New Experimental verification of 4D Monte Carlo simulations of …eheath/Publications/... · 2017....
Transcript of New Experimental verification of 4D Monte Carlo simulations of …eheath/Publications/... · 2017....
1
Experimental verification of 4D Monte Carlo simulations of dose delivery to a
moving anatomy
S. Gholampourkashi Carleton University, 1125 Colonel By Drive, Ottawa, Ontario K1S 5B6, Canada
M. Vujicic Department of Medical Physics, The Ottawa Hospital Cancer Centre, 501 Smyth Road, Box 927, Ottawa, Ontario K1H 8L6, Canada
J. Belec Department of Medical Physics, The Ottawa Hospital Cancer Centre, 501 Smyth Road, Box 927, Ottawa, Ontario K1H 8L6, Canada
Joanna. E. Cygler Carleton University, 1125 Colonel By Drive, Ottawa, Ontario K1S 5B6, Canada
Department of Medical Physics, The Ottawa Hospital Cancer Centre, 501 Smyth Road, Box 927, Ottawa, Ontario K1H 8L6, Canada
E. Heath Carleton University, 1125 Colonel By Drive, Ottawa, Ontario K1S 5B6, Canada
Purpose: To evaluate a novel 4D Monte Carlo simulation tool by comparing calculations to physical
measurements using a respiratory motion phantom.
Methods: We used a dynamic Quasar phantom in both stationary and breathing states (sinusoidal motion of
amplitude of 1.8 cm and period of 3.3 sec) for dose measurements on an Elekta Agility linear accelerator.
Gafchromic EBT3 film and the RADPOS 4D dosimetry system were placed inside the lung insert of the
phantom to measure dose profiles and point-dose values at the center of the spherical tumour inside the insert.
Both a static 4x4 cm2 field and a VMAT plan were delivered. Static and 4D Monte Carlo simulations of the
treatment deliveries were performed using DOSXYZnrc and a modified version of the defDOSXYZnrc user
code that allows modeling of the continuous motion of both machine and patient. DICOM treatment plan files
and linac delivery log files were used to generate corresponding input files. The phantom motion recorded by
RADPOS during beam delivery was incorporated into the input files for the 4DdefDOSXYZnrc simulations.
Results: For stationary phantom simulations, all point-dose values from MC simulations at the tumour center
agreed within 1% with film and within 2% with RADPOS. More than 98% of the voxels from simulated dose
profiles passed a 1D gamma of 2%/2 mm criteria against measured dose profiles. Similar results were
observed when applying 2D gamma of 2%/2 mm to compare 2D dose distributions of Monte Carlo
simulations against measurements. For simulations on the moving phantom, MC simulated dose values at the
center of the tumour were found to be within 1% of film and within 2 of experimental uncertainties (2.8%)
of RADPOS measurements. 1D gamma comparisons of the dose profiles were better than 91% and 2D gamma
comparisons of the 2D dose distributions were found to be better than 94%.
Conclusion: Our 4D Monte Carlo method using defDOSXYZnrc can be used to accurately calculate the dose
distribution in continuously moving anatomy for various treatment techniques. This work, if extended to
2
deformable anatomies, can be used to reconstruct patient delivered dose for the purpose of adaptive
radiotherapy techniques.
Key words: Monte Carlo, respiratory motion, 4D dose calculation, VMAT
I. INTRODUCTION
The ultimate goal of radiation therapy is to deliver a planned dose to a target volume while minimizing dose to
normal tissue. This objective, however, is limited by the fact that the dose distribution from the treatment plan is
based on a single state of the patient anatomy while patient’s anatomy and position often vary between treatment
fractions (inter-fraction motion) or during one fraction (intra-fraction motion).1-3
As a result of these positional
uncertainties, a significant deviation could be observed between the dose prescribed and received by a target
volume.3
Respiratory motion is an intra-fraction motion that contributes largely to the geometric uncertainties in
thoracic and abdominal sites.2-4
The main impact of the respiratory motion is the blurring of the dose distribution
along the path of the motion. The amount of this blurring is independent of the treatment technique and depends
on the properties of the respiratory motion.1,2
Localized deformations in the dose distribution are another effect
that occur due to the motion and deformation of the internal anatomy during respiration and can result in up to 5%
error in the localized dose values.1,5
When dynamic beam delivery techniques are used, an additional effect caused
by the interplay between motion of the machine components (e.g. multi-leaf collimators) and target motion is
observed inducing further deviation of 1%-10% in the delivered dose compared to the prescribed dose1,2,6-13
Many researchers have developed methods to account for these effects in the calculation of dose distributions.
One early approach is convolution14-16
of the static dose distribution with a probability distribution function (PDF)
that describes the characteristics of the respiratory motion. Although this approach sufficiently models the blurring
effect of the respiratory motion, it is limited by certain assumptions. One assumption is that the dose distribution
is shift invariant which results in inaccuracies at tissue interfaces.17,18
To accurately model the spatial dependence
of the dose distribution, a fluence-convolution method to convolve the incident beam fluence with the PDF of the
motion was suggested and implemented into Monte Carlo-based dose calculation algorithms.17,19,20
However, the
limitation that remained unresolved is that the patient motion can only be modeled as rigid body translation and
deformations in the anatomy are not considered.1,17
A more accurate approach to calculate the dose distribution, while accounting for the effects of respiratory
motion, is to accumulate the dose distributions calculated on different respiratory phases. This approach accounts
for the motion and deformation of the anatomy. The concept of 4D dose accumulation was initially introduced by
Brock et al.21
. Many other groups4,22-26
have investigated the reconstruction of accumulated dose in a deforming
anatomy using deformable image registration (DIR) to map dose calculated in each respiratory phase to a
3
reference (planning) phase. Although DIR provides a more realistic modeling of the respiratory motion, these
groups did not model the interplay effect in their dose calculations. Studies conducted by Rao et al.13
and
Paganetti et al.27
investigated 4D dose calculations with interplay effect during lung treatments. The first study
(Rao et al.) focused on IMRT and VMAT deliveries and determined the beam segment corresponding to each
respiratory phase. The dose from each segment was calculated on the appropriate respiratory phase. The second
study (Paganetti et al.) modeled proton therapy delivery. They used a Monte Carlo (MC) simulation method
where both the beam delivery and dose calculation geometries were continuously updated as a function of time
spent in each respiratory phase.
