- Email: [email protected]
Influence of calculation model on dose distribution in stereotactic radiotherapy for pulmonary targets
Int. J. Radiation Oncology Biol. Phys., Vol. 61, No. 1, pp. 239 –249, 2005 Copyright © 2005 Elsevier Inc. Printed in the USA. All rights reserved 0360-3016/05/$–see front matter
doi:10.1016/j.ijrobp.2004.03.028
PHYSICS CONTRIBUTION
INFLUENCE OF CALCULATION MODEL ON DOSE DISTRIBUTION IN STEREOTACTIC RADIOTHERAPY FOR PULMONARY TARGETS ULRICH HAEDINGER, PH.D., THOMAS KRIEGER, M.S., MICHAEL FLENTJE, M.D.,
AND
JOERN WULF, M.D.
Department of Radiotherapy, University of Wuerzburg, Wuerzburg, Germany Purpose: To compare the pencil beam (PB) and collapsed cone (CC)-based three-dimensional dose calculation used for stereotactic irradiation of pulmonary targets. Methods and Materials: Three-dimensional conformal dose distributions (using 6-MV and 18-MV photon beams) were generated for 33 pulmonary targets using the PB algorithm implemented in the Helax-TMS treatment planning system and then recalculated with the CC algorithm of TMS using an identical beam setup and parameters. The differences were analyzed by evaluating the dose–volume histograms for the planning target volume (PTV) and clinical target volume (CTV) and evaluating the computed absolute monitor units (MUs). The influence of the photon energy was also studied. For three cases, the results were compared with Monte-Carlo calculations. Results: Use of the CC model typically showed increased dose inhomogeneity. Owing to a more accurate modeling of secondary charged particle disequilibrium at the tumor–lung interface, the beam penumbra is broadened. The median and mean target dose decreased by 13.9% and 11.2% for the PTV and 9.2% and 9.4% for the CTV, respectively, using the CC algorithm. Consequently, the average PTV dose coverage decreased by 7.1% (SD, 6.5%). On average, the MUs calculated to achieve the prescribed dose were 5.4% (SD, 5.8%) greater for the CC algorithm. The difference in MUs between the PB and CC increased with decreasing PTV size and high photon energy (18 MV; r ⴝ ⴚ0.68), reaching 26% at the maximum. Conclusion: The absorbed dose at the lung–tumor interface calculated by the PB algorithm was considerably greater than the dose calculated using the CC algorithm. In small targets (PTV <100 cm3) and for 18-MV photons, the MUs calculated with PB may lead to an insufficient dose to the target volume. © 2005 Elsevier Inc. Stereotactic radiotherapy, Lung tumors, Three-dimensional conformal radiotherapy, Pencil beam model, Collapsed cone algorithm, Monte-Carlo calculation, Target dose coverage.
INTRODUCTION Originally developed for tumors in the brain, stereotactic radiotherapy (RT) is now successfully applied to extracranial lesions. Targets in the lung, liver, abdomen, and pelvis are treated in either single fractions of 14 –26 Gy or using a hypofractionated treatment approach with two to eight fractions of 5–20 Gy (1–10). Various clinical groups have described their treatment approaches and presented their results with respect to local tumor control and treatmentrelated toxicity. A comparison of the results is difficult because of the different tumor doses, normalization procedures, and prescription options used. In addition, most of the publications did not provide any detailed information about the calculation model used to compute the dose distributions. For targets in the lung, the computed dose distributions strongly depend on the calculation model. Pulmonary tumors are embedded within lower density lung tissue (density, 0.1– 0.3 g/cm3), and, therefore, the charged secondary
particle equilibrium at the interface between the tumor and lung tissue is violated. The absorbed dose distribution in the patient can differ significantly from the calculated dose distribution if a calculation model that is too simplistic is used. This discrepancy might be of more importance in stereotactic RT for pulmonary targets. These tumors are usually very small, and the ratio of tumor surface and volume increases with decreasing target volumes, so that surface effects become more and more important. In this study, the influence of the calculation model on the relative dose distributions and computed absolute monitor units (MUs) was analyzed by comparing a pencil beam (PB) model with a collapsed cone (CC) implementation of the point kernel approach for a broad range of clinical setups. The evaluation was based on the comparison of the minimal, median, and mean relative doses in the planning target volume (PTV) and clinical target volume (CTV), as well as the dose coverage of both volumes. The influence of photon energy (6 MV vs. 18 MV) on these parameters was also
Reprint requests to: Ulrich Haedinger, Ph.D., Department of Radiotherapy, St. Vincentius Hospitals Karlsruhe, Steinhaeuserstrasse 18, D-76135 Karlsruhe, Germany. Tel: (⫹49) 721-8108-5170; Fax:
(⫹49) 721-8108-5152; E-mail: [email protected] Received Oct 28, 2003, and in revised form Mar 11, 2004. Accepted for publication Mar 11, 2004. 239
240
I. J. Radiation Oncology
● Biology ● Physics
analyzed. Monte-Carlo (MC) calculations were performed for three cases to compare the results with the most accurate dose calculation model available. METHODS AND MATERIALS The concept of extracranial stereotactic RT (ESRT) was introduced in our department in November 1997 on the basis of the experience of Blomgren et al. (1) and Lax et al. (11) of the Karolinska Hospital, Stockholm, Sweden. The treatment of pulmonary targets—together with irradiation of liver tumors—is the main application of this new therapy approach. A detailed description of the treatment procedure in our institution has been previously reported (10, 12–14). For patient immobilization and target motion minimization, a stereotactic body frame (SBF; Elekta Instruments AB, Stockholm, Sweden) was used. The SBF allows precise and reproducible positioning by noninvasive fixation of the patient using an individually shaped vacuum pillow mounted to the SBF. The breathing mobility of the target can be reduced mechanically to 5– 6 mm by increasing the abdominal pressure by pressing a SBF-attached template on the patient’s abdomen (11, 12).
