Professional Documents
Culture Documents
Description and Dosimetric Verification of The PEREGRINE Monte Carlo Dose Calculation System For Photon Beams Incident On A Water Phantom
Description and Dosimetric Verification of The PEREGRINE Monte Carlo Dose Calculation System For Photon Beams Incident On A Water Phantom
Description and Dosimetric Verification of The PEREGRINE Monte Carlo Dose Calculation System For Photon Beams Incident On A Water Phantom
Abstract
PEREGRINE is a 3D Monte Carlo dose calculation system written specifically for
radiotherapy. This paper describes the implementation and overall dosimetric
accuracy of PEREGRINE physics algorithms, beam model, and beam
commissioning procedure. Particle-interaction data, tracking geometries, scoring,
variance reduction, and statistical analysis are described. The program BEAM is
used to model the treatment-independent accelerator head resulting in the
identification of primary and scattered photon sauces and an electron
contaminant source. The magnitude of the electron source is increased to improve
agreement with measurements in the buildup region in the largest fields.
Published measurements provide an estimate of backscatter on monitor chamber
response. Commissioning consists of selecting the electron beam energy,
determining the scale factor that defines dose per monitor unit, and describing
treatment-dependent beam modifiers. We compare calculations with
measurements in a water phantom for open fields, wedges, blocks, and a multi-leaf
collimator for 6 and 18 MV Varian Clinac 2100C photon beams. All calculations are
reported as dose per monitor unit: no additional, field-specific normalization is
included in comparisons with measurements. Maximum discrepancies were less
than either 2% of the maximum dose or 1 mm in isodose position for all field sizes
and beam modifiers.
calculation in radiotherapy,8 9 10 11 12 13
I. Introduction and rapidly-increasing computer
speeds it is now possible to use 3D
Dose calculation accuracy is a
Monte Carlo methods for day-to-day
critical part of radiation therapy.
treatment planning.
Combining first-principles physics
Accurate dose calculation
with physical descriptions of the
requires accurate characterization of
radiation source and patient, Monte
the radiation source. This can be
Carlo transport methods have the
accomplished with high accuracy and
potential to calculate dose accurately
exceptional detail by simulating the
over a wide variety of treatment
transport of particles through the
delivery and patient conditions.1 2 3 4 5
6 7 accelerator head and beam delivery
Owing to development of faster
system.14 15 16 17 18 19 20 Recent work
codes designed specifically for dose
2
transport mesh using the delta electron tracking cut off. All energies
scattering method.35 This allows stated are kinetic energy.
PEREGRINE to avoid computationally In the patient mesh, termination
expensive distance-to-boundary of the electron trajectory is
calculations without biasing the determined by its kinetic energy and
resulting dose calculations. In this its location in the geometry. The
study, the minimum photon tracking track terminates when the electron’s
energy was 100 eV. energy falls below the energy needed
to traverse one-third of the voxel’s
Charged particles minimum dimension (approximately
PEREGRINE uses class-II 180 keV kinetic energy for 1
condensed history method for millimeter voxels in water). Electrons
charged particle transport,36 are never transported below 10 keV
modeling knock-on and kinetic energy. Once the particle
bremsstrahlung processes above reaches the minimum tracking
specified cut off energies as discrete energy, its residual energy is
events. The Moliere37 multiple deposited at a random location along
scattering method is employed, a straight-line trajectory of length
implemented as in the EGS4 code. equivalent to its residual range. The
The condensed history electron step termination of a positron trajectory
size taken is determined by the results in the emission of two 511 keV
minimum of the step size necessary kinetic energy annihilation photons.
to create a bremsstrahlung photon, to
generate a knock-on electron, to 3. Geometry
reach the next spatial boundary, to Beam modifiers are described as
reach the next energy bin boundary, collections of six-sided prisms, with
or 1 mm. The energy bins used in boundaries defined by the physical
PEREGRINE are such that the dimensions of the modifier. Collimator
fractional energy loss in crossing a jaws can move along a line or arc,
bin varies between about 8% and matching their physical
20%. The electron step is typically implementation in the accelerator
broken into two substeps by treatment head. Aperture blocks are
deflecting the particle by the multiple modeled as a diverging raster of voxel
scattering angle for the full step at a prisms with lateral dimensions
random point along the step. In the defined by the user. In multileaf
patient mesh, energy is deposited at collimators, the convex leaf ends
a random location along each (Varian design) and tongue-and-groove
substep. features are modeled explicitly.
