Acta Oncologica

ISSN: 0284-186X (Print) 1651-226X (Online) Journal homepage: https://www.tandfonline.com/loi/ionc20

Magnetic resonance only workflow and validation

of dose calculations for radiotherapy of prostate
cancer

Rasmus L. Christiansen, Henrik R. Jensen & Carsten Brink

To cite this article: Rasmus L. Christiansen, Henrik R. Jensen & Carsten Brink (2017) Magnetic
resonance only workflow and validation of dose calculations for radiotherapy of prostate cancer,
Acta Oncologica, 56:6, 787-791, DOI: 10.1080/0284186X.2017.1290275

To link to this article: https://doi.org/10.1080/0284186X.2017.1290275

View supplementary material

Published online: 20 Feb 2017.

Submit your article to this journal

Article views: 1279

View related articles

View Crossmark data

Citing articles: 10 View citing articles

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=ionc20
ACTA ONCOLOGICA, 2017
VOL. 56, NO. 6, 787–791
http://dx.doi.org/10.1080/0284186X.2017.1290275

ORIGINAL ARTICLE

Magnetic resonance only workflow and validation of dose calculations for

radiotherapy of prostate cancer
Rasmus L. Christiansena, Henrik R. Jensena and Carsten Brinka,b
a
Laboratory of Radiation Physics, Department of Oncology, Odense University Hospital, Odense, Denmark; bInstitute of Clinical Science,
University of Southern Denmark, Odense, Denmark

ABSTRACT ARTICLE HISTORY

Background: Current state of the art radiotherapy planning of prostate cancer utilises magnetic reson- Received 29 December 2016
ance (MR) for soft tissue delineation and computed tomography (CT) to provide an electron density Accepted 29 January 2017
map for dose calculation. This dual scan workflow is prone to setup and registration error. This study
evaluates the feasibility of an MR-only workflow and the validity of dose calculation from an MR
derived pseudo CT.
Material and methods: Thirty prostate cancer patients were CT and MR scanned. Clinical treatment
plans were generated on CT using a single 18 MV arc volumetric modulated arc therapy (VMAT) with a
prescription of 78 Gy/39 fractions. Dose was recalculated on pseudo CT and assuming uniform water
density. Pseudo CT and uniform density based dose calculations were compared to CT dose calcula-
tions by gamma analysis. One patient was treated with a plan based solely on MR and pseudo CT
including daily image guided radiotherapy (IGRT) performed by manual match of implanted gold
markers.
Results: A pseudo CT was generated for 29 of the 30 patients. Median gamma pass rates for 1%/1 mm
passing criteria for dose calculated on pseudo CT when compared to CT were 100% for most evaluated
structures. Dose calculated on uniform density also yielded high median pass rates, but with a higher
occurrence of pass rates below 95%. Cases of pass rate below 95% on pseudo CT proved to originate
from the presence of rectal air on CT, not represented by the pseudo CT. Treatment based on MR
alone was successfully delivered to one patient, including manually performed daily IGRT.
Conclusions: Median gamma pass rates were high for pseudo CT and proved superior to uniform
density. Local differences in dose calculations were concluded not to have clinical relevance. Feasibility
of the MR-only workflow was demonstrated through successful delivery of a treatment course planned
based on MR alone.

Background recommended (or indeed available) for clinical use. However,

Magnetic resonance (MR) images provide the gold standard
for defining soft tissue structures during radiotherapy plan-
ning for prostate cancer [1–3]. However, in order to calculate
a dose distribution the electron density map of a computed
tomography (CT) scan is required. This necessitates a dual
modality workflow and image co-registration which is cur-
rently standard in many clinics. However, this workflow intro-
duces the risk of setup error between the scans as well as
registration error. An MR-only workflow eliminates set-up and
registration error while also reducing workload and strain on
the patient. The use of MR guided treatment delivery in the
years to come provides further argument for use of an MR-
only workflow [4,5]. Therefore, several groups have under-
taken work to create an electron density map from MR; a
pseudo CT [6–11]. These studies evaluate different
approaches to pseudo CT generation and report high
degrees of similarities of dose calculations compared to CT.
Most of these are still under development and not explicitly
Ko€hler et al. [10] report on the same pseudo CT algorithm as
evaluated in the current study published by the manufac-
turer, investigating only the high dose volume and not spe-
cific organs at risk (OAR). A consensus on evaluating and
reporting these evaluations does not currently exist.
Furthermore, different institutions apply different treatment
techniques which may impact on the dose calculations,
depending on the given pseudo CT. Therefore, a validation
of dose calculated on the specific pseudo CT in concordance
with the treatment technique used is necessary. To the
extent of our knowledge, previous studies have until now
only been performed for the target region and the whole
body volume (with a low-dose threshold), thus not sufficient
for clinical implementation. This study investigates the valid-
ity of dose calculations based on the first commercially avail-
able pseudo CT derived from MR in comparison to CT
without any modifications to the pseudo CT and with respect
to target volumes and specific OAR. Furthermore, the

