The importance of accurate linear accelerator head modelling for IMRT Monte Carlo

Transcription

1 Home Search Collections Journals About Contact us My IOPscience The importance of accurate linear accelerator head modelling for IMRT Monte Carlo calculations This article has been downloaded from IOPscience. Please scroll down to see the full text article. 5 Phys. Med. Biol. 831 ( View the table of contents for this issue, or go to the journal homepage for more Download details: IP Address: The article was downloaded on 15// at 15:31 Please note that terms and conditions apply.

2 INSTITUTE OF PHYSICS PUBLISHING Phys. Med. Biol. (5) PHYSICS IN MEDICINE AND BIOLOGY doi:.88/ //5/8 The importance of accurate linear accelerator head modelling for IMRT Monte Carlo calculations N Reynaert 1, M Coghe 2, B De Smedt 1, L Paelinck 2, B Vanderstraeten 1,2, W De Gersem 2, B Van Duyse 2, C De Wagter 2, W De Neve 2 and H Thierens 1 1 Department of Medical Physics, Ghent University, Proeftuinstraat 86, B-9 Gent, Belgium 2 Division of Radiotherapy, Ghent University Hospital, De Pintelaan 185, B-9 Gent, Belgium Received 31 March 4, in final form 17 August 4 Published 17 February 5 Online at stacks.iop.org/pmb//831 Abstract Two Monte Carlo dose engines for radiotherapy treatment planning, namely a beta release of Peregrine and MCDE (Monte Carlo dose engine), were compared with Helax-TMS (collapsed cone superposition convolution) for a head and neck patient for the Elekta SLi plus linear accelerator. Deviations between the beta release of Peregrine and MCDE up to % were obtained in the dose volume histogram of the optical chiasm. It was illustrated that the differences are not caused by the particle transport in the patient, but by the modelling of the Elekta SLi plus accelerator head and more specifically the multileaf collimator (MLC). In MCDE two MLC modules (MLCQ and MLCE) were introduced to study the influence of the tongue-and-groove geometry, leaf bank tilt and leakage on the actual dose volume histograms. Differences in integral dose in the optical chiasm up to 3% between the two modules have been obtained. For single small offset beams though the FWHM of lateral profiles obtained with MLCE can differ by more than 1.5 mm from profiles obtained with MLCQ. Therefore, and because the recent version of MLCE is as fast as MLCQ, we advise to use MLCE for modelling the Elekta MLC. Nevertheless there still remains a large difference (up to %) between Peregrine and MCDE. By studying small offset beams we have shown that the profiles obtained with Peregrine are shifted, too wide and too flat compared with MCDE and phantom measurements. The overestimated integral doses for small beam segments explain the deviations observed in the dose volume histograms. The Helax-TMS results are in better agreement with MCDE, although deviations exceeding 5% have been observed in the optical chiasm. Monte Carlo dose deviations of more than % as found with Peregrine are unacceptable as an influence on the clinical outcome is possible and as the purpose of Monte Carlo treatment planning is to obtain an accuracy of 2%. We would like to emphasize that only the Elekta MLC has been tested in this work, so it is certainly /5/831+16$3. 5 IOP Publishing Ltd Printed in the UK 831

