Professional Documents
Culture Documents
Original article
Monte Carlo-based analytical model for small and variable fields delivered
by TomoTherapy
Edmond Sterpin a,c,*, Brian T. Hundertmark c, Thomas R. Mackie b,c, Weiguo Lu b, Gustavo H. Olivera b,c,
Stefaan Vynckier a
a
Université Catholique de Louvain, Department of Radiotherapy and Oncology, Brussels, Belgium; b TomoTherapy Inc., Madison, WI, USA; c Department of Medical Physics,
University of Wisconsin, Madison, WI, USA
a r t i c l e i n f o a b s t r a c t
Article history: Background and purpose: Extend to very small fields the validity of a Monte Carlo (MC) based model of Tomo-
Received 19 October 2009 Therapy called TomoPen for future implementation of the dynamic jaws feature for helical TomoTherapy.
Received in revised form 15 December 2009 Materials and methods: First, the modelling of the electron source was revisited using a new method to mea-
Accepted 20 December 2009
sure source obscuration for very small fields (<1 cm). The method consisted in MC simulations simulations
Available online xxxx
and measurements of the central dose in a water phantom for a 10 cm FW field scanned to deliver a
10 10 cm2 fluence. FW, the longitudinal field width, was varied from 0.4 to 5 cm. The second part of the work
Keywords:
consisted of adapting TomoPen to account for any configuration of the jaws in a fast and efficient way by using
Monte Carlo simulation
TomoTherapy
routinely only the phase-space file of the largest field (5 cm) and interpolated analytical information of phase-
Small fields space files of smaller field widths.
Results: For the electron source fine tuning, it was shown that the best results were obtained for a 1.1 mm
wide spot. Our single phase-space method showed no significant differences compared to MC simulations
of various field widths even though only longitudinal intensity and angular analytical functions were applied
to the 5 cm phase-space.
Conclusion: The designed model is able to simulate all jaw openings from the 5 cm field phase-space file by
applying a bi-dimensional analytical function accounting for the fluence and the angular distribution in the
longitudinal direction.
Ó 2009 Elsevier Ireland Ltd. All rights reserved. Radiotherapy and Oncology xxx (2009) xxx–xxx
Hi-ArtTM TomoTherapy allows high modulation in the trans- in tissue [3]. However, for very small fields (<1 cm), the validity
verse plane by means of the simultaneous movement of the gantry of MC models are strongly dependent on the validity of the hypoth-
and a binary multileaf collimator (MLC) [1]. The longitudinal reso- esis made on the electron source model. Output factors diminish
lution is controlled by the size of the field in this direction, which is rapidly because of source occlusion effects [4] and therefore, the
limited in the present configuration of the TomoTherapy machine accuracy of the source model is of primary importance [5–7].
to three values: 1, 2.5 and 5 cm. However, more freedom could Two previously published papers describe extensively the diffi-
be given to the user by increasing the number of available field culty for MC models to accurately predict output factors for small
sizes, or better, by allowing the jaws to move dynamically during fields without a measurement specific and sensitive to the electron
irradiation. Doing so would sharpen the treatment penumbra in source model [8,9]. In the present study, however, another method,
the inferior–superior direction and speed up TomoTherapy deliv- designed for the TomoTherapy machine was chosen to quantify the
ery by two to three times over the current implementation [2]. effect of source occlusion.
Arbitrary jaw openings involve new challenges in dose compu- Following confirmation that electron source was accurately
tation for TomoTherapy. Efficient and realistic modeling of the modeled, the second objective of the work was to extend the scope
jaws may be required, especially for the smaller fields, which of validity of a previously published MC model called TomoPen
may be achieved by models based on Monte Carlo (MC) simula- [10–12] to all possible jaw field widths up to 5 cm, without modi-
tions. Indeed, those are widely recognized as the most accurate fying its general structure and maintaining efficiency. The model
method available to model treatment delivery and compute dose developed here was also used as an analysis and a development
tool during the implementation of variable jaw openings in the
TomoTherapy treatment planning system (TPS). Since symmetric
* Corresponding author. Address: Université Catholique de Louvain, Department
jaw movement is sufficient to gain the advantages of dynamic
of Radiotherapy and Oncology, St-Luc University Hospital, 10 av. Hippocrate, 1200
10 Brussels, Belgium.
jaws, only symmetric fields are considered in this paper (Private
E-mail address: esterpin@yahoo.fr (E. Sterpin). communication (Gustavo Olivera, January 2009)).