CONTACT Rasmus L€ ubeck Christiansen Rasmus.Lubeck.Christiansen@rsyd.dk Laboratory of Radiation Physics, Department of Oncology, Odense University
Hospital, Odense, Denmark
Supplemental data for this article can be accessed here.
ß 2017 Acta Oncologica Foundation
788 R. L. CHRISTIANSEN ET AL.
planning and treatment workflow were validated end-to-end
to evaluate the feasibility of MR-only planning.
Material and methods
Patients
Thirty prostate cancer patients referred for localised, curative
radiotherapy (78 Gy in 39 fractions) were included in the
study. All patients included in the study were referred for CT
and MR scans on the same day as part of the standard pro-
cedure. Written informed consent to acquire the additional
MRCAT sequence during the MR scan and approval for use in
the current study was obtained from all patients.
Pseudo CT generation
Magnetic Resonance for Calculating Attenuation (MRCAT)
(Philips, Helsinki, Finland) is a commercially available solution
for generation of pseudo CT for dose calculation purposes
for prostate cancer. MRCAT consists of a T1 weighted (T1w),
dual echo mDixon scan sequence providing the source
images for the subsequent reconstruction algorithm which
outputs a pseudo CT.
The MRCAT pseudo CT image is not identical to a CT as it
comprises five discrete Hounsfield Units (HU) ( 968, 86, 42,
198 and 949) to discriminate between air, fat, water/muscle,
spongious bone and cortical bone (Figure 1). The MRCAT
reconstruction algorithm utilises the information of the
mDixon MR images to adapt a generic anatomical model to
the individual patients. The resulting pseudo CT does not
represent non-tissue densities within the patient outline, e.g.,
volumes of air or metallic implants, such as prosthetics or
fiducial markers.
Planning procedure
The planning workflow for included patients was as per the
standard protocol, except addition of the MRCAT scan proto-
col to MR. Dose comparisons were performed retrospectively.
CT scans were performed on a Philips Big Bore Brilliance 16
slice CT scanner (Philips, Best, The Netherlands) with 140 kVp
and reconstructed to a slice thickness of 3 mm, 512  512
matrix. MR scans were performed on a Philips Ingenia 1.5 T
MR scanner (Philips, Best, The Netherlands). Both scanners
Figure 1. CT (left) comprises a range of 4000 HU, whereas the MRCAT generated pseudo CT represents the range of attenuations with five HU bins. A gold fiducial
marker and rectal air are visual on CT in contrast to pseudo CT.
were equipped for radiotherapy with a flat, indexed table
top, knee-and foot fixation device (Bionix, Toledo, OH) and
external laser positioning system (LAP, Lu€neburg, Germany).
All acquired MR sequences are listed in Table S1.
Small FOV T2w MR scans were rigidly co-registered with
CT for the purpose of delineating the target structures and
selected OAR. The CT was used to delineate the remaining
OAR.
Treatments were planned on CT in Pinnacle v9.10 (Philips,
Milwaukee, WI, USA) as single arc volumetric modulated arc
therapy (VMAT) with an 18 MV beam. All dose calculation
was based on the standard CT to density conversion table in
our clinical treatment planning system, thus no correction of
this table was performed for the pseudo CT data.
Comparison of dose calculations
To minimise errors in dose calculation related to setup uncer-
tainties, the planning CT was rigidly co-registered to the
pseudo CT using Elastix [12] and reformatted to fit identically
in terms of voxel size, slice positioning and field of view
(FOV) of the pseudo CT. In this process, the voxel dimensions
of the CT were changed from 0.976 mm  0.976 mm  3 mm
to 1.04 mm  1.04 mm  2.5 mm.
Beam parameters, points and regions of interest (ROI) and
dose grid of the clinical plan were transferred to three image
sets: pseudo CT, the reformatted clinical CT and a version of
the pseudo CT where the body was assigned to a uniform
density of 1 g/cm3. The dose was recalculated with a dose
grid resolution of 3 mm on each image set, yielding three
dose distributions for each patient. For each patient, the simi-
larity of dose distribution calculated based on both pseudo CT
and uniform density, respectively, were evaluated against that
of CT using pass rate of the 3D Gamma index, as described
by Low et al. [13]. The dose acceptance criterion was eval-
uated as a 2% or 1% percentage of the prescribed dose and
distance criteria of 2 mm and 1 mm, respectively. The calcula-
tions were performed by in-house developed MatLab script
(Matlab version 2014a, The MathWorks Inc., Natick, MA). The
calculation was performed for the entire scanned volume,
thus no dose threshold was used in the current study.
Verification of workflow end-to-end
To verify that the ultimate goal of delivering a full course of
treatment planned solely on MR is feasible, the entire
 ACTA ONCOLOGICA 789

