time or increase spatial resolution over reduced field of view.

Yet, reduced FOV imaging is challenging due to system and B1+ calibration

imperfections, as well as RF pulse design techniques which result in incomplete

suppression of the unexcited regions. Signal from imperfectly suppressed regions can

138
alias into a target reduced FOV and result in significant image artifacts. Reduced FOV

imaging may benefit from combining inner volume excitations with compressed

sensing (128) methods to overcome these problems. In addition, the size of the

excitation profile has been shown to affect the SAR consequences of parallel RF

excitation pulses (51). It would be beneficial for UHF-MRI to investigate the SAR

consequences of inner-volume excitations and, more importantly, to compare with

conventional RF excitation pulses, such as slab selective sinc pulses.

Power prediction and monitoring techniques have been used extensively during

the course of this thesis. Such techniques have been used to decrease global SAR and

to design RF pulse designs conforming to safety limits as well as strict operational

limits of RF power amplifiers. An extension of the global SAR prediction techniques

described here has been shown to enable prediction of the local SAR consequences of

any parallel RF excitation pulse (62,63). The techniques described in this thesis will

also be applicable for local SAR management, which will further enable exploration

and maximization of the benefits of UHF MRI.

139
 APPENDIX

Parallel transmission experiments require knowledge of the B1+ distributions of

individual coil transmit elements in order to tailor the RF excitation as desired. In

addition to B1+ distributions, B0 maps and, if needed, power correlation matrices must

be measured / calibrated before designing parallel transmission RF pulses. Even after

scanner-related-measurements are acquired, RF pulse design requires inputs such as

the desired excitation profile, the choice of RF pulse design method, the excitation k-

space trajectory and so on. Since parallel transmit systems are still in the development

stage, the workflow of obtaining the abovementioned inputs and designing parallel

transmission RF pulses is not yet supported with intuitive graphical user interfaces

(GUIs) of the sort used in clinical MRI scans. The lack of application specific GUIs

for parallel transmit systems results in inefficiencies in the MRI scan workflow, e.g.

longer experiments and operator errors. In order to increase the efficiency and

accuracy of parallel transmit experiments, we developed and used custom-designed

GUIs in the Matlab programming environment in the course of this thesis. These GUIs

can be used not only for parallel transmit experiments but also for transmit coil design

(e.g. using electromagnetic simulation results as inputs to test the suitability of

prospective coil designs) and for educational purposes (e.g. for practice in RF pulse

design). The GUIs described here have been made available for download using the

following web link: http://www.cemnaz.com/~cem/projects/GUI. In this appendix, we

140
describe the workflow of parallel transmit experiments with the guidance of the

developed GUIs.

A.1 RF Shimming GUI

An RF shimming GUI was developed for and used in the maximum efficiency

RF shimming study described in Chapter 2. Figure A.1 shows the workflow of the RF

shimming experiment with numbers in parentheses indicating corresponding locations

in the GUI that can be found in Figure A.2.

B1+ distributions of individual transmit channels can be visualized inside the

GUI after including them from either MR images obtained with turbo FLASH based

flip angle mapping techniques as described in Section 1.3.6 or Matlab .mat files which

contain flip angle information from any imaging method or simulation. Using a data

cursor, flip angles inside figures can be displayed and, if needed, a colorbar can be

included with any figure within the GUI.

RF shimming requires a desired shimming ROI to be defined. Four different

ROI selection mechanisms are implemented as shown in the Shim ROI Select panel

(Figure A.2-2). The shimming ROI can be interactively selected from an additional

MR image or an image obtained with sum of squares (SoS) combination of the B1+

profiles. Additionally, the shimming ROI can be defined as the whole sample or

incorporated from a saved .mat file. After choosing the shimming ROI, the user needs

to specify the coils to be used in RF shimming weights calculation. Initially all eight

coils (the maximal coil set of our 8-channel parallel transmit system) are pre-selected.

141
The user can choose any subset of transmit coils by using check boxes and for

convenience odd and even coils can be selected easily by pressing the corresponding

button from the Select channels panel (Figure A.2-3).

Four different RF shimming methods were implemented in the current GUI.

Two of them are amplitude and phase RF shimming methods indicated by the

RF Shim panel (Figure A.2-4a) and the rest are phase only RF shimming methods

indicated by the Phase Only RF Shim panel (Figure A.2-4b). Push buttons (Figure

A.2-4) initiate calculation of the RF shimming values aiming to match the shim ROI

defined in Figure A.2-2. Pushing the Calculate Shim button calculates RF shimming

weights using a regularization parameter, and if available, the Φ-matrix as explained

in Section 1.3.4 using Eq. [1.14]. When only the magnitude of the shim profile is

targeted, whim weights are calculated by an iterative search algorithm using Matlab's

fminsearch function. Results of both methods are shown in Figure A.2-5. Maximum Tx

Efficiency button calculates the maximum and minimum transmit efficiency RF

shimming values as described in Section 2.3.1. The No Amplitude Target button

calculates unit amplitude phase only shim weights aiming to align the phases of the

transmit elements inside the chosen ROI. In this type of RF shimming the amplitude

of the ROI is not considered but uniform phase distribution is targeted. We

successfully used this shimming approach to obtain Birdcage-type profiles by

prescribing a small ROI at the center of the phantom on a transmit array with

azimuthally distributed individual elements. The With Amplitude Target button

142
calculates the unit amplitude RF shimming weights aiming to match only the

amplitude of the shim ROI. Both Phase Only RF Shim panel algorithms use iterative

search algorithms employing Matlab's fminsearch function. Transmit efficiencies of

the calculated RF shimming weights aligned with minimum and maximum possible

transmit efficiencies are displayed in the Transmit Efficiency panel.