0167-8140/$ - see front matter Ó 2009 Elsevier Ireland Ltd. All rights reserved.
doi:10.1016/j.radonc.2009.12.018
Please cite this article in press as: Sterpin E et al. Monte Carlo-based analytical model for small and variable fields delivered by TomoTherapy. Radiother
Oncol (2010), doi:10.1016/j.radonc.2009.12.018
ARTICLE IN PRESS
To account for all jaw openings, the most intuitive method is to the phantom, where FW is the longitudinal field width. The ma-
simulate the treatment head according to the physical position of chine used for the experiment was located at Radiotherapy Clinic
the jaws [13]. This technique has the disadvantage of being time of the University of Wisconsin Hospital, Madison. The values of
consuming and it is not satisfying the various requirements for FW ranged from 0.4 to 5 cm. For a point source and neglecting dif-
the MC model in our case: ferences in phantom scattering, the dose would be proportional to
FW, and so the ratio of D/FW and D/5 cm would equal unity. How-
(1) The model has to be compatible with the structure of the ever, because of source occlusion, it was expected that this ratio
existing MC code TomoPen, which is phase-space file (PSF) would decrease significantly for small fields. The method was pro-
based. ven to be robust, i.e. it was detector independent and therefore it
(2) In the TomoTherapy convolution/superposition (C/S) [14], could be used for testing the accuracy of the source model for
the longitudinal field width and shape are determined by any kind of dose computation engine, including MC-based models.
measured longitudinal intensity measurements which cor- The MC calculations consisted of a direct simulation of the
respond to a precise and reproducible value of the jaw set- experimental setup. PSF calculations were performed for various
tings. However, the physical aperture of the jaws itself is field sizes and electron source spot sizes using MC HAMMER
hard to measure. The MC model must be able to handle [10]. The resulting PSF were used thereafter in TomoPen. It is
the same data, which make it portable from one machine worth mentioning that equivalent results were obtained by simu-
to another without additional tuning. lating the actual experiment and by convolving a longitudinal dose
(3) The model must be as fast as possible, rely on a minimum of profile computed for a static field with a unit-high step function
precalculated data with no or a negligible cost in accuracy. over 10 cm (and zero elsewhere).
Please cite this article in press as: Sterpin E et al. Monte Carlo-based analytical model for small and variable fields delivered by TomoTherapy. Radiother
Oncol (2010), doi:10.1016/j.radonc.2009.12.018
ARTICLE IN PRESS
u2 þ v 2 þ w2 ¼ 1 ð3Þ
then only the behavior of two direction cosines must be analyzed. In
this case, u and w were chosen. The reasons of this choice will ap-
pear later. Therefore, let us assume that the function PFS can be sep-
arated into two functions PRFS, the phase space for a Reference Field
Size (RFS) which contains most of the information for any field size,
and FFS which corrects the weight of each particle to account for the
differences from one field size to another:
PFS ðE; x; y; z; u; v ; w; WGHT; NSHIÞ
¼ P RFS ðE; x; y; z; u; v ; w; WGHT F FS ðE; x; z; u; w; WGHTÞ; NSHIÞ
ð4Þ
It will be shown that the dependence of E, WGHT, x and u with
field size can actually be neglected and that the problem may be
reduced to
PFS ðE; x; y; z; u; v ; w; WGHT; NSHIÞ
¼ P RFS ðE; x; y; z; u; v ; w; WGHT RFS ðz; wÞ; NSHIÞ ð5Þ
where RFS is a ratio accounting for the variation of the energy flu- Fig. 2. Examples of JIF (a) and JAF (b) ratios calculated with respect to the 5 cm
ence of particles and their angular distribution along the longitudi- field.
nal direction.
