BJR © 2014 The Authors.

Published by the British Institute of Radiology

Received: Revised: Accepted: doi: 10.1259/bjr.20140163

24 February 2014 10 July 2014 14 July 2014

Cite this article as:

Papagiannis P, Pantelis E, Karaiskos P. Current state of the art brachytherapy treatment planning dosimetry algorithms. Br J Radiol 2014;87:
20140163.

BRACHYTHERAPY DOSIMETRY SPECIAL FEATURE: REVIEW

ARTICLE
Current state of the art brachytherapy treatment planning
dosimetry algorithms
P PAPAGIANNIS, PhD, E PANTELIS, PhD and P KARAISKOS, PhD
Medical Physics Laboratory, Medical School, National and Kapodistrian University of Athens, Athens, Greece

Address correspondence to: Mr Panagiotis Papagiannis

E-mail: [email protected]

ABSTRACT
Following literature contributions delineating the deficiencies introduced by the approximations of conventional brachytherapy
dosimetry, different model-based dosimetry algorithms have been incorporated into commercial systems for 192Ir brachy-
therapy treatment planning. The calculation settings of these algorithms are pre-configured according to criteria established by
their developers for optimizing computation speed vs accuracy. Their clinical use is hence straightforward. A basic
understanding of these algorithms and their limitations is essential, however, for commissioning; detecting differences from
conventional algorithms; explaining their origin; assessing their impact; and maintaining global uniformity of clinical practice.

Conventional, Task Group (TG)43-based1 dosimetry marked results of their performance. Monte Carlo (MC) simu-
an improvement over prior dose calculation formalisms for lation is also briefly discussed since, besides being a
brachytherapy treatment planning by advocating the use of a candidate MBDCA for clinical implementation, it is used
source strength quantity traceable to international standards, for obtaining input data for MBDCAs, as well as for their
the introduction of two-dimensional (2D) source anisotropy, testing.
and global uniformity in source characterization as well as
clinical dosimetry practice. In the past decade, brachytherapy DOSIMETRY IN THE BRACHYTHERAPY
has progressed from the traditional surgical paradigm to PHOTON ENERGY RANGE
modern three-dimensional (3D) image-based treatment The vast majority of brachytherapy sources can be con-
planning systems (TPSs) and dose delivery. The information sidered pure photon emitters.1–3 Figure 1 readily implies
available through patient imaging, however, had not been that mainly density heterogeneities affect dosimetry of
fully exploited since TG43-based dosimetry relies on source- higher photon energy-emitting sources like 192Ir, since at-
specific data pre-calculated in a standard homogeneous water tenuation and energy absorption per unit mass is compa-
geometry.1–3 Hence, it disregards patient-specific radiation rable for all materials except for bone. On the contrary,
scatter conditions and the radiological differences of tissue or attenuation and energy absorption differences exist be-
applicator materials from water. tween different tissue compositions for lower photon en-
ergies. Literature on the effect of tissue composition
In response to literature on the effect of these shortcomings, heterogeneities in the energy range of 125I and 103Pd has
which has been reviewed in several recent publications,4–7 been recently reviewed.6,7 Average elemental composition
TPSs have become commercially available that include im- cannot, however, be obtained through routine clinical
proved dosimetry algorithms, collectively referred to as imaging methods, with promising alternatives such as dual
model-based dosimetry algorithms (MBDCAs). At the time energy or spectral CT requiring further research.6
of writing, these include a deterministic solver of the linear
Boltzmann transport equation (LBTE)8–10 and a collapsed The men/m ratio plotted in the inset of Figure 1 vs energy
cone superposition (CCS) algorithm11–17 for 192Ir high-dose- quantifies the average energy absorbed per interaction. The
rate (HDR) applications. minimum of men/m at energies around 60–100 keV implies
that incoherently scattered radiation will play an important
This work reviews the basic features of these algorithms role in dose deposition for sources emitting photons of
and their clinical implementation and presents illustrative energy higher than that. This is because a large number of
 BJR P Papagiannis et al

Figure 1. Mass attenuation and mass energy absorption coefficients plotted as a function of photon energy in the range of
brachytherapy for selected media: water, lung, prostate and cortical bone.6 The inset presents corresponding ratios of the linear
energy absorption to attenuation coefficients in the same energy range.

incoherent interactions of minimal energy transfer are expected should not be confused with the effect of placing a source close to
for photons accumulated between 60 and 100 keV, and the mean a geometry boundary. The latter is position dependent (Figure 3)
free path of 60 keV photons is significant (about 3.4 cm in with the greatest differences between dose rate distributions
water). obtained with a source situated centrically and eccentrically, ob-
served at points where scatter predominates.8,20,21
Figure 2 presents dose rate distributions for point sources
emitting photons of energy equal to the average photon energy For photon energies pertinent to brachytherapy, the ranges of
of 125I and 192Ir. MC results in this figure, and throughout this secondary electrons are small relative to photon mean free paths
work, were calculated using MCNP v. 6.1.18 The factors de- (e.g. Carlsson Tedgren et al22). Charged particle equilibrium (CPE)
termining the dose rate distributions, besides source strength can be therefore assumed to exist at all points separated from the
and the inverse square law that are factored out in Figure 2, are geometry boundaries or the source by a distance at least equal to
attenuation and scatter. In Figure 2 dose rate is separated into the aforementioned range. Under conditions of CPE, dose equals
its primary and scatter components. Scatter dose is the greatest collision kerma in millimetre-sized voxel elements and can be
at points at distances beyond 2.5 and 6.5 cm for the low- and approximated by kerma since radiative energy loss of secondary
high-energy source, respectively. Dose rate distributions are electrons is negligible in tissue media.6 In short, to know the dose
shown for spherical water geometries of different radii. The distribution one needs to know the energy distribution of fluence,
increasing lack of backscatter affects points at distances from FE,23 at all points of a geometry:
the edge of the geometry smaller than the mean free path of
backscattered photons and makes each bounded geometry ð
unique in terms of dosimetry.13 The effect of phantom DðrÞ5 KðrÞ5 EFE ðrÞ½men ðEÞ=rdE (1)
dimensions,19 which is a one-dimensional effect in the ex- E
ample of Figure 2, which corresponds to a symmetric geometry,

2 of 20 birpublications.org/bjr Br J Radiol;87:20140163
 Review Article: Current state of the art brachytherapy treatment planning dosimetry algorithms BJR

Figure 2. Monte Carlo primary, scatter and total dose rate results calculated using a general purpose code plotted against radial
distance from a point source centred in a water sphere of radius of 5, 10 and 15 cm, emitting photons with energies of (a) 28 and
(b) 355 keV. All results are multiplied by the square of distance from the source.

This is trivial for primary photons since: element, reduced by the number of photons that were absorbed
or scattered out of this phase space element. This is the time-
prim Qprim expð2mrÞ independent LBTE for photon transport only:
FE 5 (2)
4pr 2
 
where Qprim is the number of photons emitted from a mono-     Qprim E; Ω ^  
energetic source. The spectrum of scatter radiation (Figure 4) ^
Ω×=FΩ;E
^ 5Qsc r; E; Ω
r; E; Ω ^ 1 d r 2 rp
formed by photon transport in the geometry also needs to be 4p
 
taken into account. ^
2 st ðr; EÞFΩ;E r; E; Ω
Photon transport in a geometry can be described by the LBTE. (3)
Assume a point photon emitting source in a bounded homogeneous
geometry of volume V. Let FΩ;E ðr; E; ΩÞ^ be the energy distribution where Qprim/4p is the number of photons per solid angle ele-
of the time-integrated particle radiance commonly referred to as ment emitted from the point source situated at position rp, st is
angular fluence (FΩ;E equals dF/dΩdE, where F 5 dN/dA is the the macroscopic total cross-section and Qsc is the photon scat-
particle fluence23). It is a scalar function of position r 5 (x, y, z), tering source:
^
energy and direction given by the direction cosines Ω5ðu; v; wÞ.
These variables are collectively referred to as phase space; the six-
dimensional space of photon states (recall u2 1 v2 1 w2 5 1). Hence,   ð‘ ð    
FΩ;E denotes the number of photons in phase space ele- Qsc ^ 5
r; E; Ω ^ Ω
ss r; E9→E; Ω× ^ 9 FΩ;E r; E; Ω
^ 9 dΩ
^ 9dE9
^ that is, passing through area dA normal to Ω
mentðr; E; ΩÞ, ^ and 0 4p
located at r, with Ω between Ω and Ω 1 dΩ and E between E and (4)
E 1 dE. It follows that FΩ;E from points outside V is zero.
where ss is the macroscopic differential scattering cross-section.
At any part of the phase space, conservation of energy dictates
particle balance so that the streaming of photons through phase
space element ðr; E; ΩÞ^ (i.e. photons of energy E crossing Since Equation (3) is linear, primary scatter separation (PSS) can
^ equals photons scattered in it from all prim
be employed: FΩ;Ε 5FΩ;Ε 1 Fsc
dV 5 dAdr along Ω) Ω;Ε ; where the arguments were
others plus photons emitted by the source into this phase space discarded for simplicity.