All of the previously mentioned 4D dose accumulation studies made use of dose interpolation methods in
combination with DIR to map the dose back to the reference geometry.22-27
Previous studies have assessed the
accuracy of dose interpolation methods for dose accumulation.28-31
Deformable image registration algorithms
might split or merge voxels while determining the correlation between images and thus cause a lack of a one-to-
one voxel correspondence between each respiratory phase and the reference phase (i.e. dose calculation grid is not
conserved). As pointed out by Siebers et al.28
, this can lead to inaccuracies when dose interpolation is used in
regions of dose or density gradients. More accurate approaches are the energy/mass transfer (EMT)28,29
and voxel
warping methods (VWM)30
. EMT uses deformation vector fields (DVFs) generated by DIR to map energy
deposited on each respiratory phase to a reference phase. VWM uses DVFs to deform voxels from the reference
geometry to reproduce each respiratory phase. Both methods assure conservation of the energy deposited as well
as the mass on each respiratory phase and as a result a more accurate cumulative dose calculation on the reference
phase.
A previous study by Belec et al.32
used the EMT method adopted from Siebers et al.28
and Zhong et al.29
to
calculate accumulated dose distributions from VMAT lung treatments. They modeled the continuous motion of
the beam during VMAT treatments using a position-probability sampling (PPS) technique.33
A randomly sampled
time variable between 0 and 1 was associated with each particle. This time variable was used to sample linac
geometry settings specified as a function of normalized cumulative MU. The time variable was then used to
interpolate DIR vectors and the voxel densities to more closely model changes associated with continuous motion.
The authors acknowledged the limitation of using this linear interpolation of the voxel densities and attempted to
minimize its impact by reducing the motion between respiratory phases to be smaller than voxel sizes. They
reported results from comparison of simulation with measurements for a sweeping field delivered to a respiratory
motion phantom moving a sinusoidal motion with 2 cm amplitude and 8 sec period. Delivery log files recorded at
a sampling interval of 250 ms were captured using an iCom listening software. The log files contained planned
and actual values for gantry, multi-leaf collimator leaves, jaws and table settings, cumulative MU delivered, dose
rate, and linac status (beam holds). The authors also applied the same MC method to calculate cumulative
delivered dose for VMAT plans for several lung patients. For these studies plan information was extracted from
the treatment planning system (TPS) and MC calculated dose were compared against the TPS dose.
4
Recently, we have developed a 4D Monte Carlo simulation tool34
for reconstructing the delivered dose
accounting for respiratory motion. This approach uses the voxel warping method and delivery log files to model
continuous motion of machine components and patient voxel boundaries. The aim of the current work was to
validate this simulation tool using a lung motion phantom. Dose delivery was performed on an Elekta Agility
linac for both static and dynamic (VMAT) delivery techniques. Dose distributions from Monte Carlo simulations
were compared with measurements to evaluate the accuracy of simulations. We demonstrate how motion recorded
during dose delivery as well as the delivery log files, using a 4D dosimetry system, are incorporated into our
simulations.
II. MATERIALS AND METHODS
II. A. Measurements
II. A.1. Quasar respiratory motion phantom
Measurements were performed in a Quasar respiratory motion phantom (Modus Medical, London, ON,
Canada), containing a cylindrical lung insert made of cedar wood with a 3 cm diameter solid water sphere to
imitate a tumour (Fig. 1).
FIG. 1. (a) The Quasar respiratory motion phantom, (b) wooden lung insert, (c) solid water tumour.
The Quasar phantom simulates the breathing motion by allowing the lung insert to move rigidly in one
direction according to a programmed motion function. A single, reproducible, sinusoidal motion function was
used for all experiments.
II. A.2. 3DCT acquisition
3DCT scans of the Quasar phantom were acquired using a helical CT scanner (Brilliance CT Big Bore,
Philips, Amsterdam, Netherlands). Reconstructed images had a standard resolution of 0.0625 0.0625 0.1 cm3
resulting in an image matrix of 512 512 246 voxels. The center of the sphere inside the lung insert was
(a) (b)
3 cm
(c)
5
positioned at the zero amplitude position. Pitch values and the gantry rotation times for the scans were 0.95 0.75
sec, respectively.
II. A.3. Treatment plans
Two treatment plans using a 6 MV photon beam were created on the static 3DCT scans of the Quasar phantom
to deliver 100 cGy to the center of the tumour. A GTV was created by delineating the tumour. No margins were
added to compensate for motion. Both treatment plans were designed to cover the GTV. The first treatment plan
was a static 4 4 cm2 square field that was created using XiO V.4.7 (Elekta AB, Stockholm, Sweden). The
resultant dose distribution, calculated using a superposition algorithm, is shown in Fig. 2.
FIG. 2. Static 4 4 cm2 square plan from XiO: dose distributions on (a) axial, (b) coronal and (c) sagittal planes.
The second treatment plan was a VMAT plan with 45 control points that was created in Monaco V.5.10.02.
The start and stop angles of the VMAT delivery arc were 180o with angular spacing of 8
o between most control
points. The dose calculation algorithm used by Monaco is XVMC35
(X-ray voxel Monte Carlo). A 2 mm dose
calculation grid with a statistical uncertainty of 1% was used to calculate the dose to water (Dw) to comply with
measurements (film and RADPOS) and MC calculations. Figure 3 shows the dose distribution corresponding to
the VMAT treatment on the static 3DCT image of the Quasar phantom.
(a) (b)
(c)
6
FIG. 3 Static VMAT plan from Monaco: dose distributions on (a) axial, (b) coronal and (c) sagittal planes.
II. A.4. Dose measurements: Film and RADPOS
Dose measurements inside the Quasar lung insert were performed using calibrated Gafchromic film (EBT3,
Ashland, Wayne, NJ, United States) and the RADPOS 4D dosimetry system36,37
. The RADPOS probe consists of
a microMOSFET dosimeter and an electromagnetic positioning sensor and is capable of simultaneous real-time
monitoring of dose and position. The MOSFET dosimeter and position sensor are separated by 8 mm to minimize
the perturbation of the incident radiation fluence at the MOSFET location.