3 832 N Reynaert et al possible that alpha releases of Peregrine provide more accurate results for other accelerators. 1. Introduction The treatment head of linear accelerators has been modelled with Monte Carlo for several decades. An interesting and extensive overview has recently been provided by Verhaegen and Seuntjens (3). The importance of an accurate modelling of the multileaf collimator (MLC) has been recognized by several investigators. The importance of the curvature of the leaf tips was demonstrated by De Vlamynck et al (1999), which led to the introduction of MLCQ into the BEAMnrc software (Rogers et al 2). A little later VARMLC was introduced in the BEAMnrc software. In VARMLC the leaf sides are modelled in detail as well as the curved leaf tip of the Varian MLC. The leaf sides need to be modelled in detail to obtain accurate results concerning leakage, transmission and the so-called tongue-and-groove effect. Recently MLCE, a component module for the Elekta SLi plus MLC (Van de Walle et al 3), was developed. Next to the curved leaf tip and the step in the leaf side, a tilt in the MLC bank was included. The model is able to reproduce leakage and transmission very accurately and demonstrates that MLCQ only predicts 44% of the total transmission. Next to that a spectral hardening due to inter-leaf leakage was demonstrated for 6 MV. Also in Monte Carlo systems designated for radiotherapy treatment planning, where speed is the first goal, the effects of a detailed modelling of the MLC were investigated. Deng et al (1) state that the tongueand-groove effect is negligible for an intensity modulated radiotherapy (IMRT) treatment with more than four gantry angles, especially taking into account patient set-up uncertainty. For an individual IMRT field the tongue-and-groove design could amount to %, as illustrated in the work of Deng et al. In the fast MLC module, developed by Keall et al (1) and Siebers et al (2), the Varian MLC leaves are modelled in detail and the beam hardening effect of the MLC at 6 MV is taken into account. In current Monte Carlo treatment planning systems several procedures to speed up the simulation are used. Next to variance reduction techniques in the phantom, special attention has been paid to the transport through the accelerator head, which has led to the development of the virtual source model (Ma and Rogers 1995) and several fast transport techniques in the MLC. The purpose of this paper is to determine the importance of an accurate modelling of the MLC for IMRT treatment planning in the case of the Elekta SLi plus linear accelerator. In a previous paper of our group (Van de Walle et al 3) attention was only paid to leakage and transmission for a totally closed MLC and the tongue-and-groove design of a non-realistic beam set-up, without evaluating the consequences for realistic beams used in a clinical context. Dose volume histograms of a head and neck (Sino nasal) cancer patient treated with IMRT obtained with a beta release of Peregrine (Hartmann Siantar et al 1) are compared to results obtained with our recently introduced dose engine MCDE (Reynaert et al 4). Treatment consists of resection of tumour creating large air cavities. Close-shave resection margins are likely sites of sub-clinical disease. As a result, regions nearby airsoft tissue interfaces are clinical target volume and accurate dose computation is needed for dose prescription, optimization and reporting. Accurate dose computation is also needed to secure organs at risk sparing, because of the steep dose-toxicity relationships for structures like lachrymal glands, retina, optic pathways, brain and brainstem. For IMRT applications, small and irregular beam apertures are shaped to create dose gradients nearby organs at risk.

4 Accurate linac head modelling for IMRT Monte Carlo calculations 833 Leakage and transmission of leaves and jaws must be known to compute appropriate shielding of small organs at risk, like optic nerves and chiasm. With precise modelling of the head and collimator, Monte Carlo techniques allow us to address all these challenges regarding the dose computation for the treatment of Sino nasal cancer. In MCDE no variance reduction techniques or virtual source model are used and the MLC is handled with full Monte Carlo. Consequently this code is slower than other Monte Carlo systems introduced for treatment planning, but it can be used as a verification tool for results obtained with other Monte Carlo systems as Peregrine. To investigate the influence of the MLC modelling, two MLC modules, namely MLCE and MLCQ are implemented in MCDE. The comparison between MCDE MLCE, MCDE MLCQ and Peregrine is also performed for small offset fields to obtain an explanation for the observed differences in the dose volume histograms for the studied clinical case. 2. Materials and methods 2.1. MCDE The 6 MV beam of the Elekta SLi plus linear accelerator was modelled in detail with BEAMnrc as described in previous papers (De Vlamynck et al 1999, Van de Walle et al 3, Reynaert et al 4). A phase-space file below the mirror is determined and used as input in MCDE. Two MCDE codes were built, one with the MLCQ module and one with MLCE, which includes an accurate modelling of the leaf side and a tilt in the MLC leaf bank. In both systems the particle transport is handled with full Monte Carlo in the MLC. In the accelerator head the electron cut off energy (ECUT) is.7 MeV while this parameter has a value of.561 MeV in the phantom/patient. No variance reduction techniques are used, except the recycling of particles from the original phase-space file and particles entering the phantom, to increase efficiency (up to cycles). The original phase-space file contains 58 6 particles per node (18 nodes) Peregrine The department of radiotherapy of the Ghent University Hospital has an ongoing collaboration with Nomos Corporation (Sewickley) to make Peregrine work adequately with Elekta linear accelerators for IMRT. This work is still in progress. Therefore, we speak of a beta release of Peregrine as this version (Peregrine version 1., PRDCT.9 BETA) is not released for clinical use Clinical case study A patient with advanced (pt3) adenocarcinoma of the ethmoid sinuses was treated with IMRT using up to 6 MV beam segments (nine beam directions, non-coplanar), to a total dose of 7 Gy. For a more extensive description of the treatment the reader is referred to Claus et al (1). A set of up to 1 CT slices with a variable slice thickness ranging from 2 to 5 mm was converted to a Monte Carlo phantom with an in-plane resolution of 2 mm The following materials were defined: air, adipose tissue, muscle, soft bone, cortical bone. The density was scaled linearly with the Hounsfield number. The skin contour was used to delimit the voxels that need to be taken into account for scoring dose information. A scoring grid consisting of partially overlapping spherical voxels with a radius of 2 mm was superimposed on the