Changing the amplitude and phase of calculated RF shimming weights, which

are displayed in Figure A.2-5, automatically updates the displayed RF shimming

results and transmit efficiency metrics. This feature enables interactive changes to the

calculated RF shimming weights, with immediate visualization. This helps users to

understand the phase and amplitude relationships in a multi-channel transmit system.

If those changes result in unsatisfactory results, the user can press the Go to Original

Shim button to recover previously calculated RF shimming weights. Additionally, RF

shimming weights can be saved to any folder by using the Save to .txt button.

Bloch simulations for adiabatic RF pulses were used in Section 2.3.3 in order

to compare RF power benefits of using the maximum transmit efficiency RF

shimming method (Chapter 3). Choosing an adiabatic RF pulse using Choose RF from

.mat button in Figure A.2-6 enables the Bloch Simulations for Adiabatic RF Pulse

panel. This panel can be used to run Bloch simulations of the chosen adiabatic RF

pulse with the specified maximum voltage. An image of z-magnetization in the sample

resulting from Bloch simulations is shown in Figure A.2-6 with the mean and standard

deviation of the z-magnetization. The frequency response of the adiabatic RF pulse for

143
the given voltage is displayed for the weakest B1+ spatial location with an additional

plot in order to determine whether the adiabatic condition over the ROI is met or not.

In Section 2.3.3, the adiabatic condition in the sample was met by increasing or

decreasing the maximum RF voltage and checking the Bloch simulation results

interactively. Peak and mean power of the RF pulse is predicted and displayed by

including a power correlation matrix, Φ, with the defined calibration voltage.

144
 Obtain B1+ distribution and add to GUI from (1)
an MR experiment (1a) a simulation (1b)

Select targeted RF shimming ROI (2) from

an MR the whole an image obtained with SoS
a saved .mat file
image sample combination of B1+

Select channels to be used in RF shimming (3)

Select type of the RF shimming method to be used (4)

Magnitude and phase RF shimming (4a) Phase only RF shimming (4b)
• Maximum efficiency RF shimming • Aiming to align only the phases of
(Section 2.3.1) transmit elements
• RF shimming aiming target • Aiming to align phases and match the
distribution (Section 1.3.4) uniform amplitude distribution

Visualize and change the calculated RF shimming values (5)

Bloch simulations, if needed, for an RF pulse to obtain frequency response (6)

Use calculated RF shimming coefficients in experiments

Figure A.1 Workflow of an RF shimming experiment. Relation to the RF shimming GUI is

indicated by the numbers in parentheses.

145
146

Figure A.2 Screenshot of RF Shimming GUI

 A.2 Parallel Transmit GUI

A parallel transmit GUI was developed and used for the work reported in

various chapters of the thesis. Figure A.3 shows the workflow of a parallel

transmission experiment with the numbers in the parentheses indicating corresponding

locations in the GUI that can be found in Figure A.4.

B1+ distributions of individual transmit channels can be visualized inside the

GUI after including them from either MR images obtained with turbo FLASH based

flip angle mapping technique as described in section 1.3.6 or .mat files which contain

flip angle information from any imaging method or simulations. Parallel transmission

requires desired excitation profile to be defined. Three different ROI selection

mechanisms are implemented as shown in the Desired Profile Selection panel (Figure

A.4-2). The desired excitation profile can be selected interactively from an image

obtained with sum of squares (SoS) combination of the B1+ profiles. Additionally, the

desired excitation profile can be defined as the whole sample or incorporated from a

saved .mat file. After choosing the desired excitation profile, the user must specify the

coils to be used in parallel transmission RF pulse calculation. Initially all eight coils

are pre-selected. As for RF shimming, the user can choose any subset of transmit coils

by using check boxes and for convenience odd and even coils can be selected easily by

pressing the corresponding button from the Select channels panel (Figure A.4-4). Main

magnetic field inhomogeneities can be included in RF pulse design using the B0 Map

panel in Figure 1.1-3 by including a predefined .mat file or by first choosing names of

147
two GRE images with different TEs from a pop-up menu and then pushing the

Calculate B0 Map button.

Within the pTx Pulse Info panel, the k-space Trajectory panel enables users to

choose the type of excitation k-space trajectory for RF pulse design. Three different

k-space trajectories are implemented:

1. Constant Density Spiral

2. Variable Density Spiral

3. EPI like (Echo Planar)

Constant density spirals were used in Chapters 1 & 3 and variable density

spirals were used in Chapter 1. EPI-like trajectories were not incorporated in RF pulse

design in the thesis, but were implemented in the GUI since they provide a useful

educational perspective for accelerated excitations. Changing parameters in the GUI,

such as excitation resolution, excitation field of view and acceleration, enables

automatic calculation and display of the selected type of k-space trajectory. k-space

trajectory calculation aims for the shortest possible RF pulse length while obeying

gradient specifications defined in the GUI. Additionally, the RF pulse length for the

calculated k-space trajectory is displayed in the k-space Trajectory panel.