planning process was performed by MR alone for one
patient. A full FOV T2 weighted MR provided the basis for
OAR delineations. Prior to treatment delivery, the clinical
dose calculation based on pseudo CT was verified against CT
by setting up beam parameters, etc. and recalculating the
dose, similar to the workflow described above. Fiducial
markers were identified on MR and delineated for daily setup
verification with cone beam CT (CBCT).
Results
Of the 30 patients included, 29 had successfully generated
pseudo CTs which were included in the analysis.
Results of gamma analysis are summarised for the patient
cohort in Tables 1 and 2. Median values were 100% for all
evaluated structures under 2% and 2 mm passing criteria
both for pseudo CT and uniform density based calculations.
Under the stricter passing criteria of 1% and 1 mm, median
values were also 100% for all structures calculated on the
pseudo CT. A few individual structures yielded pass rates
below 95%. These incidents were linked to rectal air being
present on CT, but not on the pseudo CT (see Figure 2).
Dose calculations based on uniform density resulted in
median gamma pass rates of 97.5% and above for all eval-
uated structures under 1% and 1 mm criteria. However, a
large portion of evaluated structures had lower pass rates
than for the pseudo CT data, as illustrated by the 16.7–83.3
percentiles in Table 1.
Visual inspection of gamma index overlay on CT (as
shown in Figure 2) for each patient also revealed that gener-
ally areas near the skin surface were prone to failure.
For one patient, all delineations of target and OAR were
performed on MRI for dose calculation and plan optimisation
on pseudo CT. Daily online verification of patient setup was
performed by CBCT prior to treatment delivery. Gold seeds
implanted in the prostate visual on the CBCT were manually
matched to the corresponding delineations of the markers
from the planning MR (Figure S1).
Discussion
This study has evaluated the quality of dose calculation
based on an unmodified commercially available pseudo CT
to that of CT. Evaluation was performed by analysis of the
gamma index for the entire scanned volume as well as indi-
vidual target and OAR to assess clinical impact of MR-only
radiotherapy planning. Subsequently, one patient was treated
based on MR images alone to prove the feasibility of an MR-
only planning workflow.
A pseudo CT could not be calculated for one patient but
it was not possible to investigate why this was so within the
current study as we did not have access to the pseudo CT
source code. However, this was most likely related to a
‘sanity check’ performed in the software, evaluating whether
the obtained pseudo CT seems plausible. If not no output is
generated. The cause of the non-successful generation of
pseudo CT was thus likely an atypical anatomy of that
patient.
Very high median pass rates were seen for both pseudo
CT and uniform density for 2% and 2 mm passing criteria,
although outliers were less prominent for pseudo CT. This
tendency was more pronounced when the stricter 1%/1 mm
passing criteria were applied; where pseudo CT still showed
consistently high pass rates for all patients, uniform density
yielded several gamma index failures, as illustrated by the
ranges of 16.7–83.3 percentiles of Table 1. The small ranges

Table 1. Median gamma pass rates shown for selected structures for CT vs pseudo CT and CT vs uniform density dose calculations.
16.7–83.3 percentiles mimicking 1 standard deviation, had data been normally distributed are shown in brackets.
2% and 2 mm 1% and 1 mm
Pseudo CT Uniform density Pseudo CT Uniform density
PTV78 100 (99.7–100) 100 (98.4–100) 99.9 (96.7–100) 97.1 (56.5–100)
Prostate 100 (100–100) 100 (99.9–100) 100 (98.7–100) 98.5 (23.7–100)
Seminal vesicles 100 (100–100) 100 (100–100) 100 (99.3–100) 100 (71.1–100)
Rectum 100 (99.1–100) 100 (98.2–100) 100 (94.9–100) 100 (89.9–100)
Small bowel incl. sigmoid 100 (99.6–100) 100 (99.9–100) 100 (97.5–100) 100 (97.6–100)
Bladder 100 (100–100) 100 (100–100) 100 (99.9 –100) 100 (95.8–100)
Penile bulb 100 (100–100) 100 (100–100) 100 (100–100) 100 (100–100)
Femoral heads 100 (100–100) 100 (100–100) 100 (100–100) 100 (99.1–100)
Body 99.7 (99.3–99.8) 99.7 (99.4–99.9) 98.9 (98.3–99.2) 98.9 (98.2–99.3)