RADPOS dose measurements are based on measuring the change in the threshold voltage (∆Vth) of the
MOSFET detector before and after irradiation. Hence, a calibration coefficient is required to convert readings in
mV to absorbed dose in cGy. The position sensor, preamplifier, transmitter and 3D-guidance tracker are
responsible for position tracking of RADPOS. The RADPOS transmitter generates a 3D DC magnetic field. This
signal is received by the position sensor and its response to the signal is amplified through a pre-amplifier and
monitored by the guidance tracker to determine the spatial coordinates of the RADPOS probe. More details about
the system can be found in the publication by Cherpak et al.37
The EBT3 film and RADPOS detector used in this work were cross-calibrated against an A1SL ionization
chamber (Standard imaging Inc., Middleton, WI, USA) in a 6 MV photon beam at a field size of 10 10 cm2. Film
and RADPOS were placed at a depth of 5 cm and the A1SL chamber was placed at a depth of 10 cm in a Solid
(a) (b)
(c)
7
Water (RMI457 Gammex, Wisconsin, USA) phantom at an SSD of 100 cm. After appropriate irradiations (three
irradiations of 100 MU for RADPOS and eight irradiations of 0, 5, 25, 50, 75, 119, 178, 240 MU’s for film), dose
to water (Dw) at the position of RADPOS and film was determined using the PDD curves available in the clinic.
The calibration coefficient ( for RADPOS was derived by calculating the ratio of the average dose
delivered to RADPOS (Dw) to the average change in the threshold voltage (∆Vth). Films were scanned (pre and
post-irradiation) using an Epson (10000XL, Suwa, Naogna, Japan) scanner. The film analysis and calibration (to
convert film readings to Dw) was performed according to the procedures described in Devic et al.38
using an in-
house MATLAB (R2013a, Mathworks, MA, USA) script.
For dose measurements with the Quasar phantom, film and RADPOS were placed inside the lung insert of the
phantom as shown in Fig. 4. The RADPOS probe was fixed in a special groove that was cut into the spherical
solid water tumour and the lung insert so that its point of measurement was located at the center of the tumour.
Film was placed and taped on top of the RADPOS probe.
FIG. 4. Film and RADPOS inside the lung insert of the Quasar phantom. RADPOS was placed inside a special groove and film was
placed on the top of RADPOS probe.
The total dosimetric uncertainty of RADPOS measurements was determined to be 1.40%, including the
uncertainties in the dosimeter calibration and beam delivery conditions. Sources that contributed to the dosimeter
calibration uncertainty (1.39%) included uncertainties corresponding to IC calibration (NDw from clinical data
~1%, kQ from literature39
), IC reading (0.16%), RADPOS reproducibility (0.48%) and Solid Water phantom
material variability40
. With respect to the beam delivery conditions, uncertainties corresponding to SSD, depth and
field size settings as well as temperature and pressure correction factors were the main contributors to the total
uncertainty of 0.25%.39
The total dosimetric uncertainty of the film measurements was determined to be 2.3%.
Items such as scanner uniformity, lateral correction, film inhomogeneity, energy and angular dependence, fit
accuracy (netOD to Dose) and intra-batch variations were considered to be main contributions to uncertainty
associated with EBT3 films (1.83%).41
Regarding the calibration, total uncertainty (1.3%) was determined based
on parameters such as IC calibration (NDw from clinical data ~1%, kQ from literature39
) and Solid Water phantom
material variability40
. The uncertainty related to beam delivery (0.25%) was determined to be similar for film and
Film
RADPOS probe
8
RADPOS measurements. Total dosimetric uncertainties of film and RADPOS measurements were obtained by
quadrature sum of all contributing sources of uncertainty.
II. A.5. Quasar phantom irradiations
The treatment plans described in section II.A.3 were transferred to the Elekta MOSAIQ RadOnc system and
delivered to the Quasar phantom on an Agility linac (Elekta AB., Stockholm, Sweden). The 4 4 cm2 square plan
delivered 123.2 MU at a nominal dose rate of 600 . The VMAT plan delivered 172.3 MU at a varying
dose rate. Delivery log files (IAN V.2, Elekta AB., Stockholm, Sweden) containing cumulative delivered MU,
gantry, multi-leaf collimator, jaws and table positions at 40 ms intervals were saved after each delivery and
processed as explained in section II.B.1.
For both treatment plans, irradiations were performed with the Quasar in stationary (no motion) and breathing
(sinusoidal motion, 1.8 cm respiratory amplitude, 3.3 s period) states. To align the phantom, the center of the
sphere inside the lung insert was positioned at the zero amplitude position and was aligned with the beam
isocenter. During all irradiations, film and RADPOS were placed inside the lung insert (as explained in section
II.A.4) to measure the point dose at the center of the tumour (film and RADPOS) as well as dose profile (film)
along the motion path of the tumour. For irradiations with breathing motion, the RADPOS positioning sensor was
used to record the motion at a sampling interval of 100 ms. In order to synchronize the beam-on time with the
starting phase of the Quasar motion, it was assured that computer clock times of both MOSAIQ and RADPOS
were synchronized.
II. B. Monte Carlo simulations
II. B.1. Monte Carlo user codes and their simulation parameters
All Monte Carlo simulations were performed using EGSnrc42
(V4-2.4.0, National Research Council of
Canada, Ottawa, ON, Canada). The BEAMnrc43
user code was used to model the Elekta Agility linear accelerator.
Dose distributions on stationary and moving phantoms were calculated with the DOSXYZnrc44
and
defDOSXYZnrc30
user codes, respectively. Calculated dose from Monte Carlo simulations was converted to
absolute dose using the following formulation:
(1)
where, is the monitor units (MU) delivered by a linear accelerator. In this formula
represents the dose scored per number of incident particles in a Monte Carlo simulation. The calibration
simulation was performed in water for a square field of 10 10 cm2 and SSD of 100 cm and dose was scored at a
depth of 10 cm.
9
DICOM plan files exported from the TPS as well as the linac delivery log files were used to generate all the
input files required for Monte Carlo simulations. An in-house Python script was written to perform these
functions. The photon cutoff energy (PCUT) and electron cutoff energy (ECUT) were set to 0.01 and 0.7 MeV,
respectively and the electron range rejection was set to 2 MeV for all Monte Carlo simulations. Dose calculations
were performed with 300,000,000 histories to achieve a mean relative statistical uncertainty45
of 0.4% over all
voxels with doses greater than 50% of the maximum dose.