Three different RF pulse calculation algorithms are implemented and shown in

the Pulse Design Type panel (Figure A.4-6). Depending on the type of solution, they

are classified as regularized and not regularized. Regularization based RF pulse

designs are implemented in STA and LTA regime. On the other hand, strict constraint

148
RF pulse design (not regularized) is implemented only in the STA regime. Regularized

RF pulse design methods and strict constraint RF pulse design methods were used in

Chapter 1 and 3, respectively. For constrained RF pulse design, the user must choose a

constraint type from the Constraints panel. All options include peak and average

forward and reflected power constraints and global SAR constraints as defined in the

Power Constraints panel. The power correlation matrix must be imported, using the

Include PHI button, for accurate power prediction which is essential in constrained RF

pulse design. Additional parameters for RF pulse design e.g. flip angle, smoothing

profile, and regularization parameter, can be included from Figure A.4-7. After all the

parameters are selected the RF pulse will be designed and displayed in Figure A.4-8

after pushing the Calculate RF Pulse button. Relevant information such as NRMSE

and number of iterations (e.g. conjugate gradient iterations) used in pulse design is

displayed under the amplitude of the designed RF pulse. Calculated RF pulse can be

exported to different formats which can be used in the MR scanner. Write In Float

button generates .float files of individual channel RF pulses and gradients, then puts

them in an ".Output/RFPulses" directory with separate folder names for each

individual transmit channel (TX1-TX8). The WriteIn_pTXRFPulse button generates

the 'pTXRFPulse0.ini' file under the directory ".Output/RFPulses". All variables inside

the GUI can be saved for future reference using the Save Variables button.

149
 Obtain B1+ distribution and add to GUI from (1)
an MR experiment (1a) a simulation (1b)

Select the desired excitation profile (2) from

an image obtained with SoS
a saved .mat file the whole sample
combination of B1+

Include the B0 map (3) from

two GRE MR images precalculated .mat file

Select channels to be used in pTx experiment (5)

Select type of the excitation k-space to be used (4)

Constant Density Spiral Variable Density Spiral EPI

Select type of RF pulse design method (6)

Using regularization (Chapter 1 ) Without using regularization (STA, Chapter 3)
• STA • Unconstrained
• LCLTA • Only global SAR constrained
• Constrained

Define additional parameters for RF pulse design (7)

Calculated RF pulses and Bloch simulation results are shown in (8)

Figure A.3 Workflow of a parallel transmission experiment. Relations to the parallel

transmission GUI are given by the numbers in parentheses.

150
151

Figure A.4 Screenshot of Parallel Transmit GUI

 REFERENCES

1. Purcell EM, Torrey HC, Pound RV. Resonance Absorption by Nuclear

Magnetic Moments in a Solid. Physical Review 1946;69(1-2):37-38.

2. Bloch F. Nuclear Induction. Physical Review 1946;70(7-8):460-474.

3. Rabi II, Zacharias JR, Millman S, Kusch P. A New Method of Measuring

Nuclear Magnetic Moment. Physical Review 1938;53(4):318-318.

4. Lauterbur PC. Image Formation by Induced Local Interactions: Examples

Employing Nuclear Magnetic Resonance. Nature 1973;242(5394):190-191.

5. Kumar A, Welti D, Ernst RR. NMR Fourier zeugmatography. Journal of

Magnetic Resonance (1969) 1975;18(1):69-83.

6. Twieg D. The k-trajectory formulation of the NMR imaging process with

applications in analysis and synthesis of imaging methods. Med Phys
1983;10(5):610.

7. Ljunggren S. A simple graphical representation of fourier-based imaging

methods. Journal of Magnetic Resonance (1969) 1983;54(2):338-343.

8. Mansfield P. Multi-planar image formation using NMR spin echoes. Journal of

Physics C: Solid State Physics 1977;10(3):L55.

152
9. Hennig J, Nauerth A, Friedburg H. RARE imaging: A fast imaging method for
clinical MR. Magnetic Resonance in Medicine 1986;3(6):823-833.

10. Ahn CB, Kim JH, Cho ZH. High-Speed Spiral-Scan Echo Planar NMR
Imaging-I. Medical Imaging, IEEE Transactions on 1986;5(1):2-7.

11. Hutchinson M, Raff U. Fast MRI data acquisition using multiple detectors.
Magnetic Resonance in Medicine 1988;6(1):87-91.

12. Sodickson DK, Manning WJ. Simultaneous acquisition of spatial harmonics

(SMASH): Fast imaging with radiofrequency coil arrays. Magnetic Resonance
in Medicine 1997;38(4):591-603.

13. Pruessmann KP, Weiger M, Scheidegger MB, Boesiger P. SENSE: Sensitivity

encoding for fast MRI. Magnetic Resonance in Medicine 1999;42(5):952-962.

14. Jakob PM, Griswold MA, Edelman R, Sodickson DK. AUTO-SMASH: A self-
calibrating technique for SMASH imaging. Magnetic Resonance Materials in
Physics, Biology and Medicine 1998;7(1):42-54.

15. Griswold MA, Jakob PM, Nittka M, Goldfarb JW, Haase A. Partially parallel
imaging with localized sensitivities (PILS). Magnetic Resonance in Medicine
2000;44(4):602-609.