Table 2. Mean gamma pass rates (and their standard deviation) shown for selected structures for CT vs pseudo CT and CT vs uniform
density dose calculations for comparison with results of other groups, although data are not normally distributed.
2% and 2 mm 1% and 1 mm
Pseudo CT Uniform density Pseudo CT Uniform density
PTV78 99.8 (0.5) 98.9 (2.7) 97.3 (7.1) 82.5 (24.3)
Prostate 99.9 (0.3) 97.6 (7.7) 97.4 (10.4) 74.3 (35.5)
Seminal vesicles 99.9 (0.3) 99.8 (0.7) 96.7 (10.0) 88.7 (19.3)
Rectum 99.2 (2.7) 98.7 (3.2) 96.8 (7.0) 95.4 (8.6)
Small bowel incl. sigmoid 99.7 (1.0) 98.2 (8.8) 98.9 (2.3) 95.9 (17.8)
Bladder 100.0 (0.0) 100.0 (1.8) 98.9 (4.8) 96.2 (9.9)
Penile bulb 100.0 (0.0) 100.0 (0.0) 100.0 (0.0) 99.0 (5.5)
Femoral heads 100.0 (0.0) 99.9 (0.6) 100.0 (0.2) 97.6 (10.2)
Body 99.5 (0.4) 99.5 (0.7) 98.7 (0.9) 98.1 (3.0)
790 R. L. CHRISTIANSEN ET AL.
Figure 2. Gamma index (1%/1 mm) shown as coloured overlay on CT.
of 16.7–83.3 percentiles of pseudo CT based calculations indi-
cate that calculations are accurate on patient level, not just
the cohort.
For pseudo CT calculation, gamma failures were seen pri-
marily in conjunction with rectal air being present on CT but
not pseudo CT. Given that the full treatment course is deliv-
ered over 39 fractions, identical occupation of the rectal vol-
ume by air is unlikely. Therefore, it is likely that it is at least
as good as an approximation to overwrite these moving air
pockets with water density rather than assuming stationary
air throughout the treatment course.
Failures of the gamma index in the skin areas occurred pri-
marily due to the inevitable setup error between CT and MR.
Furthermore, the volumes of failure are small in comparison
to the entire body volume (5%), and so is the absolute
dose deposited in these areas. Therefore these gamma fail-
ures are not considered to lessen the validity of the MRCAT-
CT based dose calculation. On the other hand, when a thresh-
old is applied, such as Korhonen et al. [9], Dowling et al. [11]
and Ko €hler et al. [10] do, the effect is not at all evaluated.
There are variations in the method of reporting gamma
values among studies. Some groups use mean instead of
median gamma pass rates although data are unlikely to be
normally distributed. For the purpose of comparison to those
results, mean pass rates of the gamma indexes in this study
are given in Table 2. The dual model HU conversion of
Korhonen et al. [9] yielded mean 2D gamma pass rates for
VMAT plans of 99.9% for 2% and 2 mm passing criteria and
93.2% for 1% and 1 mm for the D10% volume. The mean
pass rate of 98.7% for 1% and 1 mm found in the current
study for the entire body is superior to this. However, accord-
ing to Pulliam et al. [14], their use of a 2D gamma analysis is
likely to result in slightly lower pass rates than if a 3D ana-
lysis had been performed, such as in the current study.
Dowling et al. [11] evaluated dose calculations based on
pseudo CT generated from a standard T2w sequence and
reported mean 3D gamma pass rates of 100% for the D90%
volume (corresponding roughly to the PTV78 in the present
study) under 2% and 2 mm as well as 1% and 1 mm criteria.
Mean pass rates for D10% were reported to be 95% (2% and
2 mm) and 93% (1% and 1 mm). As in the present study, a
global dose was used as reference for the gamma index cal-
culation, although it is unclear whether they also chose the
prescription dose.
The MRCAT generated pseudo CT was evaluated by the
manufacturer on clinical data resulting in 99.83% gamma
pass rate of D75% for 1% and 1 mm criteria. In their study,
differences in outer contour between CT and pseudo CT
were overwritten with water density to minimise the effect
of setup and registration error [10]. Similarly, Dowling et al.
[11] added 1 mm to the patient circumference to adjust a
drop in signal on the skin surface [11]. In contrast to these
studies, no changes in the obtained pseudo CT were made
in the present study, as the purpose was to test the validity
of the dose calculation based on the pseudo CT generated
by the commercially available software package. For the
same reason, the gamma index was calculated and reported
for the entire scanned volume.
This validation of the dose calculation utilising electron
density information from the pseudo CT was undertaken
for a specific treatment technique; 18 MV full arc VMAT,
which will de-emphasise local HU errors, since only a small
part of the dose is delivered in a specific direction. The
high dose volume is placed distant from the skin, thus this
will not be as sensitive to small differences in the defin-
ition of the outer contour of the patient. The same argu-
ments to some extent explain why homogeneous water
performs as well as it does in the comparison. Other treat-
ment sites and techniques will require individual
validations.
For a daily online setup verification to be practical, an
automatic match of CBCT to the planning image should be
possible. Currently, the vendor does not provide such a solu-
tion. To manage this, an in-house developed solution is thus
required and has now been developed locally to facilitate an
automatic match in the future workflow.
In conclusion, dose calculated based on MRCAT pseudo
CT was practically indistinguishable from those of CT and as
such considered approved for clinical application for prostate
cancer with the described treatment technique. Gamma pass
rates were consistently higher for pseudo CT than uniform
density when passing criteria of 1% and 1 mm are applied.
One patient was successfully treated with a dose plan gener-
ated from MR only, demonstrating the practical feasibility of
an MR only work flow for prostate cancer treatment.
However, automatic matching between CBCT and pseudo CT
was not possible due to implanted fiducials not being shown
by the pseudo CT.
Disclosure statement
The authors report no conflicts of interest.
 ACTA ONCOLOGICA 791