5 834 N Reynaert et al Figure 1. Delineation of PTV and optical chiasm superimposed on the CT images (inverse grey scale). The figure on the left is a CT slice through the isocentre illustrating the large air cavities within the PTV, while the picture on the right illustrates the short distance between the contour of the small optical chiasm and the PTV contour. (This figure is in colour only in the electronic version) geometrical grid. The obtained dose distribution was converted to dose to water (Siebers et al ) to enable a comparison with results obtained with the collapsed cone convolution (CCC) algorithm of HELAX-TMS 6.1.A (Nucletron, Veenedal, The Netherlands) and results obtained with the beta release of Peregrine (NOMOS Corporation, Sewickley, PA, USA). In Peregrine the same scoring grid is used and all tissues (except for air and cortical bone) are modelled as water. To link the electron density with the Hounsfield number extra bins are defined. For the evaluation of the three systems, dose volume histograms (DVHs) in the planning target volume (PTV) and the optical chiasm, which is an important critical organ, were determined. To have a better idea of the location and the dimensions of these anatomical structures, the reader is referred to figure 1. From this figure it is clear that the optical chiasm is a relatively small anatomical structure. Therefore, dose gradients (e.g. beam penumbra) in that region may have a large influence on the DVH. This is illustrated in figure 2 where several beams only partially overlap the optical chiasm. To ensure that eventual differences between the codes are not caused by a different positioning of the scoring voxels, the scoring grid is in all three systems identical. The influence of a small shift (1 mm) in the position of the score voxels in the three directions on the DVH in the optical chiasm was investigated and was found to be non-existent. For the determination of the DVHs from the dose distributions, the DVH software of PLUNC (PLan UNC, Cullip 3) was used, by uniformly selecting points in the anatomical structures of interest (PTV and optical chiasm) with an inter spacing of.5 cm (1 bin per Gy). In this way, identical points were used for the three systems, although the influence of the point selection method was negligible. In Peregrine and MCDE calculations were also performed assuming water in all CT voxels within the skin contour, as a first order separation of the influence of the particle transport in the patient and the modelling of the accelerator head.

6 Accurate linac head modelling for IMRT Monte Carlo calculations 835 Figure 2. Illustration of the partial overlap of beam segments and optical chiasm. The optical chiasm in different CT planes is drawn grey. The solid lines represent the MLC and JAW settings (all dimensions are in cm) Standard offset beam segments In order to find the explanation for any differences between MCDE and Peregrine, the two codes were used to determine lateral dose profiles for small beam segments. A 2 cm 2cm beam segment was studied for different offsets. Both jaw pairs were opened totally and thus do not influence the beam collimation. Even for IMRT the studied beam segment is relatively small, but it provides interesting information concerning the modelling of the leaves, as we only want to find an explanation for observed deviations. For the rest of this paper the direction of motion of the leaves when the collimator is at zero degrees will be defined as the X-direction while the Y-direction is the gun-target (GT) direction. For the studied beam segments X- and Y- profiles are computed with MCDE with both MLC models (MLCE and MLCQ) and with Peregrine. The profiles in the X-direction enable a check of the modelling of the curvature of