16. Griswold MA, Jakob PM, Heidemann RM, Nittka M, Jellus V, Wang J, Kiefer
B, Haase A. Generalized autocalibrating partially parallel acquisitions
(GRAPPA). Magn Reson Med 2002;47(6):1202-1210.

153
17. Tsao J, Boesiger P, Pruessmann KP. k-t BLAST and k-t SENSE: dynamic
MRI with high frame rate exploiting spatiotemporal correlations. Magn Reson
Med 2003;50(5):1031-1042.

18. Huang F, Akao J, Vijayakumar S, Duensing GR, Limkeman M. k-t GRAPPA:

a k-space implementation for dynamic MRI with high reduction factor. Magn
Reson Med 2005;54(5):1172-1184.

19. Hoult DI. Sensitivity and power deposition in a high-field imaging experiment.
J Magn Reson Imaging 2000;12(1):46-67.

20. Center for Devices and Radiological Health. Guidance for the Submission Of
Premarket Notifications for Magnetic Resonance Diagnostic Devices. Food
and Drug Administration; 1998. p 23.

21. International Electrotechnical Commission. Medical electrical equipment - Part

2-33: Particular requirements for the basic safety and essential performance of
magnetic resonance equipment for medical diagnosis. IEC 60601-2-33 ed3.0;
2010.

22. Vaughan JT, Hetherington HP, Otu JO, Pan JW, Pohost GM. High frequency
volume coils for clinical NMR imaging and spectroscopy. Magnetic
Resonance in Medicine 1994;32(2):206-218.

23. Ibrahim TS, Lee R, Baertlein BA, Abduljalil AM, Zhu H, Robitaille PM.
Effect of RF coil excitation on field inhomogeneity at ultra high fields: a field
optimized TEM resonator. Magn Reson Imaging 2001;19(10):1339-1347.

154
24. Metzger GJ, Snyder C, Akgun C, Vaughan T, Ugurbil K, Moortele P-FVd.
Local B1+ shimming for prostate imaging with transceiver arrays at 7T based
on subject-dependent transmit phase measurements. Magnetic Resonance in
Medicine 2008;59(2):396-409.

25. Deniz CM, Brown R, Alon L, Sodickson DK, Zhu Y, Lattanzi R. MRI of the
Hip at 7T using RF shimming with 4-channel excitation. 2011; Lake Louise

26. Saekho S, Boada FE, Noll DC, Stenger VA. Small tip angle three-dimensional
tailored radiofrequency slab-select pulse for reduced B1 inhomogeneity at 3 T.
Magnetic Resonance in Medicine 2005;53(2):479-484.

27. Saekho S, Yip CY, Noll DC, Boada FE, Stenger VA. Fast-kz three-
dimensional tailored radiofrequency pulse for reduced B1 inhomogeneity.
Magnetic Resonance in Medicine 2006;55(4):719-724.

28. Katscher U, Bornert P, Leussler C, van den Brink JS. Transmit SENSE.
Magnetic Resonance in Medicine 2003;49(1):144-150.

29. Zhu Y. Parallel excitation with an array of transmit coils. Magnetic Resonance
in Medicine 2004;51(4):775-784.

30. Setsompop K, Wald LL, Alagappan V, Gagoski B, Hebrank F, Fontius U,

Schmitt F, Adalsteinsson E. Parallel RF transmission with eight channels at 3
Tesla. Magnetic Resonance in Medicine 2006;56(5):1163-1171.

155
31. Schneider J, Kalayciyan R, Haas M, Ruhm W, Doessel O, Hennig J, Ullmann
P. Inner-Volume-Imaging Using Three-Dimensional Parallel Excitation.
Proceedings 18th Scientific Meeting, International Society for Magnetic
Resonance in Medicine. Stockholm; 2010. p 103.

32. Setsompop K, Wald LL, Alagappan V, Gagoski BA, Adalsteinsson E.

Magnitude least squares optimization for parallel radio frequency excitation
design demonstrated at 7 Tesla with eight channels. Magn Reson Med
2008;59(4):908-915.

33. Griswold M, Kannengiesser S, Muller M, Jakob P. Autocalibrated Accelerated

Parallel Excitation (Transmit-GRAPPA). In Proceedings of the 13th Scientific
Meeting, ISMRM. Miami; 2005. p 2435.

34. Grissom W, Yip CY, Zhang Z, Stenger VA, Fessler JA, Noll DC. Spatial
domain method for the design of RF pulses in multicoil parallel excitation.
Magn Reson Med 2006;56(3):620-629.

35. Kerr AB, Zhu Y, Pauly JM. Phase constraint relaxation in parallel excitation
pulse design. In Proceedings of the 18th Scientific Meeting, ISMRM. Berlin;
2007. p 1694.

36. Brunner DO, Pruessmann KP. Optimal design of multiple-channel RF pulses

under strict power and SAR constraints. Magn Reson Med 2010;63(5):1280-
1291.

37. Cloos MA, Boulant N, Luong M, Ferrand G, Giacomini E, Le Bihan D,

Amadon A. kT-points: Short three-dimensional tailored RF pulses for flip-

156
 angle homogenization over an extended volume. Magnetic Resonance in
Medicine 2012;67(1):72-80.