Target characteristics For this study, 33 consecutive pulmonary targets were evaluated. Of these, 13 were primary non–small-cell lung cancer (cT1T3N0), 5 were local recurrences of lung cancer, and 15 were lung metastases. The CTV ranged from 2 to 256 cm3 (median, 58 cm3), and the PTV ranged from 12 to 384 cm3 (median, 122 cm3). Although 24 targets were located centrally in the lung, 5 were attached to, or very close to, the thoracic wall and another 4 were located close to hilar or mediastinal structures. Mechanical breathing control was used in 11 targets (33%).
Treatment planning and dose prescription The CTV was defined by adding a margin of 2–3 mm to the gross tumor volume as it appeared in the lung window of the CT scan for treatment planning. The PTV was defined by adding another 5-mm margin in the axial direction and a 5–10-mm margin in the longitudinal direction to the CTV. After institutional practice, the CTV was defined manually slice by slice, and the margins for the PTV definition were generated using the automatic threedimensional (3D) contour generation tool of the Helax-TMS system. The fractionation pattern for the patients evaluated in this study was 3 ⫻ 10 Gy, 3 ⫻ 12 Gy, or 3 ⫻ 12.5 Gy prescribed to the PTV enclosing the 100% isodose. To achieve equivalent dose prescriptions, treatment plans were optimized until the 100% isodose was the most conformal isodose relative to the PTV. The dose level was fixed to 150% at the isocenter to achieve an inhomogeneous dose distribution. In contrast to the International Commission on Radiation Units and Measurements Report 50 concept (15) and according to the concept of Lax (16), inhomogeneous dose distributions with a dose enhancement of about 50% in the center of the target are intended to increase the effectiveness of RT in potentially hypoxic central tumor areas. The Helax-TMS system, version 6.1A (Nucletron BV, Veenendaal, The Netherlands), was used for 3D conformal treatment planning, after acquisition of the planning CT study (Philips Tomoscan AV, slice thickness 3–5 mm). Pulmonary targets are often convex and approximately spherically or cylindrically shaped. Beam configurations of five to eight,
Volume 61, Number 1, 2005
equally spaced, coplanar, fixed photon beams (Siemens Primus) of 6 MV (n ⫽ 14) or 18 MV (n ⫽ 19) were used as the treatment technique of choice. The rationale for using 18 MV was the greater available dose rate (200 MU/min at 6 MV vs. 300 MU/min at 18 MV), which decreased the radiation and immobilization time by one-third. In some cases, rotational arcs were used for treatment of the patient. Because TMS does not allow the calculation of rotational beams with the CC algorithm, such treatment plans were remodeled using multiple static beams until the original dose distribution was reproduced. The accuracy of the remodeling was proven by comparing isodoses and dose–volume histograms (DVHs) for the CTV and PTV. Typically, one static beam replaced a 30° angular segment of a rotational field. Noncoplanar beams and the extensive use of wedges were avoided to decrease the irradiation time. Beam shapes were adapted to the target projections using the beam’s eye view tool of TMS. The treatment plans were generated using the interactive—PBbased— beam modeling tool of TMS (calculation grid size 4 –10 mm) to select the appropriate number of beams, gantry angles, beam shapes, and weights. The dose distributions were then recomputed using the same algorithm but with a finer calculation grid (maximal possible resolution 2 mm). The maximal number of calculation points is fixed in the TMS, and, therefore, the actual grid size depends on the patient’s outline contour (including the SBF). Consequently, calculation dose grid sizes of 2.0 – 4.8 mm were used in this study. No statistically significant differences were found in the DVH when the dose grid size was varied between 2 and 5 mm for 3 randomly selected patients (data not shown). The PB implementation in TMS has been described by Ahnesjo¨ (17) and Helax (18). An equivalent path length method was used for one-dimensional longitudinal inhomogeneity corrections. Separate correction factors for primary and scattered photons were applied, and lateral tissue inhomogeneities were not taken into account. An evaluation of the performance of the algorithm in the clinical environment has been reported by Kno¨o¨s et al. (19). In a second step, the treatment plans were recalculated using the collapsed cone option of TMS (20). In this approach, the dose deposition kernels are approximated as a set of discrete cones of a fixed solid angle. Energy transport within each cone is collapsed to the cone axis. During the transport along the cone axis, absorption and attenuation is scaled by local electron density as converted from Hounsfield numbers. To evaluate whether the CC calculation model in the TMS provides accurate results in the vicinity of large heterogeneities, ionizing chamber measurements in a phantom consisting of slabs of rigid water and Styrofoam were performed at our institution. The CC calculations agreed with the measurements within 3%. At Sion, Switzerland, comprehensive quality assurance tests of the CC algorithm in TMS have been performed using different phantom setups to simulate missing lateral and back scatter material and two-dimensional tissue heterogeneities. In addition, film and thermoluminiscent dosimetry (TLD) measurements were performed in a breast phantom and showed satisfactory agreement with the CC dose calculations (21).