In the beam delivery system, Wedges are defined as a series of
electrons can be tracked with contiguous trapezoidal prisms. All
different levels of approximation, block and wedge trays are modeled
depending on the beam modifier. as uniform slabs of material, with
While user-defined tracking and cutoff lateral extents corresponding to their
options can be modified, we used the physical dimensions. All space not
following options for this work: 0.01 occupied by solid, beam-modifying
MeV bremsstrahlung creation material is filled with air, in which
threshold, 0.1 MeV Møller electron particles are allowed to interact. The
creation threshold, and 0.1 MeV user assigns all beam-modifier
6
materials and densities in the device volume than the cube defined by the
description input file. grid points. This causes the
The patient mesh is taken directly calculation to reach a given
from the CT scan, with no reduction statistical noise level faster than it
in resolution. The user assigns would if the dose were collected in
material to each voxel by specifying the cubes. Overlapping spheres
pre-defined materials for ranges of CT causes neighboring points to be
number. The user specifies density correlated, and results in the
from the CT number with a resolution being slightly lower than
monotonically-increasing, piecewise- the grid spacing.
continuous linear function. Each
material can also be defined with a 5. Variance reduction
default density. Several variance reduction
techniques are used in PEREGRINE.
4. Scoring They are: source particle reuse, range
Dose is scored in PEREGRINE on a rejection, and splitting/Russian
grid that is independent from the CT rouletting in the patient.
derived Cartesian grid that is used for
particle transport. This permits Source particle reuse
PEREGRINE to speed up the Each particle that survives
calculation by using variance transport through the modifiers is
reduction techniques in the regions reused a set number of times. If the
that lie outside of the dose-scoring source photon is reused so many
grid. times that the ranges of the electrons
In PEREGRINE, dose is scored in a set created by different reuses overlap, it
of spheres that are centered on the will make non-independent
points of a grid with cubical unit depositions to the same dosel. This
cells. The spheres, referred to as amounts to expending computer time
dosels, fill all of space since they just without a reduction in statistical
touch along the cube diagonals. The noise. Photons are reused upon
spheres overlap elsewhere. The entering the CT grid. Electrons are
transport algorithms in PEREGRINE reused upon entering the air column
deposit energy at a random point below the last modifier. This permits
along the electron step, as opposed the reused electrons to spread out
to depositing energy inside a given CT before entering the CT grid, which
voxel. When energy is deposited at a prevents them from making non
point in space, the energy is added to independent depositions into the
every sphere that contains that point. same dosel. The number of times
To obtain dose, the energy is divided that source particles can be reused is
by the mass of the sphere, which is also limited by the requirement that
determined in a setup calculation that the source be adequately sampled
utilizes a closed form solution for the during the course of the run.
common volume of a sphere and a
rectangular parallelepiped. Spheres Range rejection
were selected as a way of An electron with CDSA range that
approximating the dose reported at a is less than 1/3 the smallest
point for a region of non-uniform dimension of a CT voxel is terminated
density. The spheres have a larger and its energy is dumped at a random
7
point along a straight line of length For dosels with similar density,
equal to its residual range. In a setup the variance tends to be proportional
calculation, each voxel is assigned a to the dose. If the variance were
range rejection energy using this exactly proportional to the dose for
criterion. Electrons are never every dosel in the problem, then the
transported below 10 keV, regardless dosel with the maximum dose would
of the range rejection energy. also be the dosel with the maximum
standard deviation. If this were true,
Russian rouletting and splitting the termination criterion based on the
Photons that are outside the standard deviation of the watch dosel
dose scoring region and are moving would be equivalent to the
away from the dose scoring region termination criterion based on the
are Russian rouletted. If descendents standard deviation of every dosel.
of rouletted photons move towards or The standard deviation that
enter the dose scoring region, they goes with a given dose tends to
are split. The number of times that decrease as the density of the dosel
roulletting and splitting can happen to increases. For this reason, the watch
the descendents of a photon is dosel is selected from dosels that
limited to avoid the generation of very have a mass that is between one
high and very low weight particles. fourth and 3 times the mass of a
water dosel.