38. Zelinski AC, Wald LL, Setsompop K, Goyal VK, Adalsteinsson E. Sparsity-
enforced slice-selective MRI RF excitation pulse design. IEEE Transactions on
Medical Imaging 2008;27(9):1213-1229.

39. Schneider JT, Haas M, Ruhm W, Hennig J, Ullmann P. Robust spatially

selective excitation using radiofrequency pulses adapted to the effective
spatially encoding magnetic fields. Magnetic Resonance in Medicine
2011;65(2):409-421.

40. Xu D, King KF, Zhu Y, McKinnon GC, Liang ZP. A noniterative method to
design large-tip-angle multidimensional spatially-selective radio frequency
pulses for parallel transmission. Magn Reson Med 2007;58(2):326-334.

41. Xu D, King KF, Zhu Y, McKinnon GC, Liang ZP. Designing multichannel,
multidimensional, arbitrary flip angle RF pulses using an optimal control
approach. Magn Reson Med 2008;59(3):547-560.

42. Grissom WA, Yip CY, Wright SM, Fessler JA, Noll DC. Additive angle
method for fast large-tip-angle RF pulse design in parallel excitation. Magnetic
Resonance in Medicine 2008;59(4):779-787.

43. Setsompop K, Alagappan V, Gagoski B, Witzel T, Polimeni J, Potthast A,

Hebrank F, Fontius U, Schmitt F, Wald LL, Adalsteinsson E. Slice-selective
RF pulses for in vivo B1+ inhomogeneity mitigation at 7 tesla using parallel

157
 RF excitation with a 16-element coil. Magnetic Resonance in Medicine
2008;60(6):1422-1432.

44. Wu X, Vaughan JT, Uğurbil K, Van de Moortele P-F. Parallel excitation in the
human brain at 9.4 T counteracting k-space errors with RF pulse design.
Magnetic Resonance in Medicine 2010;63(2):524-529.

45. Malik SJ, Keihaninejad S, Hammers A, Hajnal JV. Tailored excitation in 3D

with spiral nonselective (SPINS) RF pulses. Magnetic Resonance in Medicine
2011:doi: 10.1002/mrm.23118.

46. Graesslin I, Niemann M, Harvey P, Vernickel P, Katscher U. SAR and RF

power reduction with parallel excitation using Non-Cartesian Trajectories.
MAGMA. Volume 18; 2005. p S251.

47. Liu Y, Feng K, McDougall MP, Wright SM, Ji J. Reducing SAR in parallel
excitation using variable-density spirals: a simulation-based study. Magn
Reson Imaging 2008;26(8):1122-1132.

48. Homann H, Graesslin I, Nehrke K, Findeklee C, Dössel O, Börnert P. Specific

absorption rate reduction in parallel transmission by k-space adaptive
radiofrequency pulse design. Magnetic Resonance in Medicine
2011;65(2):350-357.

49. Katscher U, Röhrs J, Börnert P. Basic considerations on the impact of the coil
array on the performance of Transmit SENSE. MAGMA 2005;18(2):81-88.

158
50. Deniz CM, Lattanzi R, Zhu Y, Wiggins G, Sodickson DK. RF Energy
Deposition and RF Power Requirements in Parallel Transmission with
Increasing Distance from the Coil to the Sample. In Proceedings of the 17th
Scientific Meeting, ISMRM. Honolulu; 2009. p 4802.

51. Zelinski AC, Angelone LM, Goyal VK, Bonmassar G, Adalsteinsson E, Wald
LL. Specific absorption rate studies of the parallel transmission of inner-
volume excitations at 7T. J Magn Reson Imaging 2008;28(4):1005-1018.

52. Cloos MA, Luong M, Ferrand G, Amadon A, Le Bihan D, Boulant N. Local

SAR reduction in parallel excitation based on channel-dependent Tikhonov
parameters. Journal of Magnetic Resonance Imaging 2010;32(5):1209-1216.

53. Lee J, Gebhardt M, Wald LL, Adalsteinsson E. Local SAR in parallel

transmission pulse design. Magnetic Resonance in Medicine 2011:doi:
10.1002/mrm.23140.

54. Sbrizzi A, Hoogduin H, Lagendijk JJ, Luijten P, Sleijpen GLG, van den Berg
CAT. Fast design of local N-gram-specific absorption rate–optimized
radiofrequency pulses for parallel transmit systems. Magnetic Resonance in
Medicine 2011:doi: 10.1002/mrm.23049.

55. Waldby C. The Visible Human Project : informatic bodies and posthuman
medicine. London; New York: Routledge; 2000.

56. Christ A, Kainz W, Hahn EG, Honegger K, Zefferer M, Neufeld E, Rascher

W, Janka R, Bautz W, Chen J, Kiefer B, Schmitt P, Hollenbach H-P, Shen J,
Oberle M, Szczerba D, Kam A, Guag JW, Kuster N. The Virtual Family—

159
 development of surface-based anatomical models of two adults and two
children for dosimetric simulations. Physics in Medicine and Biology
2010;55(2):N23.

57. Homann H, Börnert P, Eggers H, Nehrke K, Dössel O, Graesslin I. Toward

individualized SAR models and in vivo validation. Magnetic Resonance in
Medicine 2011;66(6):1767-1776.