7 836 N Reynaert et al the leaf tips, while the Y-profile provides information concerning the leaf sides and the leaf bank tilt. As these 2 cm 2 cm beam segments cannot be generated on the Elekta SLi plus linear accelerator without using the collimator jaws (touching leafs from opponent bank cannot pass each other), open slits of 2 cm width were modelled to enable a comparison of both systems with measurements. All measurements were performed in an MP3 water phantom (PTW-Freiburg) with a diamond detector (PTW-Freiburg, type 3) at isocentre with 6 MV photons. The diamond detector was placed at isocentre (SSD = 95. cm, depth equal to 5. cm) and oriented for maximal spatial resolution (detector axis perpendicular to the beam axis) and a dose-rate correction is applied. A 2 cm cm (X cm Y cm) beam was used to determine the X-profile and a cm 2 cm beam for the Y-profile. For both beams the offset was varied in both directions. Especially in the Y-direction the offset is expected to play an important role due to the leaf bank tilt and tongue-and-groove design Individual clinical beam segments For a further evaluation of differences observed between MCDE and Peregrine, an individual dose map for every beam segment was determined. In MCDE this is automatically the case, while in Peregrine different calculations must be performed for each segment. With MCDE we first determined the beam segments that have an important contribution to the dose in the optical chiasm and special attention was paid to segments that partially overlap with the optical chiasm. 3. Results and discussion 3.1. Clinical case study MCDE (MLCQ) versus Peregrine versus Helax-TMS. The statistical uncertainty was below 2% with respect to the dose maximum in all voxels for all clinical cases studied. A comparison of dose volume histograms (DVHs) in the planning target volume (PTV) and optical chiasm (OC) is illustrated in figure 3 for results obtained with Peregrine, Helax-TMS (CCC) and MCDE. For this comparison the MLC is modelled with MLCQ in MCDE and the obtained dose to medium is converted to dose to water to enable a comparison with Helax-TMS (see Siebers et al ()). This stopping power conversion shifts the DVH in the PTV to the right with 3%, due to the large air cavities in this volume. In the optical chiasm the effect of the conversion is much smaller. In the PTV the results are in good agreement. In Peregrine (dose also converted to dose to water) the DVH in the PTV dose is almost identical to that of MCDE. In the optical chiasm, however, deviations are large. Especially the fact that both Monte Carlo codes deliver results that are on the opposite side of the CCC results is at first sight surprising. The integral dose obtained with Peregrine is 5% higher than that of MCDE. The MLCQ model used in MCDE does not include interleaf leakage or the tongue-and-groove geometry Homogeneous water geometry. To find an explanation of the observed differences the Peregrine and MCDE calculations were repeated with all CT voxels within the skin contour filled with water. This provides the DVHs in the optical chiasm presented in figure 4. It is clear that the observed large differences have not disappeared and are thus not caused by the particle transport in the phantom, as we do not expect that two Monte Carlo codes will deliver different results in a homogeneous water geometry when provided the same beam input (even

8 Accurate linac head modelling for IMRT Monte Carlo calculations Optical Chiasm MCDE (MLCQ) Helax Peregrine Volume (%) Dose (Gy) PTV MCDE (MLCQ) Helax Peregrine Volume (%) Dose (Gy) Figure 3. Comparison of DVHs obtained with Helax-TMS (CCC), MCDE and the beta release of Peregrine in optical chiasm and PTV. pencil beam calculations deliver correct results in a homogeneous water phantom). Therefore, for the rest of this paper attention will be focused on the modelling of the MLC MLCQ versus MLCE. To verify if the difference between results obtained with MCDE and Peregrine is caused by the MLCQ MLC model, the MCDE calculation was performed with the MLCE component module. In this model, next to the tongue-and-groove geometry of the Elekta MLC a leaf bank tilt is included as well. A comparison of the DVHs obtained with MLCE and MLCQ is shown in figure 5. In spite of a positive contribution of leakage and transmission the MLCE results are 2% lower than MLCQ. An explanation of this difference