References [8] Dowling JA, Lambert J, Parker J, et al. An atlas-based electron

density mapping method for magnetic resonance imaging (MRI)-
[1] Rasch C, Barillot I, Remeijer P, et al. Definition of the prostate in alone treatment planning and adaptive MRI-based prostate radi-
CT and MRI: a multi-observer study. Int J Radiat Oncol Biol Phys. ation therapy. Int J Radiat Oncol Biol Phys. 2012;83:e5–e11.
1999;43:57–66. [9] Korhonen J, Kapanen M, Keyril€ainen J, et al. A dual model HU
[2] Khoo VS, Joon DL. New developments in MRI for target volume conversion from MRI intensity values within and outside of bone
delineation in radiotherapy. Br J Radiol. 2006;79:S2–S15. segment for MRI-based radiotherapy treatment planning of pros-
[3] Dirix P, Haustermans K, Vandecaveye V. The value of magnetic tate cancer. Med Phys. 2014;41:011704.
resonance imaging for radiotherapy planning. Semin Radiat [10] Ko€hler M, Vaara T, van Grootel M, et al. MR-only simulation for
Oncol. 2014;24:151–159. radiotherapy planning. Koninklijke Philips N V. 2015;1–15.
[4] Lagendijk JJW, Raaymakers BW, van Vulpen M. The magnetic res- [11] Dowling JA, Sun J, Pichler P, et al. Automatic substitute com-
onance imaging-linac system. Semin Radiat Oncol. puted tomography generation and contouring for magnetic res-
2014;24:207–209. onance imaging (MRI)-alone external beam radiation therapy
[5] Mutic S, Dempsey JF. The ViewRay system: magnetic resonance- from standard MRI sequences. Int J Radiat Oncol Biol Phys.
guided and controlled radiotherapy. Semin Radiat Oncol. 2015;93:1144–1153.
2014;24:196–199. [12] Klein S, Staring M, Murphy K, et al. Elastix: a toolbox for intensity-
[6] Chen L, Price RA, Wang L, et al. MRI-based treatment planning for based medical image registration. IEEE Trans Med Imaging.
radiotherapy: dosimetric verification for prostate IMRT. Int J 2010;29:196–205.
Radiat Oncol Biol Phys. 2004;60:636–647. [13] Low DA, Harms WB, Mutic S, et al. A technique for the quantita-
[7] Johansson A, Karlsson M, Nyholm T. CT substitute derived from tive evaluation of dose distributions. Med Phys. 1998;25:656–661.
MRI sequences with ultrashort echo time. Med Phys. 2011;38: [14] Pulliam KB, Huang JY, Howell RM, et al. Comparison of 2D and
2708–2714. 3D gamma analyses. Med Phys. 2014;41:021710.