3 of 20 birpublications.org/bjr Br J Radiol;87:20140163
 BJR P Papagiannis et al

Figure 3. Monte Carlo-calculated isodose rate contours in units of cGy h21 U21 around a point source emitting photons with energies
of (a) 28 and (b) 355 keV situated 10 cm from the centre of a spherical water phantom of 15-cm radius. Corresponding results
obtained assuming 15 cm of water around the source to resemble the geometry commonly used for TG43 dosimetric calculations1
are also plotted for comparison. An appropriate colour map is used to depict the percentage differences on a pixel-by-pixel basis.

Equation (3) can then be split to its counterparts for primary encapsulation and heterogeneities must be taken into account.
and scatter radiation: This requires ray tracing from points within the source to each
point in the geometry to define the pathlength traversed in each
 
^  medium.24–26
Qprim E; Ω 
^ Fprim
Ω×= Ω;E 5
prim
d r 2 r p 2 st ðr; EÞ FΩ;E (5)
4p Only Equation (6) then remains to be solved. This is, however,
an integro-differential equation of six variables that cannot be
solved analytically. One has to seek recourse to numerical sol-
^ Fsc 5Qsc 2 st ðr; EÞ Fsc
Ω×= (6)
Ω;Ε Ω;Ε utions of either stochastic (MC) or deterministic nature (LBTE
solvers and CCS). Semi-empirical dosimetry models character-
The analytical solution of Equation (5) is analogous to that of ized by improved accuracy and computational efficiency, but not
Equation (2). If the source is polyenergetic, Equation (2) can be general applicability, have also been described in the literature
applied using an effective attenuation coefficient weighted over (e.g. Beaulieu et al6 and references therein).
the energy distribution of energy fluence for primary photons:
THE MONTE CARLO SIMULATION METHOD
+ fi Ei mi MC is a numerical method that relies on random sampling to
meff 5 i (7) produce observations on which statistical inference can be per-
+ fi Ei
i formed to extract information about a system, based on the law
of large numbers and the central limit theorem. It can be used to
where fi is the emission probability of primary photon with Ei arrive at a stochastic approximation of the LBTE solution using
and attenuation coefficient mi24 (Figure 5). When considering random sampling from the probability distributions of physical
real brachytherapy sources in a heterogeneous geometry, the processes underlying photon transport and a set of input data
effects of source geometry factor, attenuation within the source, (geometry, materials and cross-sections).27

4 of 20 birpublications.org/bjr Br J Radiol;87:20140163
 Review Article: Current state of the art brachytherapy treatment planning dosimetry algorithms BJR

Figure 4. Monte Carlo results of the energy distribution of photon fluence at 1-, 5- and 10-cm distance per starting particle emitted
from a point 125Ι (top) and 192Ir (bottom) source centred in a 15-cm radius water phantom, multiplied by distance squared.

Details on the method and its application to radiation therapy the increased number of interactions within it). Efficiency gains
dosimetry can be found in dedicated textbooks (e.g. Seco and can also be obtained through parallel computations on central as
Verhaegen28). MC has played a major role in promoting the well as graphics processing units.31 MC has been shown to be
accuracy of brachytherapy dosimetry through providing high- capable of calculation times suitable for clinical implementation.
quality single source data for use with TG43-based TPS. Several For example, MC calculation times have been reported that
general purpose codes have been benchmarked by various au- range from subminute to a few minutes for low-dose-rate bra-
thor groups, and methodological recommendations as well as chytherapy applications,32–34 from 2.5 to 17.0 min for 403–1403
relevant literature can be found in Rivard et al2 and Perez- 2-mm voxels, respectively, for an HDR rectal application,35 and
Calatayud et al.3 from subsecond for a single source to a couple of seconds for an
HDR plus shield implant using graphics processing unit
Despite being considered as a reference dosimetry method for implementation.36
brachytherapy, MC has not been implemented in commercial
TPS yet. Traditionally, this has been attributed to its computa- LINEAR BOLTZMANN TRANSPORT EQUATION
tional efficiency. Besides Type B uncertainty, including code and SOLVER BASICS AND PRACTICAL IMPLICATIONS
input data components,2,29 as an inherently stochastic method, The finite difference method, an effective iterative method for
MC is also subject to Type A uncertainty. This is characterized solving differential equations, can be used to solve Equation (6)
by the quotient of the variance of the output over the square numerically. The phase space is discretized to establish a finite
root of the number of initial photons simulated, N. Achieving difference grid of all variables, hence the commonly used term
a given factor of uncertainty reduction therefore requires an grid-based Boltzmann solvers (GBBSs). The resulting system
increase of N by the square of this factor, for the same variance. of equations is iteratively solved using the source iteration
This significantly prolongs computational time and becomes method.37
futile beyond a certain point. Obviously, MC is more time
consuming for higher photon energy brachytherapy sources The scattering source in Equation (4) is numerically integrated
owing to multiple scattering (Figures 1, 2 and 4). Variance re- ^ in spherical harmonics up to a given
by expanding FΩ;E ðr; E; ΩÞ
duction techniques exist, however, that range from elaborate30 scattering order L and combining them with cross-sections ex-
to simple (i.e. energy cut-off, scoring geometry truncation, panded using Legendre polynomials.37,38 Evaluating the integral
dose approximation by collision kerma, track-length estima- in a finite number of directions and demanding Equation (6) to
tors, analytical calculations for primary dose and use of phase hold in the same directions is an angular differencing scheme. It
space files for photons exiting a source to preclude simulating is capable of efficient treatment of anisotropic scattering, and it

5 of 20 birpublications.org/bjr Br J Radiol;87:20140163
 BJR P Papagiannis et al

Figure 5. Mean mass attenuation (left), effective mass attenuation (centre) and mean mass energy absorption (right) coefficients,
plotted against distance from a point 125Ι (top) and 192Ir (bottom) source centred in a 15-cm radius water phantom. Mean and
effective coefficients were calculated by weighting over the separate fluence and energy fluence distribution data of Figure 4.

is commonly referred to as the discrete ordinates method scattering, beginning with the use of the primary fluence in
(DOM). The set of directions and corresponding weights is the scattering integral of Equation (4) to determine the first
commonly referred to as a quadrature set of order N (corre- scatter source. The increase of quadrature set order and voxel
sponding to N2 1 2N directions). For the discretization of en- size would also mitigate ray effects, increasing, however, cal-
ergy, the range from the maximum energy of photons emitted culation time and potentially introducing volume averaging
by the source to a minimum energy is divided in groups of uncertainties.39,40
intervals that can be variable, with larger ones used for
higher energies. A LBTE is constructed for every energy group Regarding energy discretization, the multigroup approximation
and each direction. Space is also discretized in volume is realistic only if cross-sections do not vary considerably within
elements raising the number of equations to be solved to each group. This is not the case at energies and materials where
Nelement 3 Ndirections 3 Nenergy groups. After the source iteration photoelectric absorption probability is considerable. The 192Ir
method converges to the solution, dose calculations are per- spectrum is quite forgiving in this regard. Besides accuracy, time
formed in a post-processing step as the sum of results from is of the essence for clinical implementation, and the impact of
Equation (1) applied for each energy group. Heterogeneities energy groups for brachytherapy sources on both has been
are inherently accounted for in Equations (5) and (6). The studied in the literature.40 Multiple scattering with minimal
solution for the primary photon fluence at all points of the energy transfer that occurs in the 192Ir energy range (Figures 1
geometry can be calculated through ray tracing. and 6),41 as well as coherent scattering that is significant at low
energies and high Z materials, delay convergence of the source
Since a LBTE is solved at every voxel of the geometry only for iteration. Diffusion synthetic acceleration algorithms have been
a finite number of directions, an artificial build-up of particle developed to reduce calculation time in problems where scat-
fluence can occur along these directions. This gives rise to tering predominates.38 The scattering order used in the expan-
a discretization artefact in the form of rays in the calculated dose sion of the scattering source is also related to the photon energy
distribution. Ray effects are most evident for high gradients of spectrum and computation time increases with increasing order.
scatter fluence, at points where the relative importance of pri- Cross-sections recast in energy groups, and Legendre polynomial
mary dose is small (i.e. at relatively increased distances from expanded format suitable for scattering source definition are re-
a point source). Employing PSS mitigates ray effects. A further quired. These are obtained using custom software like the CEPXS.42
benefit can be obtained from separating the scatter source to
a first scatter source originating from primary photons and Since LBTEs include the flow of photons through each element
a multiple scatter one originating from higher orders of scat- surface besides photon fluence in a volume element, an ap-
tering. This is an iterative procedure referred to as successive proximate expression is required to relate volume and surface