9 838 N Reynaert et al Volume (%) MCDE (MLCQ) Peregrine MCDE_water phantom Peregrine_water phantom Dose (Gy) Figure 4. Influence of the phantom material on the DVHs in the optical chiasm for MCDE and Peregrine. The results denoted with water phantom in the legend are obtained by filling all CT voxels within the skin contour with water to obtain a homogeneous phantom. 9 8 MLCQ_Optical chiasm MLCE_Optical chiasm MLCQ_PTV MLCE_water 7 Volume (%) Dose (Gy) Figure 5. Comparison of MCDE DVHs obtained with the MLCE and MLCQ model. is given in the following paragraph. This effect increases the difference between Peregrine and MCDE though, which is illustrated in figure 6 where the MCDE dose is systematically multiplied by a factor of 1.7 (to obtain the same integral dose in the optical chiasm) to illustrate that the shape of the DVHs in the optical chiasm obtained with Peregrine and MCDE

10 Accurate linac head modelling for IMRT Monte Carlo calculations Peregrine MCDE (MLCE) +7% 8 7 Volume (%) Dose (Gy) Figure 6. Comparison of MCDE (with MLCE model) with Peregrine. All MCDE doses are scaled with a factor of 1.7 to illustrate the difference between the codes. is identical. For other patients with comparable treatment circumstances differences up to 12% are even noted Offset beam segments Comparison MLCE MLCQ. In a first set of calculations a 2 cm 2 cm beam (leaf-only) with offset 5 cm in both directions is evaluated. X- and Y- profiles are shown in figure 7. The statistical uncertainty in all voxels for all profiles shown in figure 7 is within 2% of the dose maximum. The full width half maximum (FWHM) of the X-profile obtained with MLCE is somewhat larger than that of MLCQ, which is a consequence of interleaf leakage (the profile is obtained in between two leaves). This is illustrated by the no-gap curve which is obtained by filling all interleaf air gaps with tungsten, which leads to comparable results as obtained with the MLCQ model. This is the positive leakage contribution we expected to see in the DVH of the optical chiasm. In the Y-profile the opposite occurs, namely the MLCE profile is 1.5 mm less wide than that of MLCQ. This is partly caused by the tongue-and-groove design as explained in previous work (VandeWalleet al 3) but the main reason is the leaf bank tilt as illustrated by the no-tilt curve. It is also clear that the negative effect in the Y-direction is larger than the positive leakage effect in the X-direction giving rise to an overall decrease of the integral dose by taking into account the real geometry of the leaves. This explains the shift of the DVH to the left in the optical chiasm. A number of small beam segments have a partial overlap in the Y-direction with the optical chiasm. As this is a very small volume the FWHM of a partially overlapping beam segment is extremely important Comparison Peregrine MCDE-measurement. To enable an experimental verification 2cm cm beams were used for the X-profiles and cm 2 cm beams for the Y-profiles (again leaf-only). The measurements are normalized so that the dose maximum equals the Monte Carlo results, while the Monte Carlo results are absolute for monitor units which

11 8 N Reynaert et al 9.E-3 8.E-3 7.E-3 MLCQ MLCE MLCE_nogap 6.E-3 Dose (Gy/MU) / 5.E-3 4.E-3 3.E-3 2.E-3 1.E-3.E X (cm) 9.E-3 8.E-3 7.E-3 MLCQ MLCE MLCE_notilt 6.E-3 Dose (Gy/MU) / 5.E-3 4.E-3 3.E-3 2.E-3 1.E-3.E Y (cm) Figure 7. Lateral profiles obtained with MLCE and MLCQ of 2 cm 2 cm beams at SSD = 95 cm, depth equal to 5 cm. In the X-profile the no gap curve represents results obtained with the gap between the leaves filled with leaf material, to illustrate the effect of interleaf leakage. The no tilt curve in the Y profile is obtained by removing the leaf bank tilt. would lead to a dose of Gy in the isocentre for a cm cm beam at SSD = 95 cm. The obtained results are illustrated in figure 8 where the statistical uncertainty is below 1% with respect to the dose maximum in all voxels. In the Y-direction the Peregrine results are in good agreement with the MCDE results obtained when using the MLCQ module (as can be concluded by comparing figures 7(b) and 8(b)), which demonstrates that Peregrine uses a similar model as MLCQ to model the Elekta MLC. Therefore, tongue-and-groove and leaf bank tilt effects are ignored which leads to Y-profiles with a too large FWHM. The good

