VMC++ versus BEAMnrc: A comparison of simulated linear accelerator
heads for photon beams
F. Hasenbalg,a?M. K. Fix, E. J. Born, and R. Mini
Division of Medical Radiation Physics, Insel Hospital, University of Berne, Berne 3010, Switzerland
Ionizing Radiation Standards, National Research Council of Canada, Ottawa K1A OR6, Canada
?Received 24 September 2007; revised 16 January 2008; accepted for publication 28 January 2008;
published 20 March 2008?
BEAMnrc, a code for simulating medical linear accelerators based on EGSnrc, has been bench-
marked and used extensively in the scientific literature and is therefore often considered to be the
gold standard for Monte Carlo simulations for radiotherapy applications. However, its long com-
putation times make it too slow for the clinical routine and often even for research purposes without
a large investment in computing resources. VMC++ is a much faster code thanks to the intensive use
of variance reduction techniques and a much faster implementation of the condensed history tech-
nique for charged particle transport. A research version of this code is also capable of simulating the
full head of linear accelerators operated in photon mode ?excluding multileaf collimators, hard and
dynamic wedges?. In this work, a validation of the full head simulation at 6 and 18 MV is per-
formed, simulating with VMC++ and BEAMnrc the addition of one head component at a time and
comparing the resulting phase space files. For the comparison, photon and electron fluence, photon
energy fluence, mean energy, and photon spectra are considered. The largest absolute differences
are found in the energy fluences. For all the simulations of the different head components, a very
good agreement ?differences in energy fluences between VMC++ and BEAMnrc ?1%? is obtained.
Only a particular case at 6 MV shows a somewhat larger energy fluence difference of 1.4%.
Dosimetrically, these phase space differences imply an agreement between both codes at the ?1%
level, making VMC++ head module suitable for full head simulations with considerable gain in
efficiency and without loss of accuracy. © 2008 American Association of Physicists in Medicine.
Key words: Monte Carlo simulations, VMC??, BEAMnrc, treatment head
Monte Carlo ?MC? simulations are currently considered the
most accurate way to calculate the dose to be delivered to
patients in radiotherapy treatments. Its importance is espe-
cially relevant when small fields such as those found in in-
tensity modulated radiation therapy are considered or when
inhomogeneities in the treated volumes are present. A correct
computation of the dose to be delivered to the patient not
only involves the right description of the patient in terms of
geometry and element composition but also a detailed mod-
eling of the photon source, namely, the linear accelerator
?linac? head. Although full linac head simulations have been
reported widely in the literature ?see, e.g., the review articles
by Verhaegen and Seuntjens1and by Ma and Jiang2?, they are
often considered impractical for clinical implementation due
to the long simulation times required. Another approach is
based on empirical or semiempirical source models,3–8which
substitute the whole linac head by just a few subsources. The
main advantage of empirical or semiempirical source models
is that their free parameters such as subsource intensities and
positions can be adjusted relatively easy to give dose distri-
butions matching a specific set of measured data without
the excessive simulation times needed by full linac head
In a recent publication Kawrakow and Walters9have
shown that when photon beam linac simulations with
BEAMnrc ?Refs. 10 and 11? employing directional brems-
strahlung splitting12?DBS? are combined with photon split-
ting for the transport in the phantom, the time taken by the
linac head simulation becomes only a relatively small frac-
tion of the overall calculation time. In addition, the overall
simulation time is significantly reduced compared to previ-
ous studies with BEAM and BEAMnrc. An even faster MC
simulation tool may therefore offer the opportunity of per-
forming a full linac simulation for each treatment plan with
acceptable calculation times.