Treatment plan evaluation The difference in dose distributions calculated by the PB and CC algorithms was studied by analyzing the DVHs for the CTV and PTV. For each treatment plan, the minimal, maximal, median, and mean relative doses in the CTV and PTV were compared and the differences calculated. Because the minimal target dose alone does not contain any information about the underdosed volume,
Influence of calculation model on dose in pulmonary SRT
the dose coverage of the CTV and PTV by the reference isodose was calculated as:
TC共TV兲 ⫽
V TV.ref , V TV
(1)
where TV denotes either the CTV or PTV (13, 14), VTVref specifies the volume of the target (CTV or PTV) covered by the reference isodose, VTV specifies the target volume itself, and TC (target coverage) refers to those fractions of the CTV or PTV circumscribed by the reference isodose surface (throughout this analysis, the volume included by the 100% isodose surface was defined as the reference volume). The TC was 0 in the case of a total geographic miss of the target and 1 if the CTV/PTV was completely surrounded by the reference isodose. The numbers of MUs calculated by both algorithms for delivery of the prescribed dose to the reference isodose line were also compared. To address the influence of photon energy and the target volume size on the results, the analysis was differentiated by 6-MV vs. 18-MV treatment plans and PTV size. For statistical analysis, Statistica software, version 6.0 (StatSoft), was used. All results were evaluated relative to the PB algorithm. A decrease in dose and TC, together with an increase in MUs, when calculated by the CC-based algorithm, reflects the overestimation tendencies of the PB algorithm. To compare the dose distributions achieved by the PB and CC algorithms to the standard of dose-calculation, MC– based treatment planning was performed for 3 cases. An intermediate target (CTV 21 cm3) located centrally in the lung with its base close to the diaphragm, and the smallest (2 cm3) and largest (256 cm3) CTV were chosen. MC calculations were performed using the Voxel Monte Carlo for photon and electron beams (XVMC) algorithm developed by Fippel et al. from the University of Tu¨bingen, Germany (22–24). The XVMC and its beam model have been demonstrated to reproduce measured (with a diamond detector) relative dose distributions with an accuracy of better than ⫾2% in various homogeneous and inhomogeneous phantoms (23). Because the program version implemented at our department did not allow for DVH computation, the PB, CC, and MC calculated dose distributions (6-MV photons, Siemens Primus) were compared at three representative axial slices at different target levels: the cranial edge of the target, the isocenter level in the central part of the target where the tumor was surrounded by lung tissue only laterally, and the caudal edge of the target. For quantitative comparison of the dose distributions achieved by the different calculation algorithms, dose profiles intersecting the isocenter in the craniocaudal, lateral, and AP directions were generated (see Fig. 6).
RESULTS Comparison of target dose calculated by PB and CC algorithms The comparison of dose distributions by the DVHs computed by the PB and CC algorithms revealed increased inhomogeneity for the CC calculation because of charged particle disequilibrium at the tumor–lung tissue interface. Typically, the volumes covered by the high isodoses levels (100 –140%) shrunk significantly if calculated by the CC algorithm, resulting in reduced target doses (Fig. 1). Although the maximal dose remained relatively unaffected
● U. HAEDINGER et al.
241
(average difference ⫺3.7%; SD 4.7%), the average median (PTV, ⫺13.9% and CTV, ⫺9.2%), mean (PTV, ⫺11.2% and CTV, ⫺9.4%), and minimal (PTV, ⫺5.5% and CTV, ⫺14.9%) doses decreased significantly using the CC algorithm (Table 1). Evaluation of the differences between both algorithms with respect to the PTV size and photon energy revealed that the decrease in the median relative dose was generally greater for 18 MV (mean difference, ⫺16.4%; SD, 5.1%) compared with 6 MV (mean difference, ⫺10.4%; SD, 5.4%). The difference was greater for smaller targets but without a statistically significant correlation to PTV size (18 MV r ⫽ 0.14; 6 MV r ⫽ 0.34; Fig. 2). The initial treatment plans, designed on the basis of the PB algorithm, yielded large values for TC (PTV, average, 96.1%,; SD, 3.1%; CTV, average, 99.9%; SD, 0.3%) independent of the target volume size or photon energy. Minimal values for TC were 84.7% for the PTV and 98.3% for the CTV. Recomputing the treatment plans using the CC model reduced the average PTV coverage by approximately 7%. The differentiation of this effect by photon energy revealed a volume independent (r ⫽ 0.06) decrease in the average PTV coverage of 3.3% for 6 MV (SD 3%). If 18-MV photons were used, dose calculation by the CC algorithm led to a decrease of the average PTV coverage by 9.9% (SD 7%). This effect was most evident in smaller targets (⬍200 cm3; r ⫽ 0.31). Nevertheless, the CTV coverage remained basically unaffected for both algorithms and both photon energies. Large TC values can be achieved at the expense of conformality if the dose distributions are too generous. In a recent analysis, we found that in ESRT for lung targets, highly conformal treatment techniques could be applied, primarily as a result of the relatively simple geometric shape of most pulmonary targets (13). Even as the PTV coverage dropped to a minimal value of 69.5%, the minimal CTV coverage found with CC was 93.2%. This indicates that in the worst case, approximately 7% of the CTV received less than the reference dose because of “shrinkage” of the volume covered by the reference isodose. The results are detailed in Table 1 and Fig. 3.