II. B.2. Monte Carlo model of Elekta Agility linear accelerator
Figure 5 shows an illustration of the Elekta Agility linac model in BEAMnrc including the patient independent
(target, flattening filter, etc.) and patient dependent (160 multi-leaf collimators, lower jaws) components. To
model the multi-leaf collimators and lower jaws, the SYNCMLCE and SYNCMLCQ component modules (CMs)
were used, respectively. The ‘SYNC’ versions of these component modules enable synchronization of the motion
of the multi-leaf collimators, gantry and jaws in the linac model and dose calculation geometry by using a
common, randomly generated MU index which lies between 0 and 1, to sample the configuration of the linac
components for each particle history. This was required for the VMAT delivery simulations which used Source
2146
in DOSXYZnrc. A ‘SYNC’ version of the MLCQ component module was created by modifying this CM to
read the MU index generated in the SYNCMLCE CM. The SYNCMLCE module was also modified to account
for the defocusing of the leaf bank with respect to the focal spot to limit interleaf transmission.
FIG. 5. BEAMnrc preview of the Elekta Agility linac model showing the various component modules.
The linac model was constructed based on the technical data provided by the manufacturer and previously
published work47
. Details of the model parameters and validation have been presented elsewhere48
. For the
BEAMnrc simulations, bremsstrahlung cross-section enhancement was turned on while other transport parameters
were set to default values.
Primary collimator
Flattening filter
Target
Backscatter plate
Multi-leaf collimator
Lower jaws
Monitor Ion chamber
10
II. B.3. Dose calculation geometry
The voxelized geometry for dose calculations (egsphant file) was created from the 3DCT scans of the Quasar
phantom using an in-house software. Voxel densities were assigned using the HU/density calibration data for the
Big Bore CT scanner. For material assignment to the voxels, the approach introduced by Seco et al.49
was
followed to define voxels inside the phantom as water with densities derived from the CT image. With this
approach we can calculate Dw in MC simulations which allows direct comparison against film and RADPOS
measurements. All voxels outside the phantom were assigned as air with default density. The original CT image
was re-sampled to a resolution of 0.125 0.125 0.2 cm3 to generate the dose calculation geometry as shown in
Fig. 6.
FIG. 6. Dose calculation geometry (egsphant file) generated using static CT scans of the Quasar phantom.
II. B.4. Monte Carlo simulations of stationary phantom
Two sets of input files for DOSXYZnrc were created for simulations of the 4 4 cm2 and VMAT plan
deliveries on the stationary state of the phantom. The first set of simulations used input files generated from the
DICOM plan files as described in section II.B.1. Linac delivery log files were used to generate input files for
second set of simulations.
For simulation of the 4 4 cm2 square plan delivery, the linac model described in section II.B.2 was used as a
particle source (Source 9) thus eliminating the need to store a separate phase space file44
. Source 2146
was used to
simulate the VMAT beam delivery. Source 21 also uses a full BEAMnrc linac model as a particle source, but
additionally, as mentioned earlier in section II.B.2, it uses an MU index to synchronize the motion of dynamic
components of the linac in BEAMnrc. This MU index is then used to interpolate between the control points
specifying the gantry, collimator and couch angle in the treatment plan or delivery log file (extracted as explained
in section II.B.1) as a function of fraction of the delivered monitor units to determine the source settings in
DOSXYZnrc. A photon splitting value of 10 and default values for other transport parameters were used for all
DOSXYZnrc simulations.
11
II. B.5. 4D Monte Carlo simulations of moving phantom
Simulations of plan delivery on the breathing motion state of the Quasar phantom were performed using a
modified version of the defDOSXYZrc user code30
. Organ motion is modeled in defDOSXYZnrc by displacing
the voxel nodes of the reference dose calculation geometry using displacement vectors. These displacement
vectors may be generated from a motion model or deformable image registration. New densities are calculated for
each deformed voxel to ensure conservation of mass between the reference and deformed geometries. Incident
particles are transported and their energy depositions are scored in the deformed geometry. No mapping of doses
between geometries is required since the same dose calculation grid is retained for all motion phases simulated.
The output of a defDOSXYZnrc simulation is the dose distribution in the reference geometry coordinates, which
was the stationary state of the Quasar phantom in this work.
The modified version of the defDOSXYZnrc, which we refer to as 4DdefDOSXZYnrc, enables simulation of
a continuously deforming geometry by sampling a new geometry for each incident particle. The MU index
defined in source 21 is used to determine the corresponding motion phase from a user-specified respiratory motion
trace. The voxel node displacements are linearly interpolated from a set of displacement vectors which the user
provides as input to the simulation.
The input files generated for the 4DdefDOSXYZnrc simulations used the beam delivery information from
linac delivery log files as well as the Quasar respiratory motion recorded by RADPOS. The RADPOS motion was
re-sampled to match the resolution of the log file and then normalized to give a displacement vector scaling factor
that ranged between -1 and 1 as a function of cumulative MU (Figure 7). For each incident particle, the MU index
used in source 21 is used to interpolate a vector scaling factor to be applied to the displacement vectors. Since the
Quasar lung insert moves rigidly, displacement vectors that exactly modeled this rigid translation (peak-to-peak
amplitude and direction) were generated with an in-house Python script using the recorded motion trace data.
FIG. 7 Normalized displacement (scaling factor for displacement vectors) recorded by RADPOS and re-sampled according to the
resolution of log files.
12
II.B.6. Monte Carlo simulation comparisons with measurements
2D dose distributions from film and MC simulations were extracted and compared using a 2D gamma
analysis50
with a 2% dose-difference and 2 mm distance-to-agreement criteria. The choice of these criteria was to
comply with the AAPM commissioning and QA guidelines51 for gamma comparisons of VMAT plan deliveries.
The film dose was used as the reference dose distribution for the gamma analysis. A dose calculation grid with 2
mm voxel size along the motion path of tumor was used for MC simulations. The dose grid resolution of film was
reduced (by applying a moving average filter) to match the resolution of the MC dose grid. Dose grids from film
and MC were aligned for the best agreement between the dose distributions. While this approach eliminates the
influence of setup errors, the alignment corrections were observed to never exceed 0.7 mm. Two different dose
thresholds, 5% and 50% of the evaluated maximum dose were tested to evaluate the agreement for all voxels or
only voxels in high dose regions, respectively
Dose profiles along the superior/inferior direction (motion path of the tumour) were extracted from film
measurements and were compared with profiles from MC simulations using an in-house Python script. A 1D
gamma analysis using the same parameters as the 2D analysis was performed on the dose profiles. Comparisons
of dose values at the center of the tumour were performed between measurements (film and RADPOS) and MC
simulations. RADPOS measurements were converted to absolute dose using the calibration coefficient acquired as
explained in section II.A.4.