Simulating an arbitrary field size from the 5 cm PSF An example of JIF ratios are shown in Fig. 2(a). JIF and JAF meth-
Two methods were employed here. The first method, called the ods are strictly equivalent provided that the jaws do not contribute
Jaw Angular Function (JAF) method, consisted of multiplying the to scattered radiation and that the photon source is a point source.
weight of each particle of the reference 5 cm field PSF by a precal- In the two methods, the ratios were calculated at the isocenter
culated ratio RFS at a given SSD: plane (source–surface distance (SSD) of 85 cm). In the original
TomoPen simulation, the PSFs were recorded 45 cm above the iso-
IFS ðzÞ aFS ðz; wÞ
RJAF
FS;SSD ðz; wÞ ¼ ð6Þ center plane (SSD = 40 cm). To apply the JAF and JIF methods, the
I5 cm ðzÞ a5 cm ðz; wÞ SSD
PSFs were therefore recorded at the isocenter plane. During the
where I can be either measured or calculated by integrating the simulation, the weight of each particle is multiplied by the ratio
energy times the weight of each photon with a given z in a given RFS according to the method chosen and then transported to the
PSF and dividing by the total number of primary showers. The same plane as the original TomoPen simulation before the simula-
PSF is binned along the z direction, with a resolution of around tion resumes as described in Sterpin et al. [12]. The reason why the
0.8 mm. The number of bins is limited to 128, i.e. I is defined isocenter plane was chosen is because measurements are likely
throughout 10 cm, which is large enough to encompass the vast performed at that plane and are used as primary information for
majority of the photons. Outside the region of interest, the ratio tuning the C/S algorithm. Moreover, dose computed using JAF
of I is assumed equal to one. a is also defined the same way, but and JIF will be more accurate to the plane where the ratios were
over the variables z and w and is only computable from MC simu- calculated. To minimize deviations from pure MC simulations,
lations. The number of bins is also 128 with a resolution of 0.002 positioning the PSF plane at the isocenter, the symmetry center
for w, which is justified by the narrowness of a distributions. The of the system, is the best choice.
region of interest is large enough to encompass all the values of For small fields, the method has the disadvantage of producing
a. JAF ratios for the 1 cm field with respect to the 5 cm field is a lot of particles with very small weights outside the field. To save
shown in Fig. 2(b). It is worth mentioning that the ratio of a is the time simulating particles with very small weight, Russian rou-
set to conserve energy. lette is applied with a moderate splitting factor (20) to avoid dis-
The JAF method is designed to use I ratios computed from mea- tortions due to heavy weighted particles.
surement data and interpolated a distributions determined from It is worth emphasizing the fact that the JIF ratio accounts for
PSF calculated for several jaw settings from 0 to 5 cm. most of the modifications while the ratios of a account for modifi-
To analyze the effect of removing the dependency of angular cations regarding the divergence of the beam with field size, which
distribution with field size, which is the case in a C/S algorithm, is a second-order effect.
the second method, called the Jaw Intensity Function (JIF) assumed
the ratio of a equals one and therefore RFS becomes: Testing the JAF and JIF methods
Two main sets of tests were performed:
IFS ðzÞ
RJIF
FS;SSD ðzÞ ¼ ð7Þ
I5 cm ðzÞ SSD
(1) Independence of the shape of E and x distributions. This was
verified by comparing spectral, energy fluence and dose dis-
We can also write tributions of the PSF at various positions along the longitudi-
nal direction. The invariance of the distribution of WGHT is
aFS ðz; wÞ
RJAF
FS;SSD ðz; wÞ ¼ RJIF
FS;SSD ðzÞ ð8Þ implicitly verified since it is taken into account in each
a5 cm ðz; wÞ SSD scored quantity.
Please cite this article in press as: Sterpin E et al. Monte Carlo-based analytical model for small and variable fields delivered by TomoTherapy. Radiother
Oncol (2010), doi:10.1016/j.radonc.2009.12.018
ARTICLE IN PRESS
For the sake of conciseness, the results of the first set of test are
not shown in this paper. Very good agreement was achieved for en-
ergy spectrum, which was expected according to Jeraj et al. [20]
and transverse energy fluence, within statistical uncertainties
(1%, 1 r), whatever the depth and the longitudinal position of the
acquisitions. The dose distributions in the phantom provide an
indirect verification of those results.
In the second set, two cases had to be distinguished. First JIF, JAF
and MC could be compared when I and a distributions were obtained
directly from the PSF for which dose distributions were calculated. In
this case, no interpolation was required. The second case analyzed
occurred when I and a of the considered field were obtained from
interpolation of I and a distributions computed for the closest corre-
sponding PSFs. For instance, a 0.7 cm field was calculated with a pure
MC approach but I and a distributions were interpolated between 0.6
and 0.8 cm PSFs. This test was necessary to ensure that interpolation
can be performed without a significant loss of accuracy. The only
parameter needed for the interpolation was the FWHM of the energy
fluence at the isocenter. Linear interpolation was used.