6 of 20 birpublications.org/bjr Br J Radiol;87:20140163
 Review Article: Current state of the art brachytherapy treatment planning dosimetry algorithms BJR

Figure 6. Primary, first scatter, multiple scatter and total dose rate results41 (multiplied by distance from the source squared) plotted
against distance along the transversal bisector of the VariSource™ VS2000 (Varian Medical Systems, Palo Alto, CA) (a) and the
microSelectron® mHDR-v2 (Nucletron, Veenendaal, Netherlands) (b) 192Ir high-dose-rate brachytherapy sources centred in a water
phantom of 15-cm radius.

averaged fluence and reduce the number of unknowns. Various The BV-Acuros algorithm is proprietary. Its basic features43 have
methods are used, a common approach being the use of linear- also been reviewed elsewhere.44 In short, Acuros uses an energy
discontinuous finite element methods (DFEMs).37 These discretization scheme of 37 groups for primary photons (pri-
methods allow for the calculation of fluence everywhere in each mary is used here in the sense of photons that emerge from an
192
element. Spatial discretization artefacts could, however, ensue in Ir source, see below). An efficiency gain is obtained by col-
the form of local solution over- or underestimations owing to lapsing this energy grouping scheme for the scatter radiation. A
conflicts of the approximate character of the expression used for scattering order L 5 3 is used for scattering within energy groups
the spatial variation of fluence with particle balance.43,44 (scatter events that change direction with minimal energy
transfer) and L 5 2 for scattering outside energy groups to re-
LINEAR BOLTZMANN TRANSPORT EQUATION solve the anisotropy of scatter radiation for the 192Ir energies.
SOLVER CLINICAL IMPLEMENTATION Triangular-Chebyshev quadrature sets are used for angular dis-
GBBS algorithms are not new. The DOM for angular dis- cretization and the integration for the generation of the scat-
cretization that is most commonly used in medical physics tering source. Order ranges from N 5 4 to 30 (24–960 discrete
applications39,40,45–50 was introduced in the 1950s. Since GBBS angles) both within an energy group and between energy groups.
algorithms were used primarily for neutron transport and
shielding problems to enhance calculation efficiency in view of Sources supported by BV-Acuros are represented by effective
increased absorption, they did not become particularly popular sources based on pre-calculated data on energy and angle-
within the radiotherapy (RT) community. dependent intensity of photons. Ray tracing is performed for the
definition of primary fluence in the geometry using a limited
The first GBBS algorithm to be incorporated in a commercially number of points representing each source. Successive scattering
available 192Ir brachytherapy TPS was Acuros™ [BrachyVision™ is employed wherein a first scatter source is calculated from the
(BV); Varian® Medical Systems, Palo Alto, CA].8–10,51 Acuros is primary fluence.
a version of the Attila GBBS developed at Los Alamos National
Laboratory (Los Alamos, NM)38 optimized for brachytherapy, For spatial discretization, Acuros uses a Cartesian spatially var-
and later also for external beam therapy.52,53 iable computational grid where element size is adapted based

7 of 20 birpublications.org/bjr Br J Radiol;87:20140163
 BJR P Papagiannis et al

on scatter fluence gradient to promote time efficiency while DðrÞ Dprim ðrÞ Dsc ðrÞ Dprim ðrÞ Sðr9Þ
5 1 5 1∭ kðjr 2 r9j; uÞd 3 r9
preserving accuracy and mitigating ray effects. Primary fluence is R R R R V R
mapped at several points within each computational element (8)
through ray tracing from source positions, and the Galerkin
linear DFEM is used. where R is the radiant energy of the source,23 S is the scatter
energy released per unit mass from a volume element at r9
Material properties are derived from X-ray CT imaging, except (commonly referred to as SCERMA),16 and k is the point energy
for applicators that are overlaid in the geometry using an ap- deposition kernel expressed as the fraction of SCERMA released
plicator model library containing geometry and material in- from the volume element at point r9 that is deposited per vol-
formation that are taken into account in the ray tracing for the ume element at r.
definition of the primary fluence. Individual voxel densities are
derived from a CT calibration, and density is assumed to be Primary dose can be easily calculated, and, for a monoenergetic
constant within computational grid elements. Material ele- point source, SCERMA can be calculated from the primary
mental composition is drawn from a density look-up table, in- dose as:
cluding lung, adipose, skeletal muscle, cartilage and bone
tissues with compositions from International Commision on  
m 2 men
Radiation Protection Publication 23,54 as well as air, alu- SðrÞ5 Dprim ðrÞ (9)
men
minium, titanium and stainless steel. Cross-sections are
prepared using CEPXS42 that includes all photon interactions
The point kernel, k, can be calculated analytically using ef-
except for Rayleigh scatter that is of low relative importance
fective linear absorption and attenuation coefficients pro-
for the 192Ir energies except when applicators with high Z
vided the energy spectrum of scattered photons as a function
materials are included in the geometry.
of direction u relative to the direction of primary photons is
known. In practice, however, MC simulation is used to gen-
All Acuros calculation settings are preset, and the user only erate k in water in tabular format. Results are then parame-
needs to specify a dose output grid and its resolution. The terized using exponential functions to reduce computer
output grid extent affects calculation time. Its resolution only storage requirements, mitigate the effects of MC simulation
affects accuracy since the calculation grid is defined in- type A uncertainties and facilitate efficient superposition
dependently using the spatially variable discretization scheme methods. This results to a set of angle-dependent param-
discussed above. The dose calculation grid is automatically eters for water. In order to account for heterogeneities, ray
defined as the dose output grid plus 10 cm in all directions, tracing is employed with appropriate factors to scale the
unless the CT image boundary is reached, to account for back water kernel for the difference in energy release in a medium
scatter. relative to water and attenuation of energy in different media
along r, using effective energy absorption and attenuation
Dose to any medium can be calculated using macroscopic kerma coefficients weighted over the spectrum of scattered
cross-sections. BV-Acuros v. 10.0.33 used for calculating results photons.11–16
presented in the following sections calculates dose to water in
the heterogeneous geometry. A following version incorporated
Direct superposition according to Equation (8) for every pair
the option to calculate dose to voxel medium in the heteroge-
of voxels in the geometry is impractical owing to the time
neous geometry to comply with TG186 recommendations.6
required for the N7 operations when heterogeneity correction
Acuros dosimetry remains slow for results to be used in dose
is employed for a volume partitioned to N3 voxels. Since the
optimization (from a couple of minutes for single-source cal-
scaled kernel is not spatially invariant,56 convolution tech-
culations over a 30-cm diameter water sphere with 1-mm res-
niques and fast Fourier transform algorithms of reduced
olution, up to 7 min for a multicatheter breast implant). Dose
calculation complexity cannot be used. The CCS method is an
prescription is performed using TG43 in accordance with TG186
efficient alternative.11–15 Its main assumption is that all the
recommendations.6
energy released into coaxial cones of a given solid angle is
transported, attenuated and deposited in elements on that
COLLAPSED CONE SUPERPOSITION BASICS AND axis. For implementation in the discretized Cartesian co-
PRACTICAL IMPLICATIONS ordinates inherent to patient imaging, a set of solid angle
The superposition principle has been successfully used in ex- elements are determined and each of these elements yields
ternal beam therapy for decades employing different kernel a transport direction. A lattice of parallel lines along each
representations and implementation schemes (e.g. Ahnesjö and transport direction is constructed so that every voxel in the
Aspradakis55 and references therein). Readers interested in the geometry is traversed by at least one such transport line. Ray
details of adaptation of the method in brachytherapy dosimetry tracing is performed along transport lines, and dose at a voxel
are referred to a corresponding series of publications.11–16 is given by the sum of the contributions of all directions.11–13
The complexity of a CCS algorithm is MN3, where M is the
In brief, under conditions of CPE and using PSS,16 the dose at total number of solid angle elements (or equivalent transport
point r can be calculated by integrating the contribution of directions), which is practically many orders of magnitude
energy released from all other points, r9, according to: less than that of direct superposition.