13
III. RESULTS
III. A. Stationary phantom comparisons
Figure 8(a, b) shows the comparison of dose profiles from MC simulations using log files (MC-Log file) as
well as film for the 4 4 cm2 and VMAT plan deliveries on the stationary state of the Quasar phantom.
FIG. 8. Dose profile comparisons for 4×4 cm2 (Left) and VMAT (Right) plan deliveries on the stationary phantom along the
superior/inferior direction (Top row), 2D dose distribution (Middle row) and gamma comparison maps (Bottom row) from film
measurements and MC simulations (log file). Gamma comparison maps were generated using a dose threshold = 50% Dmax.
Good agreement was observed between the measured and simulated dose profiles with more than 99% of the
voxels passing the 1D gamma of 2%/2 mm criteria, for both plan deliveries when 5% Dmax threshold was applied.
For the 50% Dmax threshold, the agreements were observed to be higher than 99% for the VMAT delivery and
(b) (a)
(f) (e)
(d) (c)
2%/2 mm 2%/2 mm
14
98% for the static plan delivery. Also, a passing rate of 99.1% and 100% (i.e. all dose points) was observed
comparing MC simulations using log files (MC-Log files) against film measurements for the static and VMAT
plan deliveries, respectively. MC simulations using log files (MC-Log files) against film measurements were
found to have 100% passing rate for the 5% Dmax threshold. In Fig. 8(b), dose points in profiles from film and MC
simulations were observed to have a distance-to-agreement of less than 1 mm.
Table I shows the RADPOS and film dose measurements taken at the center of the tumour as well as the
corresponding dose values from MC simulations. Uncertainties reported on dose values are experimental (film
and RADPOS) or statistical (MC simulations). Percent difference comparisons for dose values were calculated as
well.
TABLE I. Dose values at the center of the tumour from RADPOS, film and MC simulations for static and VMAT beam deliveries on the
stationary phantom.
Dose (cGy)
Plan MC(DICOM) MC(log file) Measurements
Film RADPOS
Static 4 4 cm2 99.50.4% 99.30.4% 100.52.3% 101.61.4%
VMAT 99.50.4% 99.80.4% 99.02.3% 98.41.4%
All dose values from MC simulations were found to be within 1% of film and 2% of RADPOS for VMAT
plan delivery. The same was true for the static plan delivery with the exception of MC-Log file against film
measurements having a discrepancy of 1.2%. Film and RADPOS doses agreed with each other at 1.1% for the
static and 1% for the VMAT plan beam deliveries.
2D dose distributions from film and MC-Log file simulations for the 4 4 cm2 and VMAT plan deliveries, as
well as the 2D gamma comparison maps for a 2%/2 mm criteria are shown in Fig. 8(c, d) and Fig. 8(e, f),
respectively. Gamma comparison maps were generated for a dose threshold of 50% of the reference maximum
dose. The passing rates of the 2D gamma comparisons for both deliveries are shown in Table II.
15
TABLE II. Passing rate of 2D gamma comparison of 2%/2 mm criteria for MC simulations (DICOM and log files) against film
measurements for static and VMAT deliveries on the stationary phantom.
As the values from Table II for VMAT delivery suggest, more than 99% of the dose points passed the 2D
gamma of 2%.2 mm for MC simulations against film measurements when a 50% dose threshold was used. At 5%
dose threshold, passing rate was reduced by less than 2% compared to a 50% threshold. With regards to the static
plan delivery, a passing rate of almost 99% was observed for both dose thresholds.
III. B. Moving phantom comparisons
For the moving state of the Quasar phantom, MC simulations used only delivery information from the log
files. Comparison of dose profiles from MC simulations using log files and film for both 4 4 cm2 and VMAT
plan deliveries on the moving state of the Quasar phantom is shown in Figure 9(a, b).
Plan Evaluated dose profile Reference dose profile
Threshold = 5%
Dmax
Threshold = 50%
Dmax
Film Film
Static 4 4 cm2
MC (DICOM) 99.5% 98.9%
MC (log file) 99.5% 98.9%
VMAT
MC (DICOM) 97.9% 99.5%
MC (log file) 97.9% 99.5%
16
FIG. 9. Dose profile comparisons for 4×4 cm2 (Left) and VMAT (Right) plan deliveries on the moving phantom along the
superior/inferior direction (Top row), 2D dose distribution (Middle row) and gamma comparison maps (Bottom row) from film
measurements and MC simulations (log file). Gamma comparison maps were generated using a dose threshold = 50% Dmax.
The level of agreement between dose profiles was verified using a 1D gamma criterion of 2%/2 mm and all
points from MC simulations agreed with the film for the static plan delivery. For VMAT delivery, the level of
agreement between MC simulation and film measurement was observed to be higher than 91% according to the
1D gamma comparisons for both dose thresholds. Discrepancies between measured and simulated profiles at the
vicinity of the maximum dose region for the VMAT delivery were found to be within the experimental
uncertainties (2.3%) and are attributed to film uncertainties.
(b) (a)
(d) (c)
(f) (e)
2%/2 mm 2%/2 mm
17
Dose values from MC simulations, film and RADPOS taken at the center of the tumor are shown in Table III.
Based on percent difference comparisons, MC simulations agreed within 1% with film and within 2 of
experimental uncertainties (2.8%) with RADPOS measurements. Film and RADPOS dose values were found to
be within 2% of each other.
TABLE III. Dose values at the center of the tumour from RADPOS, film and MC simulations for static and VMAT deliveries on the
moving phantom.
Dose (cGy)
Plan MC(log file) Measurements
Film RADPOS
Static 4 4 cm2 96.10.4% 96.42.3% 98.41.4%
VMAT 91.90.4% 91.12.3% 89.61.4%
Results of the 2D gamma comparisons with a 2%/2 mm criteria between the 2D dose distributions illustrated
in Figure 9(c, d) and Figure 9(e, f) are presented in Table IV. Gamma comparison maps were generated for a dose
threshold of 50% of the reference maximum dose.
TABLE IV. Passing rate of 2D gamma comparison of 2%/2 mm criteria for MC simulations (log files) against film measurements for static
and VMAT deliveries on the moving phantom.