Comparison of MUs calculated by PB and CC algorithms The computed MUs to achieve the reference dose were generally greater for the CC algorithm. On average, the MUs increased by 5.4% (SD 5.8%) compared with the PB-based calculation. The difference was most prominent for 18-MV photons (mean, ⫹8.4%; SD, 5.5%) compared with 6-MV photons (mean, ⫹1.5%; SD, 3.4%). The dependence of the MU ratio on the PTV size is shown in Fig. 4. Although no correlation was found between PTV size and the number of MUs for 6-MV photons (r ⫽ ⫺0.14), the number of MUs calculated by the CC method for 18-MV treatment plans were significantly greater for smaller targets (e.g., PTV ⬍100 cm3; r ⫽ ⫺0.68). For example, the smallest target in the sample (CTV 2 cm3, PTV 12 cm3) irradiated by an 18-MV dose plan showed the largest difference in calculated MUs (21%).
242
I. J. Radiation Oncology
● Biology ● Physics
Volume 61, Number 1, 2005
Fig. 1. Difference in dose–volume histograms (DVHs) from dose plans (6 MV) calculated by pencil beam (PB) and collapsed cone (CC) algorithms for lung target in right lower lobe (clinical target volume [CTV] 21 cm3; planning target volume [PTV] 42 cm3). Left curves in each DVH from CC calculations; right curves show overestimation of target dose calculated by PB algorithm. In this example (dose distributions shown in Fig. 5), minimal dose in CTV and PTV decreased from 124% and 90% to 103% and 81% (change, ⫺21% and ⫺9%) and median dose decreased from 145% and 137% to 137% and 123% (change, ⫺8% and ⫺14%), respectively, when calculated with CC algorithm.
Comparison of PB and CC algorithms to MC calculations To support the reliability of CC-based dose calculations, MC calculations were performed on 3 cases using 6-MV photons. An example of the axial dose distributions for the intermediate-size target (CTV 21 cm3, PTV 42 cm3) at the isocenter level and 1.5 cm cranial and 1.5 cm caudal of the
isocenter is shown in Fig. 5. The differences are most clearly visible in the dose profiles displayed in Fig. 6. All profiles demonstrated a decrease in the high-dose volume (isodose levels ⬎100%) from the PB to CC to MC algorithm. Although only three examples were calculated, the comparison with the MC calculation indicated that the CC
Fig. 2. Differences in median relative planning target volume (PTV) dose as calculated by pencil beam (PB) and collapsed cone (CC) algorithms. On average, median dose decreased by ⫺13.9% if CC algorithm was used. Differences increased with smaller targets and higher photon energy.
Influence of calculation model on dose in pulmonary SRT
Table 1. Comparison of dosimetric data for CTV and PTV for 33 pulmonary targets as calculated with Helax-TMS using pencil beam model and collapsed cone algorithm
Data PTV coverage Average/SD Range PTV dose (average/ SD) Minimum Maximum Median Mean SD CTV coverage Average/SD Range CTV dose (average/ SD) Minimum Maximum Median Mean SD Monitor units (average/ SD) All targets 6 MV 18 MV
Pencil beam calculation
Collapsed cone calculation
96.1/3.1 84.7/100.0
89.1/8.2 69.5/99.5
80.6/11.4 157.8/5.7 141.3/5.6 136.6/5.2 15.9/2.9 99.9/0.3 98.3/100.0
75.0/9.9 154.1/4.6 127.4/9.7 125.4/7.9 16.7/2.2 99.5/1.3 93.2/100.0
⌬ CC-PB ⫺7.1/6.5
⫺5.5/5.6 ⫺3.7/4.7 ⫺13.9/5.9 ⫺11.2/4.1 0.9/1.8 ⫺0.4/1.0
120.7/12.3 157.8/5.7 148.6/2.5 147.5/2.9 6.2/2.7
105.8/3.3 154.1/4.6 139.4/6.1 138.1/5.8 9.7/2.6
⫺14.9/9.4 ⫺3.7/4.7 ⫺9.2/4.9 ⫺9.4/4.0 3.5/2.5
100 fixed 100 fixed 100 fixed
105.4/5.8 101.5/3.4 108.4/5.5
5.4/5.8 1.5/3.4 8.4/5.5
Abbreviations: CC ⫽ collapsed cone; PB ⫽ pencil beam; PTV ⫽ planning target volume; CTV ⫽ clinical target volume; SD ⫽ standard deviation. Last column shows average difference between both calculation algorithms. Although the median and mean PTV dose decreased 13.9% and 11.2%, respectively, on average, the minimal CTV dose decreased by 14.9% and mean/median CTV doses by almost 10%. The SD of the dose in the CTV increased by 3.5%, on average, indicating a more inhomogeneous dose distribution with the CC model. The monitor units to achieve the reference dose were, on average, 5.4% greater with the CC algorithm (8.4% for targets irradiated with 18-MV photons).
algorithm is more accurate than the PB calculation in the high-dose region and that the PB-based dose calculation might overestimate the therapeutic target dose.