8 of 20 birpublications.org/bjr Br J Radiol;87:20140163
 Review Article: Current state of the art brachytherapy treatment planning dosimetry algorithms BJR

The uncertainty associated with the main assumption of the it computationally intense. The acceleration factor provided by
CCS method increases with distance from an interaction point general purpose parallel computing on graphical processing
since cone surface, or the number of voxels over which scattered units57–61 has made the method applicable. Following imple-
energy should be distributed, increases as r2DΩ. This would be mentation for external beam treatment planning, Oncentra®
manifested as a star-shaped dose pattern at large distances where Brachy v. 4.4 (Nucletron, an Elekta company, Veenendaal,
the relative importance of primary dose decreases. This un- Netherlands) was released in 2013 including the capability for
192
certainty is small when the kernel values decrease rapidly with Ir dosimetry calculations based on the CCS method, under
distance from a scatter generation point. It is further alleviated the commercial name Oncentra-ACE (Advanced Collapsed cone
when SCERMA does not vary significantly with distance so that Engine).
the lack of SCERMA deposition at voxels out of a cone direction
from voxels upstream is counterbalanced from SCERMA The algorithm is proprietary. Based on information available at
deposition from adjacent voxels. In brachytherapy, and es- the time of writing, it features a multiresolution Cartesian cal-
pecially 192Ir, the decrease of the kernel is reduced by multiple culation grid. The resolution is 1 mm in a cube containing the
scatter (Figure 6), and SCERMA decreases rapidly with dis- source dwell positions, and 2, 5 and 10 mm in three additional
tance from a source.13 Increasing the number of transport cubes outside it that extend up to the geometry boundaries, as
directions and the resolution of the transport direction lattice, the latter is defined through patient imaging. All calculation
or the voxel size, would limit this discretization uncertainty settings are preset, and the user is only presented with two
at the expense of calculation time and averaging errors, options denoted as standard and high accuracy levels. These
respectively. options control the set of margins from the source dwell
positions that define the above mentioned four cubes (stan-
The relative importance of multiple scatter in brachytherapy dard: 1, 8, 20 and 50 cm; and high: 8, 20, 35 and 50 cm).
introduces another problem since the field of scattered radiation Successive scattering superposition is employed, and the ac-
is assumed to be unidirectional within each fixed solid angle curacy level choice also controls the number of transport
element, and this applies only to once scattered photons. When directions defined by spherical tessellation and used for CCS
the contribution of multiply scattered photons to dose is sig- of scatter dose from once and multiply scattered photons
nificant, as in the intermediate and high brachytherapy energies (standard: M1sc 5 320, Mmsc 5 180 and high: M1sc 5 720,
(Figures 1, 2 and 6), accuracy deteriorates away from the sources Mmsc 5 240). The number of transport directions is also au-
and close to geometry boundaries (Figures 2 and 6). Therefore, tomatically adapted according to the number of source dwell
a successive scattering superposition approach has been in- positions since the SCERMA gradient is less steep than in the
troduced.13,14 MC simulation can be used to generate kernels case of a single source.11
from generations of scattered photons up to any given order.
The once scattered photons kernel, k1sc, can be used to calculate Pre-calculated data obtained from MC simulation are used.
dose from these photons, i.e. D1sc. The SCERMA transferred to These include kernels, spectra for the primary, once scattered,
twice scattered photons, S2sc, is calculated from D1sc using and multiple scattered photons, necessary for the calculation of
Equation (9), and the method is repeated to calculate the dose heterogeneity scaling factors, and primary dose required for the
from multiply scattered photons, Dmsc, using the corresponding estimation of S1sc. Primary dose is obtained from source-specific
kernel, kmsc. The partitioning of SCERMA also alleviates the MC calculated dose rate distributions in water scored separately
problem of the steep total SCERMA gradient discussed above. It for primary and scattered photons.15,16 Such PSS data41 are
increases calculation time, however, making it proportional to a source characterization equivalent to that of the TG43
the (M1sc 1 Mmsc)N3. formalism.16,22

The choice of M1sc and Mmsc is the result of optimizing calcu- Material definitions, and all relevant parameters, are obtained
lation time vs the distance, or equivalently the dose level, where from user-defined regions of interest and consensus data from
discretization artefacts will occur.11 The steeper SCERMA dis- the TG186 for patient tissues,6 as well as an applicator library in
tribution for once scattered photons dictates that M1sc . Mmsc.11 case one is included in the geometry. The effect of not using
individual voxel densities that can be obtained from patient
Additional refinements are necessary for CCS implementation in radiographic CT imaging is expected to be small for the 192Ir
brachytherapy dosimetry. These include accounting for the energies.62 An advantage of the method is that it is not vul-
presence of high Z shields,12 evaluating the angle between the nerable to CT artefacts and that imaging modalities other than
direction of the inherently divergent, primary radiation and each CT can also be used for dosimetry planning. Nevertheless, CT-
transport direction for every ray-trace step and selecting the derived densities will be used in a future version in compliance
appropriate angle-dependent kernel parameters13 and ray trac- with TG186 recommendations.
ing for the assignment of SCERMA values at voxels close to
a source.13 Details can be found in the cited literature.12,13 CCS reports dose to medium in the heterogeneous geometry.
The method remains slow for dose calculations to be used in
COLLAPSED CONE SUPERPOSITION: dose optimization (from minutes for a single source to tens of
CLINICAL IMPLEMENTATION minutes for a multicatheter breast implant using the high-
Although the CCS method has been shown to be accurate for accuracy option). Dose prescription is performed using TG43 in
brachytherapy dosimetry, many of the above implications made accordance with TG186 recommendations.6

9 of 20 birpublications.org/bjr Br J Radiol;87:20140163
 BJR P Papagiannis et al

ILLUSTRATIVE BRACHYTHERAPY shown in Figure 7. Significant differences are observed between

DOSIMETRY RESULTS TG43 and MBDCAs for both TPSs. For the VS2000, the ob-
Illustrative calculations were performed using BV-Acuros served differences are compatible with previous results in the
v. 10.0.33 and Oncentra-ACE v.4.4 (high-accuracy option) in the literature.8,51 In general, Acuros and TG43 agree within 62% for
same three voxelized computational models prepared in the form the majority of points away from the source longitudinal axis (z).
of a digital imaging and communications in medicine (DICOM) Differences at the drive wire side (z , 0) are attributed to the
radiographic CT image series. The purpose of these calculations is difference of cable length assumed for the calculation of the source
not a TPS intercomparison. After all, treatment plans for the same phase space in Acuros and the MC simulation for the derivation of
model differ since a different source is used with each system TG43 parameters.63 Differences at z . 0 are attributed to the
(VariSource VS2000 (Nucletron, Veenendaal, Netherlands) and anisotropy values calculated by the TG43 algorithm of the TPS.51
microSelectron, mHDR-v2 respectively) and, generally, plans can- ACE results are also generally within 62% at points at distances
not be transferred between the two systems owing to differences in up to 5 cm from the source. Further away, discretization artefacts
the implementation of conformance to the DICOM-RT protocol. become evident, and ACE generally overestimates dose. The
switch of resolution of the multiresolution Cartesian calculation
The purpose of this section is to demonstrate the strengths and grid is also discerned in Figure 7b. The pattern of dose differences
limitations of MBDCAs. Reference MC results were therefore in the radial direction at relatively increased distances is not as-
prepared using a general purpose MC code18 and a custom sociated with ACE but rather with TPS extrapolation uncertainties
made, brachytherapy-dedicated, software tool for the automatic since TG43 data are limited to distances up to 5 cm.46
preparation of input files from treatment plans exported in
DICOM-RT format. TPS calculations were also performed using
TG43. The spatial resolution was the same for all TPS results and Findings of Figure 7 might seem alarming especially for TG43
MC calculations corresponding to the same computational model. users accustomed to using TPS commissioning acceptance cri-
Further details can be found in previous published work.8–10 teria of a few percent at individual points. The reader should
recall, however, that single-source tests are the hardest ones for
Single-source and bounded both MBDCAs, that the geometry factor is the major de-
homogeneous geometries terminant of dose deposition in brachytherapy so that points at
MBDCAs should provide results compatible with TG43 for increased distances are of relatively small importance and,
single sources in the centre of the homogeneous water geometry most important, that MBDCAs require optimization to achieve
used to obtain the latter. This is not, however, the case as clinically viable calculation times. MBDCAs are optimized for

Figure 7. A comparison of TG43 and model-based dosimetry algorithm single-source dosimetry as exported from the same
treatment planning system, presented on the central coronal plane for a single dwell position brachytherapy treatment plan with the
VS2000TM (Varian Medical Systems, Palo Alto, CA) (a) and the microSelectron® mHDR-v2 (Nucletron, Veenendaal, Netherlands)
(b) 192Ir high-dose-rate brachytherapy source centred in a 15-cm radius water phantom. Besides isodose contours, percentage
differences between (a) BrachyVision™ (BV; Varian® Medical Systems, Palo Alto, CA) TG43 and BV-Acuros™ and (b) Oncentra®
Brachy v. 4.4 (Nucletron an Elekta company, Veenendaal, Netherlands) TG43 and Oncentra Brachy advanced collapsed cone engine
(TG186) are plotted on a pixel-by pixel-basis using an appropriate colour map.