12 Accurate linac head modelling for IMRT Monte Carlo calculations MCDE Measurement Peregrine 7 Dose (Gy) Dose (Gy) X (cm) MCDE Measurement Peregrine Y (cm) Figure 8. Comparison of the X-profile (top figure) of a cm 2 cm beam slit and the Y-profile (bottom figure) for a 2 cm cm beam of MCDE, Peregrine and phantom measurements with a diamond detector (SSD = 95 cm, depth equal to 5 cm). agreement between the MCDE MLCE results with the measurements is a confirmation of the tongue-and-groove results described by Van de Walle et al (3). The results in the X-direction are more surprising. As a first observation the Peregrine profile is shifted 1 mm to the left. When defining a negative X-offset of 5 cm the Peregrine profile is shifted to the right and a profile on the central axis exhibits no shift. Next to the shift, the width of the Peregrine profile is too large and the shape is wrong. These last two effects are probably caused by an overestimation of leakage and transmission. As these contributions are not inherently included in the MLC model (see Y-profile), it is difficult to account for it in an accurate way. When

13 842 N Reynaert et al profiles are determined centrally below a leaf (instead of in between two leaves), the width difference between the Peregrine and MCDE profiles decreases, which demonstrates that it is indeed a leakage effect. The shape remains different, although the same scoring volumes are used in both systems. The 1 mm shift is difficult to explain. A wrong leaf-to-source position would also lead to errors in the Y-direction profile, which seems to be correct. Another possible explanation might be that an error is present in the geometry tests of the curved leaf or in the projection of the leaf positions from isocentre to the linac head. Profiles obtained for 2 or 2 fields collimated using only the JAWS (MLC leaves opened) are in good agreement, so it is purely a MLC modelling problem. The small difference between the MCDE curve and the measured results is partially caused by the geometry of the diamond detector, which is positioned to obtain a high geometrical resolution in the X-direction, which leads to a dimension of 3 mminthey-direction. Therefore, the interleaf leakage effect is not measured accurately Effect leakage and leaf bank tilt for clinical case To study the effect of the details of the MLC model different calculations were performed for a head and neck patient. In a first calculation the interleaf regions are filled with tungsten to remove the leakage component. The leaf bank tilt was maintained and due to tongueand-groove the field width in the Y-direction is smaller than that obtained with MLCQ. This is the situation in which a minimal dose is obtained. A direct comparison with the MLCE results provides the interleaf leakage contribution which is limited to.5% for the studied clinical case. From this it is also possible to determine the negative effect on the dose due to tongue-and-groove design and leaf bank tilt. For this a factor of.97 is obtained. To separate these two effects, a calculation was performed with the leaf bank tilt removed, which led to a tilt factor of.98 for the studied case Calculation times A timing study of MLCE and MLCQ was performed. When only the transport through the MLC is taken into account, MLCE is faster than MLCQ by 17%. This is caused by the air regions between the leaves, where the particle transport is much faster. In the situation of the 2cm 2 cm leaf-only beam segments studied in this work the calculation with MLCE is only 6% slower than MLCQ, which is a consequence of the fact that more photons are exiting MLCE (interleaf leakage). Therefore, the time increase is caused by the simple fact that in the case of the MLCE model more particles have to be tracked through the phantom. For larger beams the time difference between MLCQ and MLCE becomes negligible. So, even if the effect of the detailed structure is not that large for the clinical cases, the calculation time lost by using MLCE is negligible (<5% for all studied clinical cases). Two modifications in MLCE are responsible for the relative high speed. First, the geometrical tests of the curved leaf ends are carried out in two instead of three dimensions, by neglecting the small tilt of the individual leaves. Next to that a new macro was introduced to detect when particles travel parallel to a boundary. This leads to a speed increase of a factor 2 compared to the previous MLCE version although identical results are obtained for the clinical cases and also for the 2cm 2 cm offset beam segments Individual clinical beam segments Figure 9 displays the DVHs in the optical chiasm for two individual beam segments. The first beam segment delivers the highest dose contribution to the optical chiasm and has a full