Fig. 4. Transverse dose profiles (a) and depth dose (b) comparison between the JAF
method and pure MC computations the 1 cm field at several depths.
Please cite this article in press as: Sterpin E et al. Monte Carlo-based analytical model for small and variable fields delivered by TomoTherapy. Radiother
Oncol (2010), doi:10.1016/j.radonc.2009.12.018
ARTICLE IN PRESS
Fig. 7. Longitudinal dose profiles comparison between JAF method and pure MC
computations for the 0.7 cm field at several depths (SSD = 75 cm). In this case, I and
a distributions were obtained from interpolations from I and a distributions of
adjacent PSFs (0.6 and 0.8 cm fields).
of the JAF method for an arbitrary field opening will rely directly
on the quality of the interpolation of the (z, w) distributions be-
tween the closest PSFs. If I(z) is given by measurements or interpo-
lated from a measurement database, then only a (z, w)
distributions have to be interpolated from PSF data. a is a sec-
ond-order correction, therefore the quality of the interpolation
Fig. 5. Longitudinal dose profiles comparison between the JAF (a) and JIF (b) on a will not interfere much with the accuracy of the results. How-
methods and pure MC computations for the 1 cm field at several depths
ever, the accuracy of I(z) is of primary importance and therefore it
(SSD = 75 cm).
is worth building a precise set of PSFs to minimize the errors due to
interpolations in the case of calculated I(z). Since sparse PSFs are
Discussion allowed, this is a relatively straightforward process.
One of the main objectives of the JAF and JIF methods was of
The combination of an accurate electron spot size model and a course efficiency. Calculation time between a 1 cm field calculated
fast and efficient technique to simulate any jaw opening is a prere- from a MC PSF and a JAF PSF were compared to evaluate the poten-
quisite before extending the TomoPen model to treatments involv- tial loss in speed of the JAF method. The main sources of loss of effi-
ing variable jaw openings. Fig. 3 shows a clear effect of source ciency are the fact that more particle phase space data is being read
obscuration for sizes smaller than 1.5 cm. Since the size of the pho- than necessary. For a fully open field, the loss of efficiency is
ton source is linked to the size of the electron spot [21], the exper- around 25%, which is not prohibitive considering the speed of Tom-
iment illustrates a direct effect of the accuracy of the electron spot oPen [17]. The overall efficiency could yet be improved by allowing
size. According to Scott et al. [8,9], the sensitivity of output factors TomoPen to select a PSF of another field size. For instance if the
for small fields to source occlusion is much higher than the sensi- 5 cm and 2.5 cm PSF are used altogether in the system, TomoPen
tivity of penumbras measurements, which depends on many other could select the 2.5 cm for field smaller than 2.5 cm PSF and the
parameters, like detector and MC voxel sizes. The results presented 5 cm PSF for the others. This will improve the efficiency by a factor
in our study were compatible with the observations made in Scott of 2. Another solution would be to sample the distributions of E, x,
et al. [8,9] especially if one compares Fig. 10 in Ref. [8] with Fig. 3 in z, u, v, w and WEIGHT according to the 5 cm PSF and a (z, w) of the
the present paper. considered field. The efficiency issue will be handled during the
For the modeling of the jaws, results shown in Figs. 4–7 demon- implementation of the dynamic jaw feature into TomoPen and will
strate the equivalence of JAF and MC but differences were found be the subject of another paper.
between JIF and MC, for an extreme SSD. In general, the accuracy
Conclusions
Please cite this article in press as: Sterpin E et al. Monte Carlo-based analytical model for small and variable fields delivered by TomoTherapy. Radiother
Oncol (2010), doi:10.1016/j.radonc.2009.12.018
ARTICLE IN PRESS
Acknowledgements factors and exploring miniphantom design for small-field measurements. Med
Phys 2009;36:3132–44.
[10] Sterpin E, Tomsej M, Cravens R, et al. Monte Carlo simulation of the
E. Sterpin is a research fellow of the Belgian ‘Fonds National Tomotherapy treatment unit in the static mode using MC HAMMER a Monte
pour la Recherche Scientifique’ (Charge de recherches du F.R.S – Carlo tool dedicated to Tomotherapy. J Phys: Confer Ser 2007;74:021019.