10 of 20 birpublications.org/bjr Br J Radiol;87:20140163
 Review Article: Current state of the art brachytherapy treatment planning dosimetry algorithms BJR

clinical brachytherapy where multiple source dwell positions are source catheter, increased percentage dose differences are observed
employed. only at points within the bone heterogeneities. These are
probably owing to the inability of the algorithm to account for
Multiple sources and stylized, patient equivalent the difference in the spectra of scattered photons in bone relative
computational models to water.12,68
In the following, MC, TG43 and MBDCAs are compared in two
mathematical equation-based, stylized models resembling an Results for the BV-TPS in the breast model are presented in
oesophageal and a breast brachytherapy patient.21,64 The models Figure 10. Findings are compatible with results presented in the
were voxelized through their conversion to a series of contig- literature using the same MC code without the use of the soft-
uous CT images in DICOM format to facilitate their import to ware for the automated preparation of input files.10 In short,
the TPSs.10 MC results are considered as reference. Their BV-TG43 calculations overestimate the dose distribution, espe-
comparison with TG43 dosimetry results depicts the type B cially at relatively increased distances, owing to the combined
uncertainty introduced by TG43 assumptions, and their com- effect of the lung heterogeneity and the finite patient dimen-
parison with MBDCAs is used to depict how effective these sions. A BV-TG43 dose underestimation is observed at points
algorithms are. either within the catheters, since these are not excluded in any of
the data sets, or close to the sources, probably owing to the
Results for the BV-TPS in the oesophagus model are presented extrapolation algorithm employed by the TPS. A slight dose
in Figure 8. Findings are compatible with results presented in the underestimation is also observed in the positive x side of the
literature using the same MC code without the use of the soft- planning target volume that is attributed to the different source
ware for the automated preparation of input files.10 BV-TG43 drive wire length assumed in the simulations of this work and
calculations are correct at points close to the source catheter the simulations for the data used as TG43 input to the TPS.
owing to the predominance of primary dose (Figure 2). Dose to Acuros results correctly account for lung and finite patient
the spinal cord is, however, overestimated up to 16%, mainly dimensions, and major differences with MC results are observed
owing to the inability to account for the increased attenuation of only within the catheters or close to the sources.
photon fluence in the vertebrae. Dose to the trachea and the
sternum bone is underestimated for similar reasons. Dose dif- Results for the Oncentra TPS are similar. The pattern of
ferences in the lungs depend on the relative position of the point Oncentra-TG43 differences from MC results in Figure 11b is the
of interest relative to the source. Acuros results are in excellent same as in Figure 10b besides using MC results of dose to water
agreement (62%) with MC results for the majority of points in and dose to medium, respectively. ACE succeeds in taking these
the model. Differences up to 6% are observed at a limited effects into account with major differences observed again only
number of points lying close to the directions defined by pri- within the catheters or close to the sources.
mary photon rays tangential to the spine and the trachea
structures, possibly owing to ray tracing, the increased scatter ISSUES ASSOCIATED WITH CLINICAL ADOPTION
fluence gradient in the penumbra of these structures and po- OF MODEL-BASED DOSIMETRY ALGORITHMS
tential difference in the discretization errors inherently involved As already mentioned, all calculation settings are preset in cur-
in MC and BV-Acuros results. A dose overestimation is also rent versions of TPS including an MBDCA option, according to
observed at points behind the bone heterogeneities, at increased an optimization scheme of accuracy vs calculation time. User
distance from the source catheter probably owing to imperfect control over algorithm parameters is therefore limited to spec-
resolving of the angular distribution of scatter. ifying the extent and/or the resolution of the dose output grid.
Calculations are then performed at the push of a button. Cau-
Results for the Oncentra TPS in the oesophagus model are tion is needed in that the dose calculation grid to yield results
presented in Figure 9. MC results presented in Figures 8 and 9 presented in the dose output grid is limited to the volume de-
correspond to, collision kerma approximated, dose to water and fined by the image series used for planning. This can compro-
dose to medium, respectively, to match the dose reporting mise the dosimetry accuracy at the external axial planes of the
convention of Acuros and ACE in the TPS versions used in this image series and points close to axial image boundaries of
work. The fact that the pattern of differences between MC and cropped image series, since backscatter radiation, which is sig-
TPS TG43 data are similar in Figures 8 and 9 is an indirect nificant for the 192Ir energies (Figure 2), will not be taken into
confirmation that the adopted dose reporting convention has account. The implant will always be included in the image series.
little effect for the 192Ir energies in the large cavity approxima- The effect of missing backscatter depends on geometry, mate-
tion (i.e. in terms of dose to millimetre-sized voxels). The issue rials and distance from the implant, and its significance is
of dose reporting convention with MBDCAs in brachytherapy is mitigated by the inverse square law that governs dose deposition
still open until conclusive results are available on the biologically in brachytherapy. Indicative results are presented in Figure 12
relevant target, the dose descriptor best describing its response for the oesophagus and breast models used in this work. CT over
and the implementation of cavity theory to derive this dose scanning in the z axis is not warranted unless the importance of
descriptor from MBDCA results taking into account source accuracy in dose volume histogram data for a parallel organ at
photon energy.6,65–67 risk (OAR) is deemed high. Image cropping is unacceptable.

In Figure 9, ACE results are in good agreement with the cor- Tissue segmentation defined as the spatial delineation of dif-
responding MC calculations. Except from points away from the ferent regions of a 3D image data set and assignment of elemental

11 of 20 birpublications.org/bjr Br J Radiol;87:20140163
 BJR P Papagiannis et al

Figure 8. (a) The central axial image of the voxelized oesophageal model with BrachyVision™ (BV; Varian® Medical Systems, Palo
Alto, CA)-TG43, BV-Acuros™ (Varian Medical Systems) and corresponding Monte Carlo (MC)-calculated dose results for an
oesophageal brachytherapy plan presented in the form of percentage isodose lines within the extent of the user-defined dose
output grid. (b) The spatial distribution of the percentage dose differences between BV-TG43 and MC results [100 3 (DTG43 2 DMC)/
DMC] on the central axial plane, presented on a pixel-by-pixel basis using an appropriate colour map. (c) The spatial distribution of
the percentage dose differences between BV-Acuros and MC results [100 3 (DACUROS 2 DMC)/DMC] on the central axial plane,
presented on a pixel-by-pixel basis using an appropriate colour map.

compositions and mass density6 is another issue. The TG186 representative tissues was set forth. For the 192Ir energies, tissue
report6 critically reviewed the subject with particular focus on density is the determining factor for individualized patient do-
low-energy brachytherapy where it is of increased importance simetry (Figure 1). Apart from density, all human tissues, except
(Figure 1). In the absence of a robust method for the direct for bone, are almost water equivalent under the large cavity
extraction of interaction coefficients from CT images, a table of assumption (i.e. for millimetre-sized voxels) regardless of dose

12 of 20 birpublications.org/bjr Br J Radiol;87:20140163
 Review Article: Current state of the art brachytherapy treatment planning dosimetry algorithms BJR

Figure 9. (a) The central axial image of the voxelized oesophageal model with Oncentra® Brachy v. 4.4 (Nucletron an Elekta
company, Veenendaal, Netherlands)-TG43, Oncentra-advanced collapsed cone engine (ACE) (TG186) and corresponding Monte
Carlo (MC) calculated dose results for an oesophageal brachytherapy plan presented in the form of percentage isodose lines within
the extent of the user-defined dose output grid. (b) The spatial distribution of the percentage dose differences between Oncentra-
TG43 and MC results [100 3 (DTG43 2 DMC)/DMC] on the central axial plane, presented on a pixel-by-pixel basis using an appropriate
colour map. (c) The spatial distribution of the percentage dose differences between Oncentra-ACE and MC results [100 3
(DTG186 2 DMC)/DMC] on the central axial plane, presented on a pixel-by-pixel basis using an appropriate colour map.

reporting convention, and TPS relying on individual voxel The effect of CT calibration uncertainty to 192Ir dosimetry is
density mapping combined with a limited material composition presented in Figure 13. Assuming the CT uncertainty is, on
look-up table are accurate for 192Ir dosimetry.62 average, 615 HU,69 two extreme CT calibrations were prepared.

13 of 20 birpublications.org/bjr Br J Radiol;87:20140163
 BJR P Papagiannis et al

Figure 10. (a) The central axial image of the voxelized breast model with BrachyVision™ (BV; Varian® Medical Systems, Palo Alto,
CA)-TG43, BV-Acuros™ (Varian Medical Systems) and corresponding Monte Carlo (MC) calculated dose results for a breast
brachytherapy plan presented in the form of percentage isodose lines within the extent of the user-defined dose output grid. (b)
The spatial distribution of the percentage dose differences between BV-TG43 and MC results [100 3 (DTG43 2 DMC)/DMC] on the
central axial plane, presented on a pixel-by-pixel basis using an appropriate colour map. (c) The spatial distribution of the
percentage dose differences between BV-Acuros and MC results [100 3 (DACUROS 2 DMC)/DMC] on the central axial plane, presented
on a pixel-by-pixel basis using an appropriate colour map.