14 Accurate linac head modelling for IMRT Monte Carlo calculations MCDE_MLCQ MCDE_MLCE Peregrine 7 Volume (%) Dose (Gy) 9 8 MCDE_MLCQ MCDE_MLCE Peregrine 7 Volume (%) Dose (Gy) Figure 9. DVH comparison between Peregrine, MCDE (MLCQ and MLCE) for two individual beam segments which have a high contribution to the dose in the optical chiasm. The first beam segment has of full overlap with the optical chiasm while the second segment only partially overlaps. overlap. The lower graph illustrates the second most important segment concerning the dose to the optical chiasm, and the partial overlap explains the two dose levels. From these figures it is clear that Peregrine overestimates the dose in the optical chiasm. This is partly due to the use of the MLCQ module (see MCDE MLCQ results) and partly due to the overestimation of the integral dose in the X-profile (and the shift of this profile). Integrating the surface of the DVH curves and normalizing the obtained surface to that of MCDE MLCE delivers the following quantitative results: for the first segment: Peregrine: 1.3 Gy; MCDE MLCQ: 1.14 Gy. for the second segment: Peregrine: 1.1 Gy; MCDE MLCQ: 1.65 Gy.

15 844 N Reynaert et al This illustrates clearly that the largest deviations between MCDE and Peregrine in the optical chiasm are obtained for beam segments that partially overlap this organ General discussion of the observed deviations Large dose differences (above %) between MCDE and the beta release of Peregrine are observed in the DVHs in the optical chiasm. As the optical chiasm is a small volume and as a large number of small irregular beam segments are used in the described IMRT treatment, we are indeed demanding a lot of the Monte Carlo algorithm. On the other hand, the dose to an important critical organ as the optical chiasm is of clinical importance in the planning process. Due to an overestimation of this dose, it is possible that the planning process will introduce an underdosage of the tumour volume, which might have a clinical relevance. We have found, although not described in this paper, similar results for the spinal cord for a PTV located in the neck region. It seems a general trend that small critical structures will have a partial overlap with small irregular beam segments. In the case of a wrongly modelled beam edge this might lead to severe dose calculation errors, which might have an influence on the clinical outcome. The dose will always be overestimated in critical organs due to too high integral doses for small offset segments. Even in the case of a negligible effect on the clinical outcome, it seems unacceptable that a Monte Carlo code introduces systematic errors of that magnitude as the use of Monte Carlo treatment planning is intended to obtain results within 2%. The CCC algorithm of Helax provides results that are in better agreement with MCDE. This discussion should be put in perspective though note that it is possible that this is purely a modelling problem of the MLC and we did benchmark the code only for the Elekta SLi plus linear accelerator. We believe that introducing a correction in the geometry routines of Peregrine might solve most of the observed problems, certainly the shift of the X-profile. Note that the JAWS, which have been tested in the same way, are modelled correctly. It would indeed be interesting to test the program after the application of this correction, because this could provide a discussion on the impact of the Peregrine-introduced virtual source model. For the moment we are collaborating with NOMOS to solve this problem. Measurements of individual segments of the head and neck treatment would be of value. It is not easy though to build a phantom which is representative for the studied clinical cases for, e.g., film measurements. This is a topic for further investigation. A possibility is to compare individual beam segments of the studied treatment plans. For segments with large deviations between MCDE and Peregrine, as, e.g., the second segment in figure 9, profiles could be measured in a phantom through the optical chiasm and compared with the Monte Carlo results. This could easily be performed in a homogeneous phantom as we have shown that the same deviations are obtained if all CT voxels of the patient geometry are filled with water, proving that there is a problem in the modelling of the beam. On the other hand we do believe that we have proved, with the aid of the small offset beam measurements that it is indeed in the beta release of Peregrine that errors are introduced and that the MCDE system can be used as a reliable benchmarking tool. 4. Conclusions Large dose deviations (up to %) between the Monte Carlo dose engines Peregrine (beta release) and MCDE are observed in the DVHs in the optical chiasm of a head and neck patient treated with IMRT. Filling all CT voxels with water did not solve the problem, from which we concluded that the differences are caused by a different modelling of the Elekta SLi plus accelerator head. Especially the MLC model was investigated. Introducing MLCQ in MCDE,