6. Statistical analysis
The Monte Carlo calculation is B. Source Description
considered to have converged to a The non patient specific parts of
fractional error, F, when the standard the Varian high energy family of
deviation of the dose of every dosel is accelerators (Clinacs 2100C, 2100C/D,
less than F * Md where Md is the 2300C/D) are simulated for 6 and 18
largest dose in any dosel. Calculating MV using the BEAM Monte Carlo
the standard deviation in every dosel code.Error: Reference source not
results in a significant expenditure of found The physical dimensions and
memory and time due to the large materials of the accelerator were
size of the dosel array. For this obtained from the manufacturer.*
reason, we provide a statistical figure
of merit based only on the standard 1. BEAM Simulations
deviation calculated for a single The BEAM simulation extended
dosel, which we call the watch dosel. from the top of the bremsstrahlung
The watch dosel is selected target to the bottom of the monitor
during the first part of the run as the chamber. The electron beam incident
dosel with the maximum dose after a on the target was assumed to have no
fixed number of histories, H. H is divergence, to be mono-energetic, and
chosen to be large enough that the to have a uniform spatial distribution
watch dosel will, at the end of the with 1 mm radius. The delta ray and
run, have a dose that is close to Md. bremsstrahlung production cutoffs
The run is terminated when the were taken to be AE = 0.521 MeV
standard deviation of the watch dosel (kinetic + rest mass) and AP = 0.01
is less than F * Wd, where Wd is the
dose in the watch dosel. *
Varian Oncology Systems, Palo Alto,
CA
8
5. The intersection of the particle data the best. Backscatter from the
trajectory in the virtual source upper jaws, f1, is characterized by a
plane is sampled by treating the second-order polynomial fit to
virtual source plane energy measurements made by varying the
fluence as a probability position of the upper jaw while
distribution. The trajectory of the keeping the lower jaw fixed at 40 cm.
particle is determined by Backscatter from the lower jaws is
connecting the points in the characterized by another second-
virtual source plane and the order polynomial fit, f2, to lower-jaw
isocenter plane. measurements made with the upper
Figures 3 and 4 show the jaws open to 40 cm.
magnitude and distribution of dose Backscatter Correction = f1(y) [(y
distributions resulting from individual f2(x)/40) + 1 – y/40]
subsources for a 20x20 cm2 fields where x and y are the lower and
incident on a water phantom upper jaw openings, respectively, and
positioned at at 90 cm source-to- f1 and f2 are normalized such that
surface distance (SSD). Profiles were f1(40) = f2(40) = 1.
taken at the nominal depth of Finally, comparisons with
maximum dose, dmax, (1.5 cm at 6 MV, large-field measurements reveal a
3.2 cm at 18 MV), 10, and 20 cm. The deficit in dose calculations in the
primary photon subsource provides dose-buildup region for open fields.
the largest contribution to dose, This effect, shown for a 38x38 cm2
followed by the scattered-photon field in Figure 5, is evident in depth
subsource associated with the dose curves calculated with both
flattening filter. Scattered photon BEAM and PEREGRINE. These
subsources have depth dose simulations make use of the same
characteristics similar to the primary phase space data from BEAM
photon source. Depth dose simulation of the fixed components of
distributions in Figure 4 show the the accelerator, upstream of the jaws.
electron subsource contributing Results of BEAM and PEREGRINE
significantly to the dose for 6 and 18 agree within a statistical precision of
MV beams, from the surface down less than 1% demonstrating that the
through dmax. Scattered photon and discrepancy is not due to the beam
contaminant electron subsources model or radiation transport physics
play an increasingly important role for in PEREGRINE. The magnitude of the
increasing field size. dose deficit near the surface
Because PEREGRINE does not increases with increasing field size,
account for particles scattered back and goes away for fields blocked by
into the treatment head, a wedges or trays. Based on this
measurement-based method is used evidence, we hypothesize that it is
to estimate the effect of backscatter caused by a source of electrons in the
on the over-response of the monitor accelerator head that is not fully
chamber. While backscatter factor accounted for in the treatment head
measurements have been reported by simulation with BEAM. To account for
several authors, (Reference 39, for this, we increase the weight of the
example) we used measured electron subsource by 120% and 50%
backscatter factors reported by for 6 and 18 MV beams, respectively.