DISCUSSION At different centers, primary and/or metastatic lung tumors are treated using the ESRT approach (1, 2, 4 –10). Dose-escalation studies have been performed to define the appropriate doses for achieving tumor eradication without severe treatment-related side effects (8). It is difficult to compare the results because a variety of dose fractionation schedules and different procedures for dose normalization
● U. HAEDINGER et al.
243
and prescription have been used. Furthermore, the aspect of dose calculation accuracy was often neglected. The presented results showed the largest differences between calculation models in small targets. The median CTV in this study was 58 cm3, considerably larger than the CTV in the reports of ESRT for lung tumors by other authors, which ranged from 4.0 to 22.5 cm3 (1, 2, 4, 6, 8). Under certain circumstances, the use of 18-MV photons in this study was desirable for ESRT to lung tumors, because the higher photon energy was delivered at a higher dose rate by the Siemens Primus linear accelerators at our institution. This factor becomes even more relevant if breathing-triggered RT is used (increase of RT time owing to restriction of radiation to a defined breathing phase). Because of the very small target volumes and the attraction of the higher photon energy, evaluation of the accuracy of dose-calculation algorithms becomes extremely important in ESRT for lung tumors. PB-based algorithms are widely used in computerized treatment planning systems. The limitations of this approach for dose calculation in the case of lung tumors have been reported by Kno¨o¨s et al. (19). Electronic equilibrium is generally disturbed at density inhomogeneities such as at the tumor–lung interface. This results in incorrect predictions of the calculated dose if energy deposition kernels, such as derived from homogeneous media, are used. PB models apply corrections for tissue inhomogeneities in one dimension by scaling the PB kernels in a longitudinal direction. Because MC approaches are exact mathematical models of fundamental physical interaction processes, they can, in principle, predict the dose with arbitrary accuracy. However, these approaches remain slow and are not yet widely accessible clinically. As an alternative, superposition/convolution models have been developed, which take into account tissue inhomogeneities by scaling the “pointlike” energy deposition kernels according to the local density distribution in all three dimensions (a list of references can be found in Butson et al. [25]). To speed up computation time, such a model has been implemented into the TMS in the so-called collapsed cone approach. The energy deposition kernels are approximated as a set of discrete cones, and the energy attenuation and transport is restricted to the cones axis or “collapsed to the cones axis” (the CC implementation in Helax-TMS is extensively described by Ahnesjo¨ [20]). This procedure promises a more reliable prediction of doses in heterogeneous media. Various examinations have shown satisfying agreement between the CC calculations and measurements in phantoms in which slabs of low-density material simulate lung tissue. Butson et al. (25) obtained adequate accuracy within an anthropomorphic lung phantom for 6-MV and 10-MV photon energies. Francescon et al. (26) compared a CC model with the MC code BEAM for mediastinal and breast treatment. Performing DVH analysis in real patient CT data sets, they concluded that CC and MC give absolute comparable results for common 3D-RT planning. Butts and Foster (27) concluded that the differences between diode measurements and the
244
I. J. Radiation Oncology
● Biology ● Physics
Volume 61, Number 1, 2005
Fig. 3. Planning target volume (PTV) coverage by 100% reference isodose plotted vs. PTV size using pencil beam (PB) (open circles, open triangles) and collapsed cone (CC) (solid circles, solid triangles), differentiated by photon energy of 6 MV (triangles) and 18 MV (circles). Although a volume-independent decrease in TC of 3.3% indicated at mean by CC algorithm for 6-MV photons, TC decreased by 9.9% at mean if dose was calculated by CC with 18-MV photons. This effect increased with decreasing PTV size. Difference between photon energies and its relation to target volume was not detected with PB algorithm.
Fig. 4. Ratio of monitor units (MUs) as calculated by collapsed cone (CC) and pencil beam (PB) algorithms. Amount of MUs necessary to achieve planned reference dose was significantly greater when calculated by CC algorithm, leading to ratio of ⬎100% in most cases. Results plotted vs. planning target volume (PTV) size and differentiated by photon energy, showing that MU ratio increased significantly for small targets and 18-MV photon radiation and indicating large dose overestimations by PB algorithm in these cases.
Influence of calculation model on dose in pulmonary SRT
● U. HAEDINGER et al.
245
Fig. 5. Dose distributions (6-MV photons) for target in central lung located above diaphragm (clinical target volume [CTV] 21 cm3; planning target volume [PTV] 42 cm3) as calculated with (a) Monte Carlo (MC) algorithm, (b) collapsed cone (CC) algorithm, and (c) pencil beam (PB) algorithm. Slices located at cranial edge of target (1.5 cm cranial of isocenter, left column), at isocenter level (central column), and base of target (1.5 cm caudal of isocenter, right column). Dose normalized to 150% level at isocenter of treatment plan. Volumes covered by 140%, 120%, and 100% isodoses decreased from PB to CC to MC, indicating that even CC-based calculations might overestimate doses compared with MC calculations. Dose profiles in three main directions shown in Fig. 6.