Figure 13 summarizes the percentage differences between BV- assuming a rectangular distribution for Hounsfield units. Dose
Acuros dose calculations performed for a breast and a head differences that pile up with distance from the implant are small,
and neck case when these two extreme calibrations were used, affecting only dose to OARs under 2%.

14 of 20 birpublications.org/bjr Br J Radiol;87:20140163
 Review Article: Current state of the art brachytherapy treatment planning dosimetry algorithms BJR

Figure 11. (a) The central axial image of the voxelized breast model with Oncentra® Brachy v. 4.4 (Nucletron an Elekta company,
Veenendaal, Netherlands)-TG43, Oncentra-advanced collapsed cone engine (ACE) (TG186) and corresponding Monte Carlo (MC)-
calculated dose results for a breast brachytherapy plan presented in the form of percentage isodose lines within the extent of the
user-defined dose output grid. (b) The spatial distribution of the percentage dose differences between Oncentra-TG43 and MC
results [100 3 (DTG43 2 DMC)/DMC] on the central axial plane, presented on a pixel-by-pixel basis using an appropriate colour map. (c) The
spatial distribution of the percentage dose differences between Oncentra-ACE (TG186) and MC results [100 3 (DTG186 2 DMC)/DMC] on the
central axial plane, presented on a pixel-by-pixel basis using an appropriate colour map.

15 of 20 birpublications.org/bjr Br J Radiol;87:20140163
 BJR P Papagiannis et al

Figure 12. (a) Percentage dose differences on the central axial plane of the voxelized oesophagus model between BrachyVision™
(BV; Varian® Medical Systems, Palo Alto, CA)-Acuros™ (Varian Medical Systems) calculations performed for the full model and the
model symmetrically reduced from 17 to 12 cm in the z axis. (b) Same as (a) on an axial plane 4 cm away from the central axial plane.
(c) Percentage dose differences on the central axial plane of the voxelized breast model between BV-Acuros calculations
performed for the full model and the model symmetrically reduced from 10 to 8 cm in the z axis.

16 of 20 birpublications.org/bjr Br J Radiol;87:20140163
 Review Article: Current state of the art brachytherapy treatment planning dosimetry algorithms BJR

Figure 13. The dosimetric uncertainty from an uncertainty of 615 HU (on average for all Hounsfield units69) presented using an
appropriate colour map on the central axial plane of (a) a voxelized oesophagus model and (b) a voxelized head and neck model.
Results were calculated assuming a rectangular distribution and using BrachyVision™ (Varian® Medical Systems, Palo Alto, CA)-
Acuros™ (Varian Medical Systems) for dose calculations.

In view of different methods, or even different implementations conditions (Level II) using reference, MC generated or exper-
of the same method in MBDCA-based TPS, augmenting com- imentally determined, dose distributions for particular tests.6
missioning procedures is key to its clinical integration without
compromising the global uniformity achieved with TG43. The A workflow is proposed and a working group has been set up
TG186 has proposed a graded approach to the commissioning of seeking to address limited test case availability, tools and
MBDCA-based TPS,6 similar to the procedure followed in a se- methods. Significant progress has been made by the group in
ries of works for the dosimetric benchmarking of Acuros.8–10 that a theoretical source was prepared that was included in the
Verification of the MBDCA to reproduce consensus TG43 data3 systems of both vendors of TPSs incorporating MBDCAs.70 Ex-
for sources supported by the TPS is referred to as Level I ample test case plans and reference results, to be made available
commissioning. In the absence of reference data in DICOM-RT through a web registry, are under way. It is envisioned that these
dose format for import to the TPS and appropriate TPS tools for can be uniform amongst vendors and included in their com-
2D comparisons, this could become a tedious process of point missioning test procedures to depict differences between TG43
calculations that could easily miss localized differences, as shown and MBDCAs.6 A similar, yet more flexible approach, would be
to have been carried out for TG43 TPS algorithms in the dis- the preparation of end-user-oriented tools to assist the evalua-
cussion above. Definitive acceptance criteria are also hard to tion of MBDCAs.
form since the optimization of an MBDCA could lead to
significant differences in bounded, single-source geometries
THE CLINICAL BENEFIT EXPECTED FROM MODEL-
(Figure 7). The recommendation to carefully examine dif-
BASED DOSIMETRY ALGORITHMS
ferences, understand and document the clinical impact5 is
The sensitivity of anatomic sites treated with brachytherapy to
therefore much more useful than the criterion of 2% differ-
limitations introduced by TG43 assumptions have been reviewed
ence adopted from TG43U1.2
in the literature.4,22 For the 192Ir energies where MBDCAs have
been introduced, sites potentially affected include breast, head
Level I commissioning is to be followed by checks of efficiency and neck, superficial applications and all applications involving
of the algorithm in accounting for heterogeneities and scatter shielded catheters. MBDCAs are expected to improve dosimetric

17 of 20 birpublications.org/bjr Br J Radiol;87:20140163
 BJR P Papagiannis et al

accuracy, as shown in the literature reviewed, and affirmed by tracing, PSS, successive scattering methods, use of pre-calculated
illustrative results of this work. data, increase of calculation time with number of sources, use of
applicator libraries and pre-fixed calculation settings in their
Regarding the clinical benefit expected from improved dosi- clinical implementation according to vendor optimization schemes,
metric accuracy, it must be kept in mind that MBDCAs remain to name but a few.
slow for plan optimization, and dose prescription is performed
using TG43.6 Hence the potential of MBDCAs is currently reserved Presented results confirm that both clinically available model-
for radiation oncologist assessment. Clinical protocols are not based dose calculation algorithms achieve a significant dosimetric
expected to change either, at least not until a sufficient volume of accuracy improvement compared with TG43-based calculations.
clinical data become available that justify such changes. The clinical benefit from improved accuracy remains to be
evaluated. The explanation of differences observed between
Such data could be in the form of planning dosimetry studies for model-based and TG43-based algorithms is not trivial and
patient cohorts.71,72 It is important to understand that dosi- requires a basic understanding of all methods coupled with
metric effects that are currently not taken into account but affect augmented test procedures and user-oriented tools.
the patient population in the same manner (direction and
magnitude) are included, at least to some extent, in clinical Treatment planning using model-based dose calculation algo-
protocols through clinical trial data. rithms should not be considered a simple investment on a new
TPS mandated by recommendations or current trend, but rather
The expected benefit of MBDCAs in dosimetry planning is in as an opportunity for concerted, interdisciplinary effort to reach
the amount of reduction it will achieve (through the in- improved levels of care quality in the near future.
dividualization of patient dosimetry) in the variance of the re-
sponse of clinical trial populations. This is equivalent to a FUNDING
reduction of uncertainty that must be contrasted, however, to The authors acknowledge financial support from a research
the combined dosimetric uncertainty in brachytherapy from grant cofinanced by the European Union [European Social Fund
physical and clinical sources.5 (ESF)] and Greek national funds through the Operational Pro-
gramme “Education and Lifelong Learning: investing in knowl-
Apart from that, a clear benefit can be expected from the need to edge society” of the National Strategic Reference Framework
augment commissioning and quality assurance procedures in- (NSRF). Research Funding Programme: Aristeia.
troduced by MBDCAs, and the active involvement of physicists
in examining differences and radiation oncologists in assessing ACKNOWLEDGMENTS
their impact. The assistance of V. Peppa, MSc and K. Zourari, PhD in pre-
paring data and figures for this work is gratefully acknowledged.
CONCLUSIONS Varian Medical Systems (Palo Alto, CA) and Nucletron, an
Model-based dose calculation algorithms for brachytherapy Elekta company (Veenendaal, Netherlands) are acknowledged
dosimetry share a lot of similar features: a long heritage of use for providing BrachyVisionTM-AcurosTM v. 10.0.33 and Oncentra®
outside brachytherapy, angular and spatial discretization, ray Brachy v. 4.4 for research purposes.