Dunzenli40 were used as these fit our While the 6 MV discrepancy is
11
smaller, it requires a greater beam energy for this study. The use of
proportion of added electron source, profiles to determine beam energy
because a smaller number the source relies on accurate modeling of the
electrons reach the central axis at flattener. Therefore, it is important to
100 cm SSD for 6 MV than for 18 MV. check that depth dose curves are
calculated accurately as well.
C. Beam Commissioning BEAM simulations were
The beam commissioning completed, and the corresponding
procedure consists of two parts: (1) device files generated for a set of
selecting/interpolating the initial beam energies around the nominal
electron energy incident on the energy of the machine. Using
bremsstrahlung target and (2) setting PEREGRINE, dose profiles were
the dose per monitor unit, based on calculated for each of these energies.
the specific calibration of the An interpolation was done between
accelerator. the calculated profiles to find beam
Field flatness is sensitive to beam energy. This energy was then used to
energy.Error: Reference source not generate a new device file by doing a
found Figures 6 and 7 show how the linear interpolation using the nearest
electron beam energy affects the field two device files, without rerunning
flatness and depth dose for 38x38 cm BEAM. For this study, voltages were
and 2x2 cm fields, respectively. The linearly interpolated from a library of
effect of electron beam energy on simulations at 6.0, 6.5, and 7.2 MeV
field flatness is most apparent for the and 17, 18, and 19 MeV for 6 and 18
largest field size. Figure 6 shows that MV beams, respectively. Voltages
a variation of 4% in the off-axis ratio selected were 6.2 and 18.5 MeV for 6
(defined at 10 cm from the axis) at 10 and 18 MV beams.
cm depth results from variation in Once the effective energy is
electron beam energy from 6 to 6.5 determined, we calibrate the internal
MeV, with a variation of 4% from 17 to particle fluence metric in terms of
19 MeV. Small fields have the most dose per monitor unit (MU). The user
sensitive depth dose curves to inputs the cGy/MU at 10 cm depth on
electron energy. Figure 7 shows that the central axis of a 10x10 cm2 field,
2x2 cm2 depth dose curves are most and PEREGRINE uses this number to
affected at shallower depths. For the determine the effective weight of
nominal 6 MV beam, variation in each history, so that dose is
electron beam energy from 6 to 7 MeV calculated in units of cGy/MU.
results in a difference of 7% at a For purposes of simulations, the
depth of 30 cm. A variation of 17 to 19 geometry of collimator jaws, wedges,
MeV results in a maximum difference wedge trays, block trays, and multi-
of less than 3% at a depth of 30 cm. leaf collimators (MLCs) are described
Because large-field flatness is more in terms of density, composition,
sensitive to electron energy than shape, and location. Block thickness,
depth dose for high energy x-ray material, and density, and aperture
beams and because profile shape and MLC leaf positions are
measurements are less subject to described by the user at calculation
systematic errors due to slight time. For the measurements shown in
misalignment of the gantry and beam this paper, wedges are composed of
scanner, it is used to estimate the stainless steel (density: 7.86 g/cm3,
12
block tray. Results agree with ion model. The results of a set of
chamber measurements to better calculation/measurement
than 1% at 6 MV and better than 2% comparisons show the accuracy of
at 18 MV in unblocked areas, better the overall implementation of the
than 3% (6 and 18 MV) in blocked code, including the beam model and
areas. Calculated dose outside the commissioning procedure. The only
penumbra agrees with measurements normalization done was to use a
to better than 6% (0.5% of the single-point calibration. Results,
maximum dose) on the unblocked summarized in Table 3, indicate good
side and better than 3% (0.3% of the agreement between calculations and
maximum dose) on the blocked side. measurements in dose per monitor
A complex, comb pattern, with unit for full dose distributions under
blocked and open regions was used to open fields and for a variety of beam
compare PEREGRINE calculations to modifiers. In the low-dose gradient
photon diode measurements for the regions of the field, utilizing a
multileaf collimator (Figure 15). published correction curve for
Collimators were set to 20x26 cm2. On monitor chamber backscatter and an
the side of the beam close to the 5 empirical correction to the electron
cm wide-open region, the beam is source fluence, PEREGRINE agrees
collimated by two leaves, which with measurements to within 2% of
extend 2-cm beyond the collimator the dose at the measurement point.