CC monitor unit calculation in an anthropomorphic phantom were within acceptable limits. In this study, the influence of the dose-calculation algorithms (PB vs. CC implemented in the Helax-TMS treat-
ment planning system) for a significant number (n ⫽ 33) of clinical cases in ESRT for lung tumors using 6-MV (n ⫽ 14) or 18-MV (n ⫽ 19) photons was evaluated. The analysis revealed that the dose at the interface of the tumor and
246
I. J. Radiation Oncology
● Biology ● Physics
Volume 61, Number 1, 2005
Influence of calculation model on dose in pulmonary SRT
● U. HAEDINGER et al.
247
Fig. 6. Comparison of dose profiles (6-MV photons) in craniocaudal (left), lateral (middle), and AP (right) directions calculated by Monte Carlo (MC), collapsed cone (CC), and pencil beam (PB) algorithms for three examples. (a) Largest clinical target volume (CTV) (256 cm3). (b) Intermediate-size target (dose distribution shown in Fig. 5) with clinical target volume (CTV) of 21 cm3. (c) Smallest CTV (2 cm3). Profiles intersect at isocenter (0 cm). Dose was prescribed to PTV enclosing 100% isodose with normalization to 150% at isocenter. Profiles revealed that CC-calculated dose corresponded better to MC-based calculations than did PB calculations, especially in therapeutic dose area of 100 –150%. Proportional effect of PB algorithm dose overestimation increased with decreases in target volume.
low-density lung tissue is considerably overestimated by the PB algorithm, especially in small targets ⱕ100 cm3 irradiated with 18-MV photons. On average, the CC calculations resulted in a 7% reduced PTV coverage. Differentiated by photon energy, the average PTV coverage decreased by only 3.3% for 6-MV but 9.9% for 18-MV photons. The comparison with MC calculations showed that the CC algorithm is more reliable than the PB algorithm, especially in the high-dose area between the 100% and the 150% isodose level. In relation to the total target volume, this effect increased with a decreasing target volume because of the larger interface to lung tissue. Therefore, a statistically significant overestimation of dose (average deviation of median dose ⫺13.9%) and PTV TC (average difference ⫺9.9%) and underestimation of MUs (average difference 8.4%) using the PB calculation must be considered if small target volumes are irradiated and dose prescription is performed to a PTV enclosing isodose instead of to the isocenter. The effect is further increased with the use of highenergy photons (e.g., 18 MV). No decrease in average dose coverage to the CTV was found. This indicated that the margins applied in the PTV definition to allow for target positioning/movement uncertainties—not for dose-calculation uncertainties—were sufficient to ensure CTV TC under the condition of high geometric accuracy of treatment delivery. In a recent analysis of the impact of target reproducibility on tumor dose in ESRT (calculations based on a PB model) it was shown, by matching studies from CT simulation and treatment planning, that in 91% of the lung targets CTV dose coverage ⬎95% was achieved (14). In the present study, a statistically significant reduction of the median and mean relative doses to the CTV (average reduction median ⫺9.2% and mean ⫺9.4%) and PTV (median ⫺13.9% and mean ⫺11.2%) in the CC calculation was found, indicating that about 10% less dose was delivered to the CTV than intended, even under optimal geometric conditions. Our results expand on the work of other authors, which was based on a range of clinical situations comparing different dose-calculation approaches and measurements in phantoms. Engelsman et al. (28) examined the impact of simple tissue-inhomogeneity correction algorithms on conformal RT for lung tumors in a specially designed lung phantom with an AP irradiation geometry. Using film measurements, they found that the field margins will be underestimated by PB algorithms. The actual energy-dependent broadening of the beam penumbra is not correctly described by the PB algorithm of TMS. They reported values for PTV
underdosage (with respect to the planned 95% isodose level) of between 10% and 15% for low energy beams (6 MV) and ⬎20% for higher energies. The authors stated that the underdosage found for their PTV did not necessarily lead to an underdosage of the gross tumor volume or CTV. This was also reflected in our data. Even if the PTV dose coverage drops when using the CC method, the difference in the planned CTV coverage was not statistically significant. From this point of view, there seems to be no strong influence of the dose calculation model because neither the average nor the minimally detected CTV coverage changed dramatically in our evaluation. Nevertheless, the PTV is defined to cover uncertainties in the CTV reproducibility due to setup inaccuracy and breathing mobility. It is clinically reasonable to expect the CTV to be located at the edge of the PTV during treatment. In this case, a decrease in PTV dose coverage would correspond to an equally reduced dose to the CTV. Also, although the average CTV coverage did not decrease within this study, the median and mean doses in the CTV decreased by approximately 10%, reducing the intended dose to the CTV. The increased influence of this effect was clearly demonstrated with decreasing target size. Although the clinical impact of the additional target dose achieved by the inhomogeneous dose distribution is still unclear, a considerable decrease of the dose to the CTV must be taken into account when the PB algorithm is used. Extracranial stereotactic radiotherapy is not only performed as hypofractionated treatment—such as in the present study— but also as single-dose treatment (2, 4, 6). In contrast to 3D-conformal RT in multiple fractions, underdosage of a certain proportion of the target because of setup inaccuracy or target mobility will not be compensated for by random error during the treatment course. Reliable dose calculation is then of extreme importance, primarily to avoid local failures and secondarily to allow for interpretation of failures (e.g., due to target miss or insufficient setup accuracy, target definition, or target dose). Several studies have been started to establish the most appropriate dose for ESRT. If dose calculation is possibly inaccurate, as demonstrated in this evaluation, the interpretation of doses from these studies will be difficult. Therefore, dose-calculation algorithms, as well as the photon energy used for treatment planning, should be reported. The influence of photon energy on the target dose has been evaluated by other authors using MC calculations. Saitoh et al. (29) examined the dose distribution for narrow beam irradiation of a lung phantom by comparing Electron Gamma Shower Code Version 4 (EGS4) MC simulation
248
I. J. Radiation Oncology
● Biology ● Physics
and dosimetric films and proved the broadening of dose profiles at the edges of the tumor owing to the lack of scattering material. They concluded that the variations in relative dose distributions and beam profiles change as a function of incident photon energy. Because the dispersion of absorbed energy in tumors increases with increased photon energy, they recommended the use of low-photon energies for stereotactic treatment of lung tumors. Papiez (30) studied the influence of photon energy, target size and location, tissue density, and complexity of the treatment plan (number of beams) in a mathematical model of the human chest using the PENELOPE MC code. He reported a significant dose reduction at the tumor edge (15–20% for 6 MV X-rays and 25–50% for 15 MV X-rays) that was not correctly described by commercial treatment planning systems with semiempiric beam attenuation correction or firstorder scatter corrections. Linthout et al. (31) studied the effect of the dose-calculation model for dynamic arc treatments in a humanoid head-and-neck phantom. They found discrepancies between the calculated dose at the interface of tumor–air cavities and measurements with dosimetric film, as well as TLD. In their study, the PB algorithm overestimated the dose in the PTV at the tissue–air boundaries by approximately 10%. Differences were smaller for a CC algorithm. Verellen et al. (32) reported similar results. Martens et al. (33) compared radiochromic film measurements, MC simulations, and CC calculations for intensity-modulated RT in a phantom simulating the air cavity of the trachea. They obtained very good agreement between the film measurement and MC, but the CC algorithms were not able to predict the interface dose correctly. The CC calculations showed considerable improvements compared with the PB models. In addition to the differences in relative dose distributions, large discrepancies in the numbers of MUs calculated with PB or CC algorithms can emerge. This usually happens if the dose is prescribed to a PTV-enclosing isodose (i.e., in a region in which secondary charged particle equilibrium is not maintained). In other methods that prescribe the dose to the International Commission on Radiation Units and Measurements reference point in the center of the PTV ade-
Volume 61, Number 1, 2005
quately surrounded by tumor tissue, the accuracy of the PB calculation of MUs is sufficiently high (28). As a result of this evaluation and the experiences of other authors, we now use only the CC algorithm of the HelaxTMS treatment planning system and 6-MV photons for stereotactic RT to pulmonary targets. The dose is delivered to the patient using the absolute MUs computed by the CC algorithm. To compensate for the reduction in the median and mean relative target dose, the beam widths have to be enlarged by adding an extra “penumbra broadening margin” of 3–5 mm to the beams originally designed with the interactive PB-based algorithm. Another approach is an additional, small-size beam designed to increase the dose in the “underdosed” part of the target. Alternatively, the dose normalization level could be raised from 150% to 160%, because the maximal dose is located at the center of the tumor and will not harm the patient as long as the dose is delivered precisely to the target. CONCLUSION The issue of dose-calculation is important for ESRT for targets of the lung, especially when various clinical studies are compared. Because of their intrinsic limitations, the present, widely used PB algorithms overestimate the absorbed dose at the interface of tumor and low-density lung tissue. Therefore, the more reliable CC models should be used in these cases, as long as MC calculation is not available in treatment planning systems for routine clinical use. When the dose is prescribed to the edge of the target instead of the International Commission on Radiation Units and Measurements reference point in the center of the tumor, the numbers of MUs calculated by PB algorithms can be considerably too low. This is especially the case for small targets with a PTV ⱕ100 cm3 and high-energy photons. In these cases, use of a PB algorithm can result in an overestimation of the doses at the reference isodose level of up to 20%. In a 3 ⫻ 10-Gy concept, this would be equivalent to a decrease in the total dose of 6 Gy. Additional systematic studies exploring the influence of photon energy, the specific target location within the lung, and the potential of MC-based dose calculation algorithms are needed.