2D dose distributions from MC simulations and film measurements had a good agreement and the passing rate
of gamma comparison was found to be higher that 94% for the VMAT delivery. The passing rate was found to be
slightly better when only dose points in the high dose region (i.e. 50% Dmax) were compared. All dose points (i.e.
100% passing rate) passed the gamma comparison when MC simulations were compared against films
measurements for the static plan delivery.
Plan Evaluated dose profile Reference dose profile
Threshold = 5%
Dmax
Threshold = 50%
Dmax
Film Film
Static 4 4 cm2
MC (log file) 100% 100%
VMAT
MC (log file) 94.6% 97.4%
18
IV. DISCUSSION
The aim of this study was to validate 4D Monte Carlo simulations of the dose delivered to a respiratory
motion lung phantom using the 4DdefDOSXYZnrc user code. A static 4 4 cm2 and a VMAT plan were delivered
to the phantom in stationary and breathing states. Our results show that the overall agreement of point dose values
at the center of the target volume (i.e. tumour) from MC simulations and film measurements is within 1.2%. Dose
differences between RADPOS and MC simulations are not larger than 2 of the experimental uncertainty, which
corresponds to 2.8%. These results apply to both plan deliveries and phantom motion states, i.e., stationary and
moving. Dose profiles and 2D dose distributions from film and MC simulations using delivery log files (MC-Log
file) were found to agree with more than 95% of the dose points, and hence passed the gamma comparison of
2%/2 mm placing the results in an acceptable clinical range. The level of agreement was observed to be higher for
the 4 4 cm2 square plan deliveries compared to VMAT; this can be related to the lower level of the treatment plan
complexity.
The validation work presented in this paper was performed in a rigidly moving phantom. We used manually
generated displacement vectors to accurately model the lung insert motion. The use of DIR algorithms was not an
appropriate choice considering that the motion was a purely rigid translation without any deformations occurring
in the lung insert. Work is currently underway to extend the validation work to a deformable phantom in order to
evaluate the accuracy of our 4DMC method to calculate dose delivered to a deforming anatomy. The use of DIR
to generate deformation vectors will be necessary to model deformations occurring in the phantom. This means
that the accuracy of simulation will depend on the accuracy of DIR algorithm as well. Thus, the challenge here
will be to generate DVFs that are as accurate as possible. Multiple sets of DVFs from registering different
respiratory phases (e.g. end-of-inhale or end-of-exhale) to the reference phase might be required to achieve the
most accurate model for the phantom motion and deformation.
A potential source of experimental uncertainty in this work is the temporal resolution of RADPOS and
delivery log files. The maximum sampling rate of the RADPOS position tracker is 100 ms and the sampling rate
for log files is 40 ms. Such sampling intervals act as system latencies of approximately 108 ms (summed in
quadrature) and limit our ability to accurately determine the starting phase of the respiratory cycle. Based on the
phantom motion used in these studies, this system delay was estimated to result in positional inaccuracies of about
1.1 mm. In addition to that, the uncertainty of the displacement measurements by RADPOS should be taken into
account. This uncertainty was measured to be about 0.2 mm and is in agreement with values reported by Cherpak
et al.37
. Another source of uncertainty is the accuracy of the PC clock synchronization between MOSAIQ and
RADPOS PCs. This synchronization is good to within ms and can also introduce positional uncertainties of less
than 1 mm when trying to detect the motion starting phase. Investigations of the impact of such positional
uncertainties showed that they lead to point dose uncertainties of 1%-2% at the center of the tumour. These
investigations indicate the sensitivity of our 4D dose calculations to interplay effect. For shorter delivery times
19
where only a few respiratory cycles are present or VMAT plans which are more highly modulated, accurate
detection of the starting phase of the respiratory cycle will become more important to assure better agreement
between 4DMC simulations and measurements.
A further source of uncertainty is the interpolation between the control points as performed by source 21 in
MC simulations. However, investigations showed that the effect of such interpolation is negligible. The high
sampling rate of the log files means that control points in MC input files are defined at every 40 ms and linear
interpolation occurs between these control points. Based on analyzing our log files, this corresponds to less than
0.2o of change in the gantry angle and maximum of 0.1 mm change in the MLC leaf positions between two
successive control points. This corresponds to the MU index (fraction) change of less than 0.001 between two
consecutive interpolation or control points, which minimizes the potential impact of interpolation on the dose
calculations.
The total uncertainty of the film dose measurements could have been reduced to about 1.6% if two films were
used for each individual Quasar irradiation. Using triple-channel rather than single channel film dosimetry is also
a potential film processing technique that can reduce the noise level of the film response and as a result the
dosimetric uncertainty52
.
V. CONCLUSIONS
The accuracy of a 4D Monte Carlo-based simulation tool to deliver dose to a Quasar respiratory motion
phantom using an Elekta Agility linac was investigated and verified. A sinusoidal motion with an amplitude of 1.8
cm and period of 3.3 sec was tested and RADPOS system was used to record this motion in real-time. Linac
delivery log files were used, along with the respiratory motion trace, to reproduce measurements during static 4 4
cm2 square plan and VMAT plan deliveries. Dose values at the center of the target volume inside the phantom
were observed to be within 1.2% compared to film measurements and not higher than 2 of the experimental
uncertainties (2.8%) with respect to RADPOS. Less than 5% of dose points from MC simulations using log files
failed a 2%/2 mm gamma comparison when compared against film measurements. This work has shown that our
4DMC simulations using the defDOSXYZnrc user code in combination with RADPOS (to measure motion) and
delivery log files (to reproduce beam delivery) accurately calculate realistic dose distributions in a moving
anatomy. Future work will focus on establishing the accuracy of our method in a deformable phantom as well as
investigating irregular respiratory motion patterns. If extended to deformable anatomies and more realistic
respiratory motions, this 4DMC simulation tool can be used for adaptive purposes by accurately calculating the
cumulative dose delivered to patients during treatments.
ACKNOWLEDGEMENTS
We would like to thank Jason Smale of Elekta for helping us with delivery log files. This project has been
supported by a grant from the Ontario Consortium for Adaptive Radiotherapy (OCAIRO).
20
DISCLOSURE OF CONFLICTS OF INTEREST
The authors have no relevant conflicts of interest to disclose.
21
References:
1 T. Bortfeld, S. B. Jiang and E. Rietzel, “Effects of motion on the total dose distribution”, Semin. Radiat. Oncol. 14, 41–
50 (2004). 2 P. J. Keall et al., “The management of respiratory motion in radiation oncology report of AAPM Task Group 76,” Med.