jaw, which is set to the edge of the Calculated output factors and wedge
multileaf collimator. Leakage factors are good to within 2%. In the
radiation scattering around the penumbra region, PEREGRINE
collimator jaw and multileaf predictions result in spatial isodose
collimator is responsible for the small discrepancies of less than 1 mm.
peak at the edge of the field. On the 1 Outside the penumbra, discrepancies
cm side of the comb pattern, only the are larger: PEREGRINE systematically
collimator jaw blocks the field. predicts a lower dose than measured,
Because of the large number of high- with discrepancies as high as 15%.
dose-gradient areas, we compare While these differences are large
measurements with photon diode compared to the dose at the
measurements. Agreement between measurement point, they amount to
calculations and measurements is less than 1% discrepancies expressed
generally <1% of maximum dose for as a fraction of the maximum dose.
low-gradient areas of both blocked Where tested (open fields)
and unblocked areas for the 1-cm PEREGRINE agrees with EGS4
(one-leaf-in, one-leaf-out) area of the (BEAM / DOSEXYZ), with both codes
field. PEREGRINE dose reproduces underpredicting dose in the build-up
the small peaks outside the 4-cm side region of large fields and in the area
of the beam. blocked by the collimator jaws by over
2%. This, combined with the
systematic nature of the
IV. Summary and
discrepancies, suggest that the
Conclusions remaining discrepancies over 2%
In this paper, we provide an cannot be attributed to PEREGRINE
overview of the PEREGRINE code, radiation transport or to the beam
including descriptions of the radiation model, but rather to leakage or
transport physics and the x-ray beam
17
VI. References
1
K. R. Shortt, C. K. Ross, A. F. Bielajew and D. W. O. Rogers (1986) “Electron beam dose
2
T. M. Jenkins and A. Rindi, ed (1988) “Monte Carlo Transport of Electrons and Photons”
3
T. R. Mackie (1990) “Applications of the Monte Carlo method in radiotherapy” Dosimetry of
Ionizing Radiation vol 3, ed K. Kase, B. Bjarngard and F. H. Attix (New York: Academic) pp
541-620.
4
P. Andreo (1991) “Monte Carlo techniques in medical physics,” Phys. Med. Biol. 26 861-920.
5
R. Mohan (1997) “Why Monte Carlo?” Proc. XIIth ICCR (Salt Lake City, USA) (Madison , WI:
6
J. J. DeMarco T. D. Solberg, and J. B. Smathers, (1998) “A CT-based Monte Carlo simulation
tool for dosimetry planning and analysis,” Med. Phys. 25(1) 1-11.
7
M. R. Arnfield, C. Hartmann Siantar, J. Siebers, P. Garmon, L. Cox and R. Mohan 2000 “The
impact of electron transport on the accuracy of computed dose,” 27 (6) pp. 1266-1274.
8
H. Neuenschwander, T. R. Mackie, and P. J. Reckwerdt (1995) “MMC – a high-performance
Monte Carlo code for electron beam treatment planning,” Phys Med Biol 40: 543.
9
P. J. Keall and P. W. Hoban (1996) “Super-Monte Carlo: A 3D electron beam dose calculation algorithm,” Med Phys
23:2023.
10
C. L. Hartmann Siantar, P. M. Bergstrom, W. P. Chandler, L. Chase, L. J. Cox, T, et. al. (1997)
“Lawrence Livermore National Laboratory’s PEREGRINE Project,” Proc. XIIth ICCR (Salt
11
M. Fippel, W. Labu, B. Huber and F. Nusslin (1999) “Fast Monte Carlo dose calculation for
photon beams based on the VMC elecgron algorithm” Med Phys 26: 1466-1475.
12
J. Sempau, S. J. Wilderman and A. F. Bielajew (2000) “DPM, a fast, accurate Monte Carlo
code optimized for photon and electron radiotherapy treatment planning dose
dose calculation tool for radiation therapy treatment planning” XIIIth International Conf of
14
R. Mohan, C. Chui and L. Lidofsky (1985) “Energy and angular distributions of photons
15
J. V. Siebers, P. J. Keall, B. Libby and R. Mohan (1999) “Comparison of EGS4 and MCNP4b
Monte Carlo codes for generation of photon phase space distributions for a Varian 2100C,”
16
D. W. O. Rogers, B. A. Faddegon, G. X. Ding, C-M Ma, J. S. Wei and T. R. Mackie (1995)
“BEAM: a Monte Carlo code to simulate radiotherapy treatment units,” Med. Phys. 22 503-
524.