REFERENCES

1. Nath R, Anderson LL, Luxton G, Weaver KA, 4. Rivard MJ, Venselaar JL, Beaulieu L. The current issues, and trends in brachytherapy
Williamson JF, Meigooni AS. Dosimetry of evolution of brachytherapy treatment plan- physics. Med Phys 2008; 35: 4708.
interstitial brachytherapy sources: recom- ning. Med Phys 2009; 36: 2136–53. 8. Zourari K, Pantelis E, Moutsatsos A, Petrokok-
mendations of the AAPM Radiation Therapy 5. Kirisits C, Rivard MJ, Baltas D, Ballester F, De kinos L, Karaiskos P, Sakelliou L, et al.
Committee Task Group No. 43. American Brabandere M, van der Laarse R, et al. Review Dosimetric accuracy of a deterministic radiation
Association of Physicists in Medicine. Med of clinical brachytherapy uncertainties: anal- transport based 192Ir brachytherapy treatment
Phys 1995; 22: 209–34. ysis guidelines of GEC-ESTRO and the planning system. Part I: single sources and
2. Rivard MJ, Coursey BM, DeWerd LA, AAPM. Radiother Oncol 2014; 110: 199–212. bounded homogeneous geometries. Med Phys
Hanson WF, Huq MS, Ibbott GS, et al. doi: 10.1016/j.radonc.2013.11.002 2010; 37: 649–61.
Update of AAPM Task Group No. 43 Report: 6. Beaulieu L, Carlsson Tedgren A, Carrier JF, 9. Petrokokkinos L, Zourari K, Pantelis E,
a revised AAPM protocol for brachytherapy Davis SD, Mourtada F, Rivard MJ, et al. Report Moutsatsos A, Karaiskos P, Sakelliou L, et al.
dose calculations. Med Phys 2004; 31: 633–74. of the Task Group 186 on model-based dose Dosimetric accuracy of a deterministic radi-
3. Perez-Calatayud J, Ballester F, Das RK, calculation methods in brachytherapy beyond ation transport based 192Ir brachytherapy
Dewerd LA, Ibbott GS, Meigooni AS, et al. the TG-43 formalism: current status and treatment planning system. Part II: Monte
Dose calculation for photon-emitting bra- recommendations for clinical implementation. Carlo and experimental verification of
chytherapy sources with average energy Med Phys 2012; 39: 6208–36. a multiple source dwell position plan
higher than 50 keV: report of the AAPM and 7. Thomadsen BR, Williamson JF, Rivard MJ, employing a shielded applicator. Med Phys
ESTRO. Med Phys 2012; 39: 2904–29. Meigooni AS. Anniversary paper: past and 2011; 38: 1981–92.

18 of 20 birpublications.org/bjr Br J Radiol;87:20140163
 Review Article: Current state of the art brachytherapy treatment planning dosimetry algorithms BJR

10. Zourari K, Pantelis E, Moutsatsos A, Sakelliou L, 22. Carlsson Tedgren A, Verhaegen F, Beaulieu 34. Afsharpour H, Landry G, D’Amours M,
Georgiou E, Karaiskos P, et al. Dosimetric L. On the introduction of model based Enger S, Reniers B, Poon E, et al. ALGEBRA:
accuracy of a deterministic radiation transport algorithms performing nonwater hetero- ALgorithm for the heterogeneous dosimetry
based (192)Ir brachytherapy treatment planning geneity corrections into brachytherapy based on GEANT4 for BRAchytherapy. Phys
system. Part III. Comparison to Monte Carlo treatment planning. In: Venselaar J, Baltas Med Biol 2012; 57: 3273–80. doi: 10.1088/
simulation in voxelized anatomical computa- D, Meigooni AS, Hoskin PJ, eds. Compre- 0031-9155/57/11/3273
tional models. Med Phys 2013; 40: 011712. hensive brachytherapy: physical and clinical 35. Poon E, Le Y, Williamson JF, Verhaegen F.
11. Carlsson Tedgren A, Ahnesjö A. Optimization aspects. Boca Raton, FL: CRC Press Taylor BrachyGUI: an adjunct to an accelerated
of the computational efficiency of a 3D, & Francis; 2013. pp. 145–60. Monte Carlo photon transport code for
collapsed cone dose calculation algorithm for 23. Thomas DJ. ICRU report 85: fundamental patient-specific brachytherapy dose calcula-
brachytherapy. Med Phys 2008; 35: 1611–18. quantities and units for ionizing radiation. tions and analysis. J Phys Conf Ser 2008; 102:
12. Tedgren AK, Ahnesjö A. Accounting for high Radiat Prot Dosimetry 2012; 150: 550–2. 012018.
Z shields in brachytherapy using collapsed 24. Karaiskos P, Angelopoulos A, Baras P, Rozaki- 36. Hissoiny S, D’Amours M, Ozell B, Despres P,
cone superposition for scatter dose calcula- Mavrouli H, Sandilos P, Vlachos L, et al. Dose Beaulieu L. Sub-second high dose rate
tion. Med Phys 2003; 30: 2206–17. rate calculations around 192 Ir brachytherapy brachytherapy Monte Carlo dose calculations
13. Carlsson AK, Ahnesjö A. The collapsed cone sources using a Sievert integration model. with bGPUMCD. Med Phys 2012; 39:
superposition algorithm applied to scatter Phys Med Biol 2000; 45: 383–98. 4559–67. doi: 10.1118/1.4730500
dose calculations in brachytherapy. Med Phys 25. Pantelis E, Baltas D, Dardoufas K, Karaiskos 37. Lewis EE, Miller WF. Computational methods of
2000; 27: 2320–32. P, Papagiannis P, Rosaki-Mavrouli H, et al. neutron transport. New York, NY: Wiley; 1984.
14. Carlsson ÅK, Ahnesjö A. Point kernels and On the dosimetric accuracy of a Sievert 38. Wareing TA, McGhee JM, Morel JE, Pautz
superposition methods for scatter dose cal- integration model in the proximity of 192Ir SD. Discontinuous finite element SN meth-
culations in brachytherapy. Phys Med Biol HDR sources. Int J Radiat Oncol 2002; 53: ods on three-dimensional unstructured grids.
2000; 45: 357–82. 1071–84. Nucl Sci Eng 2001; 138: 256–68.
15. Russell KR, Tedgren AK, Ahnesjö A. 26. Williamson JF. The sievert integral revisited: 39. Gifford KA, Horton JL, Wareing TA, Failla G,
Brachytherapy source characterization for evaluation and extension to125I, 169Yb, and Mourtada F. Comparison of a finite-element
improved dose calculations using primary 192Ir brachytherapy sources. Int J Radiat multigroup discrete-ordinates code with
and scatter dose separation. Med Phys Oncol 1996; 36: 1239–50. Monte Carlo for radiotherapy calculations.
2005; 32: 2739. 27. Williamson JF, Rivard MJ. Quantitative Phys Med Biol 2006; 51: 2253–65. doi:
16. Russell KR, Ahnesjö A. Dose calculation in dosimetry methods for brachytherapy. In: 10.1088/0031-9155/51/9/010
brachytherapy for a 192Ir source using Thomadsen BR, Rivard MJ, Butler WM, eds. 40. Gifford KA, Price MJ, Horton JL, Wareing
a primary and scatter dose separation tech- Brachytherapy physics. 2nd edn. Madison, WI: TA, Mourtada F. Optimization of determin-
nique. Phys Med Biol 1996; 41: 1007–24. Medical Physics; 2005. pp. 233–94. istic transport parameters for the calculation
17. Van Veelen B. Collapsed cone dose calculations 28. Seco J, Verhaegen F, eds. Monte Carlo of the dose distribution around a high dose-
for model based brachytherapy utilizing Monte techniques in radiation therapy. Boca Raton, rate 192Ir brachytherapy source. Med Phys
Carlo source characterization data. Proceed- FL: Taylor & Francis Group; 2013. 2008; 35: 2279–85.
ings of the International Workshop on Recent 29. DeWerd LA, Ibbott GS, Meigooni AS, Mitch 41. Taylor RE, Rogers DW. An EGSnrc Monte
Advances in Monte Carlo Techniques for MG, Rivard MJ, Stump KE, et al. A Carlo-calculated database of TG-43 parame-
Radiation Therapy; 8–10 June 2011; Montreal, dosimetric uncertainty analysis for photon- ters. Med Phys 2008; 35: 4228–41.
Canada. Montreal, Canada: McGill University, emitting brachytherapy sources: report of 42. Lorence LJ Jr, Morel JE, Valdez GD. Physics
2011. AAPM Task Group No. 138 and GEC- guide to CEPXS: a multigroup coupled
18. Goorley T, James M, Booth T, Brown F, Bull J, ESTRO. Med Phys 2011; 38: 782–801. electron-photon cross-section generating code
Cox LJ, et al. Initial MCNP6 release overview. 30. Sheikh-Bagheri, Daryoush Kawrakow I, version 1.0. Albuquerque, NM: Sandia Na-
Nucl Tech 2012; 180: 298–315. Walters B, Rogers DWO. Monte Carlo tional Laboratories; 1989.
19. Peŕ ez-Calatayud J, Granero D, Ballester F. simulations: efficiency improvement techni- 43. BrachyVision-Acuros algorithm reference guide
Phantom size in brachytherapy source dosi- ques and statistical considerations. Integrat- (P/N B5202151R01A). Palo Alto, CA: Varian
metric studies. Med Phys 2004; 31: 2075–81. ing new technologies into the clinic: Monte Medical Systems Inc.; 2009.
20. Lymperopoulou G, Papagiannis P, Carlo and image-guided radiation therapy— 44. Papagiannis P, Beaulieu L, Mourtada F.
Angelopoulos A, Karaiskos P, Georgiou E, Proceedings of 2006 AAPM Summer School. Computational methods for the dosimetric
Baltas D. A dosimetric comparison of Madison, WI: Medical Physics Publishing; characterization of brachytherapy sources. In:
169Yb and 192Ir for HDR brachytherapy of the 2006. pp. 71–91. Venselaar JL, Baltas D, Meigooni AS, Hoskin
breast, accounting for the effect of finite patient 31. Pratx G, Xing L. GPU computing in medical PJ, eds. Comprehensive brachytherapy: physical
dimensions and tissue inhomogeneities. Med physics: a review. Med Phys 2011; 38: and clinical aspects. Boca Raton, FL: CRC
Phys 2006; 33: 4583–9. 2685–97. Press Taylor & Francis; 2013. pp. 85–103.
21. Pantelis E, Papagiannis P, Karaiskos P, 32. Chibani O, Williamson JF. MCPI: a sub- 45. Börgers C. Complexity of Monte Carlo and
Angelopoulos A, Anagnostopoulos G, Baltas minute Monte Carlo dose calculation engine deterministic dose-calculation methods. Phys
D, et al. The effect of finite patient dimen- for prostate implants. Med Phys 2005; 32: Med Biol 1998; 43: 517–28.
sions and tissue inhomogeneities on dosim- 3688–98. 46. Daskalov GM, Baker RS, Rogers DW,
etry planning of 192Ir HDR breast 33. Thomson R, Yegin G, Taylor R, Sutherland J, Williamson JF. Dosimetric modeling of the
brachytherapy: a Monte Carlo dose verifica- Rogers D. Fast Monte Carlo dose calculations microselectron high-dose rate 192Ir source
tion study. Int J Radiat Oncol Biol Phys 2005; for brachytherapy with BrachyDose. Med by the multigroup discrete ordinates method.
61: 1596–602. Phys 2010; 37: 3910. Med Phys 2000; 27: 2307–19.