Phys. 33, 3874–900 (2006).
3 K. M. Langen and D. T. L. Jones, ‘‘Organ motion and its management,’’Int. J. Radiat. Oncol. Biol. Phys. 50, 265–278
(2001).
4 S. Flampouri, S. B. Jiang, G. C. Sharp, J. Wolfgang, A. A. Patel and N. C. Choi, “Estimation of the delivered patient
dose in lung IMRT treatment based on deformable registration of 4D-CT data and Monte Carlo simulations”, Phys. Med.
Biol. 51, 2763–2779 (2006).
5 M. Engelsman, E. M.F. Damen, K. De Jaeger, K. M. Van Ingen, B. J. Mijnheer, “The effect of breathing and set-up
errors on the cumulative dose to a lung tumor”, Radiother. Oncol. 60, 95-105 (2001).
6 T. Bortfeld, K. Jokivarsi, M. Goitein, J. Kung and S. B. Jiang, “Effects of intra-fraction motion on IMRT dose delivery:
statistical analysis and simulation”, Phys. Med. Biol. 47, 2202-2220 (2002).
7 C. Chui, E. Yorke, L. Hong, “The effects of intra-fraction organ motion on the delivery of intensity-modulated field
with a multileaf collimator”, Med. Phys. 30, 1736-1746 (2003).
8 R. I. Berbeco, C. J. Pope and S. B. Jiang, “Measurement of interplay effect in lung IMRT treatment using EDR2 films”,
Med. Phys. 7, 33-42 (2006).
9 S. B. Jiang, C. Pope, K. M. Al Jarrah, J. H. Kung, T. Bortfeld and G. T. Y Chen, “An experimental investigation on
intra-fractional organ motion effects in lung IMRT treatments”, Phys. Med. Biol. 48, 1773-1784 (2003).
10 J. Duan, S. Shen, J. B. Fiveash, R. A. Popple and I. A. Brezovich, “Dosimetric and radiobiological impact of dose
fractionation on respiratory motion induced IMRT delivery errors: A volumetric dose measurement study”, Med. Phys.
33, 1380-1387 (2006).
11 L. Court, M. Waqar, R. Berbeco, A. Reisner, B. Winey, D. Schofield, D. Ionascu, A. M. Allen, R. Popple and T.
Lingos, “Evaluation of the interplay effect when using RapidArc to treat targets moving in the craniocaudal or right-left
direction”, Med. Phys. 37, 4-11 (2010).
12 C. Ong, W. F. A. R. Verbakel, J. P. Cuijpers, B. J. Slotman and S. Senan, “Dosimetric impact of interplay effect on
RapidArc lung stereotactic treatment delivery”, ’’Int. J. Radiat. Oncol. Biol. Phys. 79, 305–311 (2011).
13 M. Rao, J. Wu, D. Cao, T. Wong, V. Mehta, D. Shepard and J. Ye, “Dosimetric impact of breathing motion in lung
stereotactic body radiotherapy treatment using image-modulated radiotherapy and volumetric modulated arc therapy”,
’’Int. J. Radiat. Oncol. Biol. Phys. 83, 251–256 (2012). 14
A. E. Lujan, E. W. Larsen, J. M. Balter and R. K. Ten Haken, “A method for incorporating organ motion due to
breathing into 3D dose calculations”, Med. Phys. 26, 715-720 (1999).
15 S. D. McCarter and W. A. Beckham, “Evaluation of the validity of a convolution method for incorporating tumour
movement and set-up variations into the radiotherapy treatment planning system”, Phys. Med. Biol.45, 923-931 (2000).
16 V. Panettieri, B. Wennberg, G. Gagliardi, M. A. Duch, M. Ginjaume and I .Lax, “SBRT of lung tumors: Monte Carlo
simulations with PENELOPE of dose distributions including respiratory motion and comparison with different treatment
planning systems”, Phys. Med. Biol.52, 4265-4281 (2007).
22
17 I. J. Chetty, M. Rosu, D. L. McShan, B. A. Fraass, J. M. Balter and R. K. Ten Haken, “Accounting for center-of-mass
target motion using convolution methods in Monte Carlo-based dose calculation of lung”, Med. Phys. 31, 925-932
(2004).
18 T. Craig, J. Battista and J. V. Dyk, “Limitations of the convolution method for modeling geometric uncertainties in
radiation therapy. I. The effect of shift invariance”, Med. Phys. 30, 2001-2011 (2003).
19 W. A. Beckham, P. J. Keall and J. V. Siebers, “A fluence-convolutoin method to calculate radiation therapy dose
distributions that incorporate random set-up error”, Phys. Med. Biol.47, 3465-3473 (2002).
20 I. J. Chetty, M. Rosu, N. Tyagi, L. H. Marsh, D. L. McShan, J. M. Balter, B. A. Fraass and R. K. Ten Haken, “A
fluence convolution method to account for respiratory motion in three-dimensional dose calculations of the liver: A
Monte Carlo study”, Med. Phys. 30, 1776-1780 (2003).
21 K. K. Brock, D. L. McShan and R. K. Ten Haken, “Inclusion of organ motion in dose calculations”, Med. Phys. 30,
290-295 (2003).
22 P. J. Keall, J. V. Siebers, S. Joshi and R. Mohan, “Monte Carlo as a four dimensional radiotherapy treatment planning
tool to account for respiraotry motion”, Phys. Med. Biol.49, 3639-3648 (2004).
23 M. Rosu, I. J Chetty, J. M. Balter, M. L. Kessler, D. L. McShan and R. K. Ten Haken, “Dose reconstruction in
deforming lung anatomy: Dose grid size effects and clinical implications”, Med. Phys. 32, 2487-2495 (2005).
24 M. Ding, J. Li, J. Deng, E. Fourkel and C. M. Ma, “Dose correlation for thoracic motion in radiation therapy of breast
cancer”, Med. Phys. 30, 2520-29529 (2003).
25 S. H. Jung, S. M. Yoon, S. H. Park, B. Cho, J. W. Park, J. Jung, J. H. Park, J. H. Kim and S. D. Ahn, “Four-
dimensional dose evaluation using deformable image registration in radiotherapy for liver cancer”, Med. Phys. 40,
01176-1-7 (2013).