17
D. M. J. Lovelock, C. S. Chui and R. Mohan (1995) “A Monte Carlo model of photon beams
18
G. Kuster, T. Bortfeld and W. Schlegel (1997) “Monte Carlo simulations of radiation beams
from radiotherapy units and beam limiting devices using the program GEANT,” Proc. XIIth
ICCR (Salt Lake City, USA) (Madison , WI: Medical Physics Publishing) 150-152.
19
B. Faddegon, J. Balogh, R. Mackenzie and D. Scora (1998) “Clinical considerations of
Monte Carlo for electron radiotherapy treatment planning,” Radiat Phys Chem 53: 217-227.
20
C.-M. Ma and S. B. Jiang (1999) “Monte Carlo modeling of electron beams from medical
accelerators,” Phys Med Biol 44 (1):R157-R189.
21
A. E. Schach von Wittenau, L. J. Cox, P.M. Bergstrom, W. P. Chandler, C.L. Hartmann
medical accelerator photon-output phase space," Medical Physics 26:7 1196-1211 (1999).
22
J. Deng, S. B. Jiang, A. Kapur, J. Li, T. Pawlicki and C-M. Ma (2000) “Photon beam
characterization and modeling for Monte Carlo treatment planning,” Phys. Med. Biol. 45
411-427.
23
B. A. Faddegon, P. O-Brien, and D. L. D. Mason (1999) “The flatness of Siemens linear
Livermore, 1989)
26
J. H. Hubbell, W. J. Veigele, E.A. Briggs, R.T. Brown, D. T. Cromer and R. J. Howerton, J.
27
C. M. Davisson and R. D. Evans, Rev Mod Phys 24, p.79 (1952)
28
J. H. Hubbell, H. A. Gimm, and I. Overbo, J. Phys. Chem. Ref. Data 9, 1023 (1980).
29
International Commission on Radiation Units and Measurements (ICRU), ICRU Report 37
30
F. H. Attix (1986) Introduction to Radiological Physics, John Wiley & Sons, New York.
31
W.R. Nelson, H.Hirayama, and D.W.O. Rogers, “The EGS4 Code System,” Report SLAC--265,
32
S. T. Perkins, D. E. Cullen and S. M. Seltzer, Lawrence Livermore National Laboratory
33
A. F. Bielajew, R. Mohan and C. Chui, “Improved Bremsstrahlung Photon Angular Sampling
in the EGS4 Code System”, National Research Council of Canada Report PIRS-0203 (1989).
34
B. A. Faddegon, C. K. Ross, D. W. O. Rodgers, “Angular distribution of bremsstrahlung from
15-MeV electrons incident on thick targets of Be, Al, and Pb”, Med. Phys. 18(4) p. 727-739
(1991).
35
I. Lux and L. Koblinger, Monte Carlo Particle Transport Methods: Neutron and Photon
36
M. J. Berger “Monte Carlo Calculation of the Penetration and Diffusion of Fast Charged
Particles” in Methods in Computational Physics, Vol1, (Academic Press, New York, 1963) p.
135.
37
G.~Z. Moliére, “Theorie der Streuung schneller geladener Teilchen. I. Einzelstreuung am
38
A. F. Bielajew and D. W. O. Rogers (1987) “PRESTA: the parameter reduced electron-step
transport algorithm for electron Monte Carlo transport,” Nucl. Instrum. Methods B 18 165-
181.
39
K. L. Lam, M. S. Muthuswamy, and R. K. Ten Haken (1998) “Measurement of backscatter
to the monitor chamber of medical accelerators using target charge,” Med. Phys. 25 (3)
334-338.
40
C. Dunzenli, B. McClean, and C. Field (1993) “Backscatter into the beam monitor chamber:
41
D. Sheikh-Bagheri, D. W. O. Rogers, C. K. Ross and J. P. Seuntjens (2000) “Comparison of
measured and Monte Carlo calculated dose distributions from the NRC linac,” submitted to
Med Phys.
42
M. Westermark, J. Arndt, B. Nilsson, and A. Brahme (2000) “Comparative dosimetry in
43
L. Wang, C-S. Chui, and M. Lovelock (1998) “A patient-specific Monte Carlo dose-