[11] Sterpin E, Salvat F, Olivera GH, Vynckier S. Analytical model of the binary
FNRS FC 73512). This work was also supported by the University
multileaf collimator of Tomotherapy for Monte Carlo simulations. J Phys:
of Wisconsin, Department of Medical Physics and TomoTherapy Confer Ser 2008;102:012022.
Inc. (Madison WI). T.R. Mackie, W. Lu and G.H. Olivera have finan- [12] Sterpin E, Salvat F, Cravens R, Ruchala K, Olivera GH, Vynckier S. Monte Carlo
simulation of helical Tomotherapy with PENELOPE. Phys Med Biol
cial interest in TomoTherapy Inc.
2008;53:2161–80.
[13] Verhaegen F, Liu HH. Incorporating dynamic collimator motion in Monte Carlo
References simulations: an application in modelling a dynamic wedge. Phys Med Biol
2001;46:287–96.
[1] Mackie TR, Holmes TW, Swerdloff S, et al. Tomotherapy: a new concept in the [14] Lu W, Olivera GH, Chen M, Reckwerdt PJ, Mackie TR. Accurate convolution/
delivery of dynamic conformal radiotherapy. Med Phys 1995;20:1709–19. superposition for mutli-resolution dose calculation using cumulative
[2] Sterzing F, Uhl M, Sroka-Perez G, et al. Dynamic jaws and dynamic couch in tabulated kernels. Phys Med Biol 2005;50:655–80.
helical tomotherapy. Int J Radiat Oncol Biol Phys 2009;75:S719. [15] Deng J, Jiang SB, Kapur A, Li J, Pawlicki T, Ma C-M. Photon beam
[3] Reynaert N, van der Marck SC, Schaart DR, et al. Monte Carlo treatment characterization and modeling for Monte Carlo treatment planning. Phys
planning for photon and electron beams. Radiat Phys Chem 2007;76:643–86. Med Biol 2000;45:411–27.
[4] Zhu TC, Bjarngard BE. The head-scatter factor for small field sizes. Med Phys [16] Fippel M, Haryanto F, Dohm O, Nusslin F. A virtual photon energy fluence
1994;21:65–8. model for Monte Carlo dose calculation. Med Phys 2003;30:301–11.
[5] McKerracher C, Thwaiters TW. Phantom scatter factors for small MV photon [17] Sterpin E, Salvat F, Olivera GH, Vynckier S. Monte Carlo evaluation of the
fields. Radiother Oncol 2007;86:272–5. convolution/superposition algorithm of Hi-ArtTM Tomotherapy in
[6] McKerracher C, Thwaiters TW. Head scatter factors for small MV photon fields. heterogeneous phantoms and clinical cases. Med Phys 2009;36:1566–75.
Part I. A comparison of phantom types and methodologies. Radiother Oncol [18] Fan J, Paskalev K, Wang, et al. Determination of output factors for stereotactic
2007;85:277–85. radiosurgery beams. Med Phys 2009;36:5292–300.
[7] McKerracher C, Thwaiters TW. Head scatter factors for small MV photon fields. [19] Salvat F, Fernandez-Varea JM, Sempau J. PENELOPE-2008: a code system for
Part II. The effects of source size and detector. Radiother Oncol Monte Carlo simulation of electron and photon transport. OECD Nuclear
2007;85:286–91. Energy Agency Issy-les-Moulineaux France; 2008.
[8] Scott AJD, Nahum AE, Fenwick JD. Using a Monte Carlo model to predict [20] Jeraj R, Mackie TR, Balog J, Olivera G, Pearson D, Kapatoes J, et al. Radiation
dosimetric properties of small radiotherapy photon fields. Med Phys Characteristics of helical tomotherapy. Med Phys 2004;31:396–404.
2008;35:4671–84. [21] Wang LLW, Leszczynski K. Estimation of the focal spot size and shape for a
[9] Scott AJD, Nahum AE, Fenwick JD. Monte Carlo modeling of small photon medical linear accelerator by Monte Carlo simulation. Med Phys
fields: quantifying the impact of focal spot size on source occlusion and output 2007;34:485–8.
Please cite this article in press as: Sterpin E et al. Monte Carlo-based analytical model for small and variable fields delivered by TomoTherapy. Radiother
Oncol (2010), doi:10.1016/j.radonc.2009.12.018