VMC++ ?Ref. 13? ?VMC++ is only available by direct re-
quest to the National Research Council of Canada? is a fast
MC code optimized for external beam radiation treatment
planning. Recently, the accuracy of VMC++ for performing
dose calculations in the patient has been validated in an ar-
ticle by Gardner et al.14The purpose of this work is to
complement this previous effort comparing VMC++ against
BEAMnrc for the simulation of medical linacs. Considering
BEAMnrc ?based on the EGSnrc ?Refs. 15 and 16? code
which uses accurate electron transport algorithms? as the
gold standard, simulations of a 6 and a 18 MV Varian-like
machines were performed covering the energy range nor-
mally encountered in radiotherapy. For each energy, simula-
15211521Med. Phys. 35 „4…, April 20080094-2405/2008/35„4…/1521/11/$23.00© 2008 Am. Assoc. Phys. Med.
tions involving the successive addition of the several ele-
ments of the linac head were done and phase space files were
generated with both codes. Phase space files contain the
charge, energy, position, direction, and statistical weight of
all particles reaching a user-defined scoring plane. These
phase space files were analyzed and compared using the
BEAMDP ?Ref. 17? utility of the 2007 BEAMnrc distribution.
Although the generic VMC++ simulation routines can also
be used to transport particles through beam modifiers such as
hard wedges, blocks, and multileaf collimators ?MLCs?, in
the present study the focus is on static fields collimated with
jaws only that are useful in the initial commissioning pro-
cess. The use of the optimized transport algorithms em-
ployed by VMC++ in the upper portion of the treatment head
and in the photon jaws for beam modifiers such as MLCs or
wedges is currently under investigation.
Since the fast VMC++ treatment head simulation has not
yet been published elsewhere, a brief description of the vari-
ance reduction techniques implemented in the code are given
in Sec. II.
II. MATERIALS AND METHODS
Our linac head model resembles a Varian accelerator head
?Varian Medical Systems, Palo Alto, CA? operated in photon
mode. In this model, the accelerator head is composed of an
electron target ?made of tungsten and copper?, a primary
tungsten collimator, a beryllium vacuum window, a flattening
filter, a simplified kapton monitor chamber, X−Y tungsten
jaws, and a light reticle. The flattening filter is made of cop-
per for the 6 MV beam and of iron and tantalum for the 18
MV beam. In this study, the correct geometry and composi-
tion of the head components are not crucial and therefore
they do not exactly correspond to that of an actual
Varian machine. The input files used for the BEAMnrc and
VMC++ simulations, on the other hand, should be as similar
as possible ?within the constraints imposed by each code?.
Therefore, for this comparative study, a simplified monitor
chamber has been assumed and the mirror, which is not
implemented in the VMC++ treatment head simulation, has
been omitted. Since VMC++ has no medium defined as
vacuum, all vacuum parts were assumed to be filled with air
in both codes.
Simulations were run with VMC++ and BEAMnrc to ob-
tain phase space files. To check VMC++ against BEAMnrc
and to observe the effect of each of the head components,
five different phase space files were generated with each
code for 6 and 18 MV for a total of 20 phase space files.
First, only the electron target is simulated surrounded by air
up to the scoring plane. Second, the electron target, the pri-
mary collimator, and the beryllium vacuum window are
simulated, surrounded by air until the scoring plane. Third,
the flattening filter is added. A fourth phase space file simu-
lates the whole head, adding monitor chamber, jaws and light
reticle for the field size of 10?10 cm2. The fifth phase
space file corresponds to the whole head simulation and a
field size of 40?40 cm2. In all these phase space files, the
scoring plane was located at a source-to-surface distance
?SSD? of 100 cm. We chose SSD=100 cm in order to mag-
nify any small differences that could show up in the phase
BEAMDP was used with 40 equal area circular bins for the
first three type of phase space files where cylindrical sym-
metry is granted and 40 equal area square bins for the phase
space files of the full head simulations. For each of the simu-
lations the quantities: all particles estimated real fluence,
??r?, all particles energy fluence, ??r?, all particles mean
energy and photon energy spectrum were studied. The pho-
ton spectrum was integrated over the radial or square dis-
tance depending on the case. To avoid redundancy, we con-
centrate on energy fluence and mean energy and show the
photon spectra when appropriate. Each phase file contained
enough particles to obtain smooth curves of the studied
quantities. The largest relative differences between the codes
were observed in the energy fluences. The relative uncertain-
ties in the energy fluences and in the mean energies varied
from case to case and between regions of high and low par-
ticle fluence. Relative uncertainties are generally larger at
large radii ?large square distances? where the bins of equal
area are narrower. To get an idea of the typical error bars
involved, these are shown in the mean energy curves in the
results section. To quantify the differences between both
codes VMC++ and BEAMnrc?, we make use of the relative
energy fluence differences between the two simulation meth-
ods with respect to the maximum of one of these
methodsconsidered as the
II.A. Simulations with VMC++
VMC++ is based on the original ideas of Kawrakow,
Fippel, and Friedrich18for electron transport and later ex-
tended and improved by Fippel19and Kawrakow20for the
transport of photons. VMC++ is written in C++ and incorpo-
rates several improvements in the transport efficiency as well
as in the modeling of the underlying physical processes ?e.g.,
the exact multiple scattering theory of Kawrakow and
Bielajew21is employed including spin effects in the same
way as the EGSnrc code,22instead of the small-angle ap-
proximation used in
VMC/XVMC, the history repetition
method found in VMC/XVMC is replaced with STOPS ?Ref. 13?