58. Voigt T, Homann H, Katscher U, Doessel O. Patient-individual local SAR

determination: In vivo measurements and numerical validation. Magnetic
Resonance in Medicine 2011:doi: 10.1002/mrm.23322.

59. Alon L, Deniz CM, Sodickson DK, Zhu Y. Do constraints on |B1+| also
constrain |E| and SAR in high field MR? In Proceedings of the 19th Scientific
Meeting, ISMRM. Montreal; 2011. p 491.

60. Zhu Y. In Vivo RF Power and SAR Calibration for Multi-Port RF

Transmission. In Proceedings of the 17th Scientific Meeting, ISMRM.
Honolulu; 2009. p 2585.

61. Alon L, Deniz CM, Lattanzi R, Wiggins G, Brown R, Sodickson DK, Zhu Y.
An Automated Method for Subject Specific Global SAR Prediction in Parallel
Transmission. In Proceedings of the 18th Scientific Meeting, ISMRM.
Stockholm; 2010. p 780.

62. Alon L, Deniz CM, Lattanzi R, Wiggins G, Brown R, Sodickson DK, Zhu Y.
Local SAR Calibration and Prediction Model in Parallel Transmit MRI. In

160
 Proceedings of the 18th Scientific Meeting, ISMRM. Stockholm; 2010. p
3869.

63. Alon L, Deniz CM, Xu J, Brown R, Sodickson DK, Zhu Y. Volumetric Local
SAR Mapping for Parallel Transmission. In Proceedings of the 19th Scientific
Meeting, ISMRM. Montreal; 2011. p 3854.

64. Graesslin I, Biederer S, Falaggis K, Vernickel P, Dingemans H, Mens G,

Roeschmann P, Leussler C, Zhai Z, Morich M, Katscher U. Real-time SAR
Monitoring to ensure Patient Safety for Parallel Transmission Systems. In
Proceedings of the 17th Scientific Meeting, ISMRM. Berlin; 2007. p 1086.

65. Gagoski B, Gumbrecht R, Hamm M, Setsompop K, Keil B, Lee J, Makhoul K,

Mareyam A, Fujimoto K, Witzel T, Frontius U, Pfeuffer J, Adalsteinsson E,
Wald LL. Real time RF monitoring in a 7T parallel transmit system. In
Proceedings of the 18th Scientific Meeting, ISMRM. Stockholm; 2010. p 781.

66. Zhu Y, Alon L, Deniz CM, Brown R, Sodickson DK. System and SAR
Characterization in Parallel RF Transmission. Magnetic Resonance in
Medicine 2012;67(5):1367-1378.

67. Haas M, Snyder J, Schneider J, Ullmann P, Kokorin D, Fautz HP, Hennig J,

Zaitsev M. Selective Excitation of Arbitrary Three-Dimensional Targets on a
Human MR System using Parallel Transmit. 2011; Montreal. p 4433.

68. Deniz CM, Chen D, Alon L, Brown R, Fautz H-P, Sodickson DK, Zhu Y.
Sparse Parallel Transmit Excitation Trajectory Design for Rapid Inner-Volume

161
 Excitation. In Proceedings of the 19th Scientific Meeting, ISMRM. Montreal;
2011. p 4434.

69. Pauly J, Nishimura D, Macovski A. A k-space analysis of small-tip-angle

excitation. J Magn Reson 1989;81(1):43-56.

70. Stenger VA, Boada FE, Noll DC. Three-dimensional tailored RF pulses for the
reduction of susceptibility artifacts in T2*-weighted functional MRI. Magn
Reson Med 2000;44(4):525-531.

71. Rieseberg S, Frahm J, Finsterbusch J. Two-dimensional spatially-selective RF

excitation pulses in echo-planar imaging. Magn Reson Med 2002;47(6):1186-
1193.

72. Lattanzi R, Sodickson DK, Grant AK, Zhu Y. Electrodynamic constraints on

homogeneity and radiofrequency power deposition in multiple coil excitations.
Magn Reson Med 2009;61(2):315-334.

73. Chen D, Bornemann F, Vogel M, Sacolick L, Kudielka G, Zhu Y. Sparse

Parallel Transmit Pulse Design Using Orthogonal Matching Pursuit Method. In
Proceedings of the 17th Scientific Meeting, ISMRM. Honolulu; 2009. p 171.

74. Zelinski AC, Goyal VK, Angelone L, Bonmassar G, Wald LL, Adalsteinsson
E. Designing RF Pulses with Optimal Specific Absoption Rate (SAR)
Characteristics and Exploring Excitation Fidelity, SAR and Duration Trade-
offs. In Proceedings of the 15th Scientific Meeting, ISMRM. Berlin; 2007. p
1699.

162
75. Hoult DI. The principle of reciprocity in signal strength calculations - A
mathematical guide. Concepts in Magnetic Resonance 2000;12(4):173-187.

76. Pauly J, Nishimura D, Macovski A. A linear class of large-tip-angle selective

excitation pulses. J Magn Reson 1989;82(3):571-587.

77. Yip CY, Fessler JA, Noll DC. Iterative RF pulse design for multidimensional,
small-tip-angle selective excitation. Magn Reson Med 2005;54(4):908-917.