26 B. Schaly, J. A. Kempe, G. S. Bauman, J. J. Battista and J. Van Dyk, “Tracking the dose distribution in radiation
therapy by accounting for variable anatomy”, Phys. Med. Biol.49, 792-805 (2004).
27 H. Paganetti, H. Jiang, J. A. Adams, G. T. Chen and E. Rietzel, “Monte Carlo simulations with time-dependent
geometries to investigate effects of organ motion with high temporal resolution”, ’’Int. J. Radiat. Oncol. Biol. Phys. 60,
942–950 (2004).
28 J. V. Siebers and H. Zhong, “An energy transfer method for 4D Monte Carlo dose calculation”, Med. Phys. 35, 4096-
4105 (2008).
29 H. Zhong and J. V. Siebers, “Monte Carlo dose mapping on deforming anatomy”, Phys. Med. Biol.54, 5815-5830
(2009).
30 E. Heath and J. Seuntjens, “A direct voxel tracking method for four-dimensional Monte Carlo dose calculations in
deforming anatomy”, Med. Phys. 33, 434-445 (2006).
31 H. S. Li, H. Zhong, J. Kim, C. Glide-Hurst, M. Gulan, T. S. Nurushev and I. J. Chetty, “Direct dose mapping versus
energy/mass transfer mapping for 4D dose accumulation: fundamental differences and dosimetric consequences”, Phys.
Med. Biol.59, 173-188 (2014).
32 J. Belec and B. G. Clark, “Monte Carlo calculation of VMAT and helical tomotherapy dose distributions for lung
stereotactic treatments with intra-fraction motion”, Phys. Med. Biol.58, 2807-2821 (2013).
23
33 J. Belec, N. Ploquin, D. La Russa and B. G. Clark, “Position-probability-sampled Monte Carlo calculation of VMAT,
3DCRT, step-and-shoot IMRT, and helical tomotherapy dose distributions using BEAMnrc/DOSXYZnrc”, Med. Phys.
38, 948-960 (2011).
34 E. Heath, I. Badragan and I. Popescu, “4D Monte Carlo simulations of beam and patient motion using
EGSnrc/BEAMnrc”, Med. Phys. 39, 3480 (2012) (Abstract).
35 M. Fippel, “Fast Monte Carlo dose calculation for photon beams based on the VMC electron algorithm”, Med. Phys.
26, 1466-1475 (1999)
36 A. Cherpak, M. Serban, J. Seuntjens and J. E. Cygler, “4D dose-position verification in radiation therapy using the
RADPOS system in a deformable lung phantom”, Med.Phys. 38, 179-187 (2011).
37 A. Cherpak, W. Ding, A. Hallil and J. E. Cygler, “Evaluation of a novel 4D dosimetry system”, Med.Phys. 36, 1672-
1679 (2009).
38 S. Devic, J. Seuntjens, E. Sham, E. B. Podgorsak, C. R. Schmidtlein, A. S. Kirov and C. G. Soares, “Precise
radiochromic film dosimetry using a flat-bed document scanner”, Med. Phys. 32, 2245-2253 (2005).
39 M. McEwen, L. DeWard, G. Ibbott, D. Followill, D. W. O. Rogers, S. Seltzer and J. Seuntjens, “Addendum to the
AAPM’s TG-51 protocol for clinical reference dosimetry of high energy photon beams”, Med. Phys. 41, 041501-1-20
(2014).
40J. Seuntjens, M. Olivaris, M. Evans and E. B. Podgorsak, “Absorbed dose to water reference dosimetry using solid
water phantoms in the context of absorbed-dose protocols”, ”, Med. Phys. 32, 2945-2953 (2005).
41L. J. Van Battum, D. Hoffmans, H. Piersma and S. Heukelom, “Accurate dosimetry with Gafchromic EBT film of a 6
MV photon beam in water: What level is achievable?”, Med. Phys. 35, 704-716 (2008).
42 E. Mainegra-Hing, D. W. O. Rogers, F. Tessier and B. R. B. Walter, “The EGSnrc code system: Monte Carlo
simulation of Electron and Photon Transport”, NRCC Report PIRS-701, NRC Canada (2013).
43 D.W. O. Rogers, B. A. Faddegon, G. X. Ding, C. M. Ma, J. We and T. R. Mackie, “BEAM: A Monte Carlo code to
simulate radiotherapy treatment units”. Med. Phys. 2, 503-524 (1995).
44 B. Walters, I. Kawarkow and D. W. O. Rogers, “DOSXYZnrc user manual”, NRCC Report PIRT-794revB, NRC
Canada (2009).
45 I. J. Chetty, M. Rosu, M. L. Kessler, B. A. Fraass, R. K. Ten Haken, F. M. Kong and D. L. McShan, “Reporting and
analyzing statistical uncertainties in Monte Carlo-based treatment planning”, ’’Int. J. Radiat. Oncol. Biol. Phys. 66,
1249–1259 (2006).
46 J. Lobo and I. A. Popescu, “Two new DOSXYZnrc sources for 4D Monte Carlo simulations of continuously variable
beam configurations, with applications to RapidArc, VMAT, TomoTherapy and CyberKnife”, Phys. Med. Biol.55, 4431-
4443 (2010).
47 S. S. Almberg, J. Frengen, A. Kylling and T. Lindmo, “Monte Carlo linear accelerator simulation of megavoltage
photon beams: Independent determination of initial parameters”, Med. Phys. 39, 40-47 (2012).
48 M. Vujicic, J. Belec, E. Heath, S. Gholampourkashi and J. Cygler, “Precision modeling of the leaf-bank rotation in
Elekta’s Agility MLC: Is it necessary?”, Med. Phys. 42, 3480 (2015) (Abstract).
49 J. Seco and P. M. Evans, “Assessing the effect of electron density in photon dose calculations”, Med. Phys. 33, 540-
552 (2006).
24
50 D. A. Low, W. B. Harms, S. Mutic and J. A. Purdy, “A technique for quantitative evaluation of dose distribution”,
Med. Phys. 25, 656-661 (1998).
51 J. B. Smilowitz et al., “AAPM medical physics practice guidelines 5.a: Commissioning and QA treatment planning
dose calculations – Megavoltage photon and electron beams”, J. Appl. Med. Phys., 16, 14-34 (2015).
52 D. F. Lewis, “A guide to radiochromic film dosimetry with EBT2 and EBT3”, Ashland Specialty Ingredients (2012)
(White paper).