19 of 20 birpublications.org/bjr Br J Radiol;87:20140163
 BJR P Papagiannis et al

47. Daskalov GM, Baker RS, Rogers DW, 55. Ahnesjö A, Aspradakis MM. Dose calcula- surrogate for the dose to cell nuclei at
Williamson JF. Multigroup discrete ordi- tions for external photon beams in radio- brachytherapy photon energies. Phys Med
nates modeling of 125I 6702 seed dose therapy. Phys Med Biol 1999; 44: R99–155. Biol 2012; 57: 4489–500. doi: 10.1088/0031-
distributions using a broad energy-group 56. Williamson JF, Baker RS, Li ZF. A convolu- 9155/57/14/4489
cross section representation. Med Phys tion algorithm for brachytherapy dose com- 66. Thomson RM, Tedgren ÅC, Williamson JF.
2002; 29: 113–24. putations in heterogeneous geometries. Med On the biological basis for competing
48. Gifford KA, Wareing TA, Failla G, Horton JL, Phys 1991; 18: 1256–65. macroscopic dose descriptors for kilovoltage
Eifel PJ, Mourtada F. Comparison of a 3-D 57. Hissoiny S, Ozell B, Despreś P. Fast dosimetry: cellular dosimetry for brachy-
multi-group SN particle transport code with convolution-superposition dose calculation therapy and diagnostic radiology. Phys Med
Monte Carlo for intracavitary brachytherapy on graphics hardware. Med Phys 2009; 36: Biol 2013; 58: 1123–50. doi: 10.1088/0031-
of the cervix uteri. J Appl Clin Med Phys 2010; 1998–2005. 9155/58/4/1123
11: 3103. 58. de Greef M, Crezee J, van Eijk JC, Pool R, Bel 67. Tedgren ÅC, Carlsson GA. Specification of
49. Han T, Mikell JK, Salehpour M, Mourtada F. A. Accelerated ray tracing for radiotherapy absorbed dose to water using model-based
Dosimetric comparison of Acuros XB de- dose calculations on a GPU. Med Phys 2009; dose calculation algorithms for treatment
terministic radiation transport method with 36: 4095–102. planning in brachytherapy. Phys Med Biol
Monte Carlo and model-based convolution 59. Zhou B, Yu CX, Chen DZ, Hu XS. GPU- 2013; 58: 2561–79.
methods in heterogeneous media. Med Phys accelerated Monte Carlo convolution/ 68. Anagnostopoulos G, Baltas D, Karaiskos P,
2011; 38: 2651–64. superposition implementation for dose cal- Pantelis E, Papagiannis P, Sakelliou L. An
50. Kan MW, Yu PK, Leung LH. A review on the culation. Med Phys 2010; 37: 5593–603. analytical dosimetry model as a step towards
use of grid-based Boltzmann equation solvers 60. Chen Q, Chen M, Lu W. Ultrafast accounting for inhomogeneities and bounded
for dose calculation in external photon beam convolution/superposition using tabulated geometries in 192Ir brachytherapy treatment
treatment planning. Biomed Res Int 2013; and exponential kernels on GPU. Med Phys planning. Phys Med Biol 2003; 48: 1625–47.
2013: 692874. doi: 10.1155/2013/692874 2011; 38: 1150–61. 69. Cozzi L, Fogliata A, Buffa F, Bieri S.
51. Mikell JK, Mourtada F. Dosimetric impact of an 61. Jacques R, Taylor R, Wong J, McNutt T. Dosimetric impact of computed tomography
192Ir brachytherapy source cable length mod- Towards real-time radiation therapy: GPU calibration on a commercial treatment plan-
eled using a grid-based Boltzmann transport accelerated superposition/convolution. Com- ning system for external radiation therapy.
equation solver. Med Phys 2010; 37: 4733–43. put Methods Programs Biomed 2010; 98: Radiother Oncol 1998; 48: 335–8.
52. Vassiliev ON, Wareing TA, McGhee J, Failla 285–92. doi: 10.1016/j.cmpb.2009.07.004 70. Ballester F, Vijande J, Granero D, Rivard MJ,
G, Salehpour MR, Mourtada F. Validation of 62. Peppa V, Zourari K, Pantelis E, Papagiannis P. Thomson R, Beaulieu L. A Generic high-
a new grid-based Boltzmann equation solver Tissue segmentation significance for individ- dose-rate 192Ir source model for model-based
for dose calculation in radiotherapy with ualized 192Ir brachytherapy dosimetry. dose calculation methods in brachytherapy
photon beams. Phys Med Biol 2010; 55: Radiother Oncol 2013; 106: S371. beyond the TG-43 formalism. Brachytherapy
581–98. doi: 10.1088/0031-9155/55/3/002 63. Angelopoulos A, Baras P Sakelliou L 2013; 12: S62–3.
53. Han T, Mourtada F, Kisling K, Mikell J, Karaiskos P, Sandilos P. Monte Carlo dosimetry 71. Desbiens M, D’Amours M, Afsharpour H,
Followill D, Howell R. Experimental valida- of a new 192Ir high dose rate brachytherapy Verhaegen F, Lavallée MC, Thibault I, et al.
tion of deterministic Acuros XB algorithm for source. Med Phys 2000; 27: 2521–7. Monte Carlo dosimetry of high dose rate
IMRT and VMAT dose calculations with the 64. Anagnostopoulos G, Baltas D, Pantelis E, gynecologic interstitial brachytherapy. Radio-
Radiological Physics Center’s head and neck Papagiannis P, Sakelliou L. The effect of ther Oncol 2013; 109: 425–9. doi: 10.1016/j.
phantom. Med Phys 2012; 39: 2193–202. patient inhomogeneities in oesophageal 192Ir radonc.2013.09.010
54. International Commision on Radiological HDR brachytherapy: a Monte Carlo and 72. Siebert FA, Wolf S, Kóvacs G. Head and neck
Protection. Reference Man: Anatomical analytical dosimetry study. Phys Med Biol (192)Ir HDR-brachytherapy dosimetry using
Physiological and Metabolic Characteristics. 2004; 49: 2675–85. a grid-based Boltzmann solver. J Contemp
ICRP Publication 23. Oxford, UK: Pergamon 65. Enger SA, Ahnesjö A, Verhaegen F, Beaulieu Brachytherapy 2013; 5: 232–5. doi: 10.5114/
Press; 1975. L. Dose to tissue medium or water cavities as jcb.2013.39444

20 of 20 birpublications.org/bjr Br J Radiol;87:20140163