thus removing the VMC/XVMC limitation to be applicable in
low-Z materials only, etc.?. A research version of the VMC+
+ code ?National Research Council of Canada? which dy-
namically loads a particle source shared library ?a full linac
head simulation module in this case? at run time is used in
this study. The linac simulation can be easily configured us-
ing a set of input files which define the head geometry and
control the MC transport and variance reduction ?VR? pa-
rameters. For details about the VR techniques already imple-
mented in previous versions of VMC++, the reader is referred
to Ref. 20 where these are described more deeply. The set of
transport and variance reduction parameters used in this
study is listed in Table I. The most important variance reduc-
tion parameter is the radiative splitting number, with the
highest value for the 6 MV 10?10 cm2field and the lowest
Hasenbalg et al.: VMC++ versus BEAMnrc: Comparison of simulated accelerator heads1522
Medical Physics, Vol. 35, No. 4, April 2008
16I. Kawrakow, “Accurate condensed history Monte Carlo simulation of
electron transport. II. Application to ion chamber response,” Med. Phys.
27, 499–513 ?2000?.
17C. M. Ma and D. W. O. Rogers, “BEAMDP Users Manual,” NRC Report
No. PIRS-509, 2006 ?unpublished?.
18I. Kawrakow, M. Fippel, and K. Friedrich, “3D electron dose calculation
using a Voxel based Monte Carlo algorithm ?VMC?,” Med. Phys. 23, 445–
19M. Fippel, “Fast Monte Carlo dose calculation for photon beams based on
the VMC electron algorithm,” Med. Phys. 26, 1466–1475 ?1999?.
20I. Kawrakow and M. Fippel, “Investigation of variance reduction tech-
niques for Monte Carlo photon dose calculation using XVMC,” Phys. Med.
Biol. 45, 2163–2184 ?2000?.
21I. Kawrakow and A. F. Bielajew, “On the representation of electron mul-
tiple elastic-scattering distributions for Monte Carlo calculations,” Nucl.
Instrum. Methods Phys. Res. B 134, 325–336 ?1998?.
22I. Kawrakow and D. W. O. Rogers, “The EGSnrc code system: Monte
Carlo simulation of electron and photon transport,” NRC Report No.
PIRS-701, 2003 ?National Research Council of Canada?.
23E. S. M. Ali and D. W. O. Rogers, “Efficiency improvements of x-ray
simulations in EGSnrc user-codes using bremsstrahlung cross-section en-
hancement ?BCSE?,” Med. Phys. 34, 2143–2154 ?2007?.
24D. Sheikh-Bagheri and D. W. O. Rogers, “Sensitivity of megavoltage
photon beam Monte Carlo simulations to electron beam parameters,”
Med. Phys. 29, 379–390 ?2002?.
25H. W. Koch and J. W. Motz, “Bremsstrahlung cross-section formulas and
related data,” Rev. Mod. Phys. 31, 920–955 ?1959?.
26F. Hasenbalg, H. Neuenschwander, R. Mini, and E. J. Born, “Collapsed
cone convolution and analytical anisotropic algorithm dose calculations
compared to VMC++ Monte Carlo simulations in clinical cases,” Phys.
Med. Biol. 52, 3679–3691 ?2007?.
Hasenbalg et al.: VMC++ versus BEAMnrc: Comparison of simulated accelerator heads 1531
Medical Physics, Vol. 35, No. 4, April 2008