78. Pauly J, Le Roux P, Nishimura D, Macovski A. Parameter relations for the

Shinnar-Le Roux selective excitation pulse design algorithm. IEEE Trans Med
Imaging 1991;10(1):53-65.

79. Fautz HP, Vogel M, Gross P, Kerr A, Zhu Y. B1 Mapping of Coil Arrays for
Parallel Transmission. In Proceedings of the 16th Scientific Meeting, ISMRM.
Toronto; 2008. p 1247.

80. Kim D, Adalsteinsson E, Spielman DM. Simple analytic variable density spiral
design. Magn Reson Med 2003;50(1):214-219.

81. Deniz CM, Alon L, Lattanzi R, Sodickson DK, Zhu Y. SAR Benefits of
Including E-field Interactions in Parallel RF Pulse Design. In Proceedings of
the 18th Scientific Meeting, ISMRM. Stockholm; 2010. p 4930.

82. Mao W, Smith MB, Collins CM. Exploring the limits of RF shimming for
high-field MRI of the human head. Magnetic Resonance in Medicine
2006;56(4):918-922.

163
83. van den Bergen B, van den Berg CAT, Bartels LW, Lagendijk JJW. 7 T body
MRI: B1 shimming with simultaneous SAR reduction. Physics in Medicine
and Biology 2007(17):5429.

84. Deniz CM, Alon L, Brown R, Sodickson DK, Zhu Y. Specific Absorption Rate
Benefits of Including Measured Electric Field Interactions in Parallel
Excitation Pulse Design. Magnetic Resonance in Medicine 2012(67):164-174.

85. Orzada S, Maderwald S, Poser BA, Bitz AK, Quick HH, Ladd ME. RF
excitation using time interleaved acquisition of modes (TIAMO) to address B1
inhomogeneity in high-field MRI. Magnetic Resonance in Medicine
2010;64(2):327-333.

86. Balchandani P, Khalighi MM, Hsieh SS, Setsompop K, Pauly J, Spielman D.

Adiabatic B1 Shimming Algorithm for Multiple Channel Transmit at 7T. In
Proceedings of the 19th Scientific Meeting, ISMRM. Montreal; 2011. p 2907.

87. Setsompop K, Wald LL, Adalsteinsson E. Reduced-Voltage RF Shimming for

Adiabatic Pulse Design in Parallel Transmission. In Proceedings of the 15th
Scientific Meeting, ISMRM. Berlin; 2007. p 1687.

88. Ganz R, Parvizi J, Beck M, Leunig M, Nötzli H, Siebenrock KA.

Femoroacetabular Impingement: A Cause for Osteoarthritis of the Hip.
Clinical Orthopaedics and Related Research 2003;417:112-120.

89. Beck M, Leunig M, Parvizi J, Boutier V, Wyss D, Ganz R. Anterior

Femoroacetabular Impingement: Part II. Midterm Results of Surgical
Treatment. Clinical Orthopaedics and Related Research 2004;418:67-73.

164
90. Kassarjian A. Hip MR Arthrography and Femoroacetabular Impingement.
Semin Musculoskelet Radiol 2006;10(03):208,219.

91. Mechlenburg I, Nyengaard JR, Gelineck J, Soballe K. Cartilage thickness in

the hip joint measured by MRI and stereology – a methodological study.
Osteoarthritis and Cartilage 2007;15(4):366-371.

92. Zhu Y, Deniz CM, Alon L, Fautz H-P, Sodickson DK. Undestanding Parallel
Transmit Array Efficiency. ISMRM. Stockholm; 2010. p 1518.

93. Brown R, Stoeckel B, Sodickson DK, Wiggins G. Investigation of Element

Designs and Construction of a Reconfigurable 8 Channel Tx, 16 Channel Rx
Torso Array for 7T. ISMRM. Stockholm; 2010. p 3807.

94. Kumar A, Bottomley P. Optimized quadrature surface coil designs. Magnetic

Resonance Materials in Physics, Biology and Medicine 2008;21(1):41-52.

95. Garwood M, DelaBarre L. The Return of the Frequency Sweep: Designing

Adiabatic Pulses for Contemporary NMR. Journal of Magnetic Resonance
2001;153(2):155-177.

96. Reeder SB, Wen Z, Yu H, Pineda AR, Gold GE, Markl M, Pelc NJ. Multicoil
Dixon chemical species separation with an iterative least-squares estimation
method. Magnetic Resonance in Medicine 2004;51(1):35-45.

165
97. Kellman P, McVeigh ER. Image reconstruction in SNR units: A general
method for SNR measurement†. Magnetic Resonance in Medicine
2005;54(6):1439-1447.

98. van den Bergen B, van den Berg CAT, Klomp DWJ, Lagendijk JJW. SAR and
power implications of different RF shimming strategies in the pelvis for 7T
MRI. Journal of Magnetic Resonance Imaging 2009;30(1):194-202.

99. Bittersohl B, Hosalkar HS, Hughes T, Kim Y-J, Werlen S, Siebenrock KA,
Mamisch TC. Feasibility of T2* mapping for the evaluation of hip joint
cartilage at 1.5T using a three-dimensional (3D), gradient-echo (GRE)
sequence: A prospective study. Magnetic Resonance in Medicine
2009;62(4):896-901.