16 Accurate linac head modelling for IMRT Monte Carlo calculations 845 thus neglecting the tongue-and-groove geometry and the leaf bank tilt, only partly decreased the differences. The MLCE routine is reprogrammed more efficiently and is now as fast as MLCQ for actual treatment planning. Differences between MLCE and MLCQ are below 3% for all clinical cases studied. For small individual beam segments though large differences are observed regarding the width of the profiles. The FWHM of the MLCE profile in the direction perpendicular to the leaf motion is up to % lower than of MLCQ. This is mainly caused by the leaf bank tilt. Therefore, we advise to use MLCE for the Elekta MLC, as this module is as fast as MLCQ. Comparison of MCDE and Peregrine with measurements for small offset beams, have illustrated that integral doses are overestimated in Peregrine and even a shift of the Peregrine profiles is observed, which provides an explanation for the deviations observed for the clinical cases, where small beam segments partially overlap with the small volume of the optical chiasm. Probably an error is present in Peregrine in the projection of the leaf settings from the isocentre to the accelerator head or in the geometry routines, which is currently under investigation in collaboration with NOMOS. Acknowledgments This research work was supported by the Ghent University grant GOA 11 and 11V172. Services provided by the Helax-TMS Software Team of Nucletron are acknowledged. We would like to acknowledge D Sheikh-Bagheri from Nomos for the discussions about the observed deviations. References Claus F et al 1 An implementation strategy for IMRT of ethmoid sinus cancer with bilateral sparing of the optical pathways Int. J. Radiat. Oncol. Biol. Phys Deng J, Pawlicki T, Chen Y, Li J, Jiang S B and Ma C-M 1 The MLC tongue-and-groove effect on IMRT dose distributions Phys. Med. Biol De Vlamynck K, Palmans H, Verhaegen F, De Wagter C, De Neve W and Thierens H 1999 Dose measurements compared with Monte Carlo simulations of narrow 6 MV multileaf collimator shaped photon beams Med. Phys Hartmann Siantar C L et al 1 Description and osimetric verification of the PEREGRINE Monte Carlo dose calculations system for photon beams incident on a water phantom Med. Phys Keall P J, Siebers J V, Arnfield M, Kim J O and Mohan R 1 Monte Carlo dose calculations for dynamic IMRT treatments Phys. Med. Biol Li J S, Pawlicki T, Deng J, Jiang S B, Mok E and Ma C-M Validation of a Monte Carlo dose calculation tool for radiotherapy treatment planning Phys. Med. Biol Ma C-M, Mok E, Kapur A, Pawlicki T, Findley D, Brain S, Forster K and Boyer A L 1999 Clinical implementation of a Monte Carlo treatment planning system Med. Phys Ma C-M and Rogers D W O 1995 Beam characterization: a multiple-source model National Research Council of Canada Report PIRS-9(D), Ottawa, Canada Reynaert N, De Smedt B, Coghe M, Paelinck L, Van Duyse B, De Gersem W, De Wagter C, De Neve W and Thierens H 4 MCDE: a new Monte Carlo dose engine for IMRT with an efficient scoring method Phys. Med. Biol. 49 N235 N241 Rogers D W O, Ma C-M, Walters B, Ding G X, Sheikh-Bagheri D and Zang G 2 Beamnrc User Manual National Research Council of Canada Schach von Wittenau A E, Bergstrom R M and Cox L J Patient-dependent beam-modifier physics in Monte Carlo photon dose calculations Med. Phys Sherouse G W, Thorn J, Novins K, Margolese-Malin J and Mosher C 1989 A portable 3D radiotherzapy design system (abstract) Med. Phys Siebers J V, Keall P J, Kim J O and Mohan R 2 A method for photon beam Monte Carlo multileaf collimator particle transport Phys. Med. Biol

17 846 N Reynaert et al Siebers J V, Keall P J, Nahum A E and Mohan R Converting absorbed dose to medium to absorbed dose to water for Monte Carlo based beam dose calculations Phys. Med. Biol Van de Walle J, Martens C, Reynaert N, Palmans H, Coghe M, De Neve W, De Wagter C and Thierens H 3 Monte Carlo model of the Elekta SLiplus accelerator: validation of a new MLC component module in BEAM for a 6 MV beam Phys. Med. Biol Verhaegen F and Seuntjens J 3 Monte Carlo modelling of external radiotherapy photon beams Phys. Med. Biol. 48 R7 64