100. Sur S, Mamisch TC, Hughes T, Kim Y-J. High resolution fast T1 mapping
technique for dGEMRIC. Journal of Magnetic Resonance Imaging
2009;30(4):896-900.

101. Hanke M. On Lanczos Based Methods for the Regularization of Discrete Ill-
Posed Problems. BIT Numerical Mathematics 2001;41(5):1008-1018.

102. Golub GH, Loan CFV. Matrix computations: Johns Hopkins University Press
1996.

103. Sturm J. Primal-Dual Interior Point Approach to Semidefinite Programming:

Tinbergen Institute; 1997.

166
104. Löfberg J. YALMIP : A Toolbox for Modeling and Optimization in
{MATLAB}. 2004.

105. Tai CT. Dyadic Green Functions in Electromagnetic Theory: Institute of

Electrical & Electronics Engineers; 1994.

106. Lattanzi R, Sodickson DK. Dyadic Green’s functions for electrodynamic

calculations of ideal current patterns for optimal SNR and SAR. In
Proceedings of the 16th Scientific Meeting, ISMRM. Toronto; 2008. p 78.

107. Wiesinger F, Boesiger P, Pruessmann KP. Electrodynamics and ultimate SNR

in parallel MR imaging. Magn Reson Med 2004;52(2):376-390.

108. Nyquist H. Certain Topics in Telegraph Transmission Theory. Transactions of

the A I E E 1928:617-644.

109. Shannon CE. Communication in the Presence of Noise. Proceedings of the IRE
1949;37(1):10-21.

110. Pauly JM, Hu BS, Wang SJ, Nishimura DG, Macovski A. A three-dimensional
spin-echo or inversion pulse. Magnetic Resonance in Medicine 1993;29(1):2-6.

111. Hardy CJ, Bottomley PA. 31P Spectroscopic localization using pinwheel NMR
excitation pulses. Magnetic Resonance in Medicine 1991;17(2):315-327.

167
112. Panych LP, Oshio K. Selection of high-definition 2D virtual profiles with
multiple RF pulse excitations along interleaved echo-planar k-space
trajectories. Magnetic Resonance in Medicine 1999;41(2):224-229.

113. Stenger VA, Boada FE, Noll DC. Multishot 3D slice-select tailored RF pulses
for MRI. Magnetic Resonance in Medicine 2002;48(1):157-165.

114. Stenger VA, Boada FE, Noll DC. Variable-density spiral 3D tailored RF
pulses. Magnetic Resonance in Medicine 2003;50(5):1100-1106.

115. Yip C-Y, Fessler JA, Noll DC. Advanced three-dimensional tailored RF pulse
for signal recovery in T2*-weighted functional magnetic resonance imaging.
Magnetic Resonance in Medicine 2006;56(5):1050-1059.

116. Zelinski AC, Wald LL, Setsompop K, Alagappan V, Gagoski BA, Goyal VK,
Adalsteinsson E. Fast slice-selective radio-frequency excitation pulses for
mitigating B1+ inhomogeneity in the human brain at 7 Tesla. Magnetic
Resonance in Medicine 2008;59(6):1355-1364.

117. Yip C-Y, Grissom WA, Fessler JA, Noll DC. Joint design of trajectory and RF
pulses for parallel excitation. Magnetic Resonance in Medicine
2007;58(3):598-604.

118. Ma C, Xu D, King KF, Liang Z-P. Joint design of spoke trajectories and RF
pulses for parallel excitation. Magnetic Resonance in Medicine
2011;65(4):973-985.

168
119. Chen D. Acceleration of the Spatial Selective Excitation of MRI via Sparse
Approximation. Munich: Technische Universitaet Munchen; 2010.

120. Mallat SG, Zhifeng Z. Matching pursuits with time-frequency dictionaries.

Signal Processing, IEEE Transactions on 1993;41(12):3397-3415.

121. Pati YC, Rezaiifar R, Krishnaprasad PS. Orthogonal matching pursuit:

recursive function approximation with applications to wavelet decomposition.
1993 1-3 Nov 1993. p 40-44 vol.41.

122. Chen S. Atomic Decomposition by Basis Pursuit. SIAM J Sci Comput

1998;20(1):33.

123. Lustig M, Donoho D, Pauly JM. Sparse MRI: The application of compressed
sensing for rapid MR imaging. Magnetic Resonance in Medicine
2007;58(6):1182-1195.

124. Kirk J. Traveling Salesman Problem - Genetic Algorithm. 2009.

125. Lustig M, Seung-Jean K, Pauly JM. A Fast Method for Designing Time-
Optimal Gradient Waveforms for Arbitrary k-Space Trajectories. Medical
Imaging, IEEE Transactions on 2008;27(6):866-873.

126. Barmet C, Zanche ND, Pruessmann KP. Spatiotemporal magnetic field

monitoring for MR. Magnetic Resonance in Medicine 2008;60(1):187-197.

169
127. Zhao L, Madore B, Panych LP. Reduced field-of-view MRI with two-
dimensional spatially-selective RF excitation and UNFOLD. Magnetic
Resonance in Medicine 2005;53(5):1118-1125.

128. Lustig M, Donoho DL, Santos JM, Pauly JM. Compressed Sensing MRI [A
look at how CS can improve on current imaging techniques]. Signal Processing
Magazine, IEEE 2008;25(2):72-82.

170