REFERENCES 1. Blomgren H, Lax I, Go¨ranson H, et al. Radiosurgery for tumors in the body: Clinical experience using a new method. J Radiosurg 1998;1:63–74. 2. Hara R, Itami J, Kondo T, et al. Stereotactic single high dose irradiation of lung tumors under respiratory gating. Radiother Oncol 2002;63:159–163. 3. Herfarth KK, Debus J, Lohr F, et al. Stereotactic single dose radiation therapy of liver tumors: Results of a phase I/II trial. J Clin Oncol 2001;19:164–170. 4. Hof H, Herfahrt KK, Munter M, et al. Stereotactic single-dose radiotherapy of stage I non-small-cell lung cancer (NSCLC). Int J Radiat Oncol Biol Phys 2003;56:335–341. 5. Nagata Y, Negoro Y, Aoki T. Clinical outcomes of 3D conformal hypofractionated single high-dose radiotherapy for one
or two lung tumors using a stereotactic body frame. Int J Radiat Oncol Biol Phys 2002;52:1041–1046. 6. Nakagawa Y, Aoki Y, Tago M, et al. Megavoltage CTassisted stereotactic radiosurgery for thoracic tumors: Original research in the treatment of thoracic neoplasms. Int J Radiat Oncol Biol Phys 2000;48:449–457. 7. Onimaru R, Shirato H, Shimizu S, et al. Tolerance of organs at risk in small-volume, hypofractionated, image-guided radiotherapy for primary and metastatic lung. Int J Radiat Oncol Biol Phys 2003;56:126–135. 8. Timmerman RD, Papiez L, McGarry R. Extracranial stereotactic radioablation: Results of a phase I study in medically inoperable stage I non-small cell lung cancer. Chest 2003;124: 1946 –1955.
Influence of calculation model on dose in pulmonary SRT
9. Uematsu M, Shioda M, Suda A, et al. Computed tomographyguided frameless stereotactic radiotherapy for stage I nonsmall cell lung cancer: A 5-year experience. Int J Radiat Oncol Biol Phys 2001;51:666–670. 10. Wulf J, Haedinger U, Oppitz U, et al. Stereotactic radiotherapy of targets in the lung and liver. Strahlenther Onkol 2001; 177:645–655. 11. Lax I, Blomgren H, Larson D, et al. Extracranial stereotactic radiosurgery of localized targets. J Radiosurg 1998;1:135– 148. 12. Wulf J, Haedinger U, Oppitz U, et al. Stereotactic radiotherapy of extracranial targets: CT-simulation and accuracy of treatment in the stereotactic body frame. Radiother Oncol 2000;57:225–236. 13. Haedinger U, Thiele W, Wulf J. Extracranial stereotactic radiotherapy: Evaluation of PTV coverage and dose conformity. Z Med Phys 2002;12:221–229. 14. Wulf J, Haedinger U, Oppitz U, et al. Impact of target reproducibility on tumor dose in stereotactic radiotherapy of targets in the lung and liver. Radiother Oncol 2003;66:141–150. 15. International Commission on Radiation Units and Measurements. ICRU report 50: Prescribing, recording and reporting photon beam therapy. Bethesda: International Commission on Radiation Units and Measurements; 1993. 16. Lax I. Target dose versus extra-target dose in stereotactic radiosurgery. Acta Oncol 1993;32:453–457. 17. Ahnesjo¨ A. A pencil beam model for photon dose calculation. Med Phys 1992;19:263–273. 18. Helax AB. Dose formalism and methods in Helax-TMS: Manual. 1998. 19. Kno¨o¨s T, Ahnesjo¨ A, Nilsson P, et al. Limitation of a pencil beam approach to photon dose calculation in lung tissue. Phys Med Biol 1995;40:1411–1420. 20. Ahnesjo¨ A. Collapsed cone convolution of radiant energy for photon dose calculation in heterogeneous media. Med Phys 1989;16:577–592. 21. Corminboeuf F, Ray JY, Bieri S. Performance evaluation of the collapsed-cone calculation on Helax-TMS. Kerzers: Verlag; Max Huber 2001, p. 128 –134. 22. Fippel M. Fast Monte Carlo dose calculation for photon beams based on the VMC electron algorithm. Med Phys 1999;26: 1466–1475.
● U. HAEDINGER et al.
249
23. Fippel M, Laub W, Huber B, et al. Experimental investigation of a fast Monte Carlo photon beam dose calculation algorithm. Phys Med Biol 1999;44:3039–3054. 24. Fippel M, Nuesslin F. Verification of a fast Monte Carlo dose calculation algorithm by EGSnrc using the statistical separation method. Z Med Phys 2001;11:152–160. 25. Butson MJ, Elferink R, Cheung T, et al. Verification of lung dose in an anthropomorphic phantom calculated by the collapse cone convolution method. Phys Med Biol 2000;45:143– 149. 26. Francescon P, Cavedon C, Reccanello S, et al. Photon dose calculation of a three-dimensional treatment planning system compared to the Monte Carlo code BEAM. Med Phys 2000; 27:1579–1587. 27. Butts JR, Foster AE. Comparison of commercially available three-dimensional treatment planning algorithms for monitor unit calculations in the presence of heterogeneities. J Appl Clin Med Phys 2001;2:32–41. 28. Engelsman M, Damen EMF, Koken PW, et al. Impact of simple tissue inhomogeneity correction algorithms on conformal radiotherapy of lung tumors. Radiother Oncol 2001;60: 299–309. 29. Saitoh H, Fujisaki T, Sakai R, et al. Dose distribution of narrow beam irradiation for small lung tumor. Int J Radiat Oncol Biol Phys 2002;53:1380–1387. 30. Papiez L. Dosimetry of solid tumors in the lung and radioablative therapy. In: Proceedings of the Second Annual Extracranial Stereotactic Radioablation: Future Directions, June 6 – 8, 2003. Halifax;2003. p. 32. 31. Linthout N, Verellen D, van Acker S, et al. Evaluation of dose calculation algorithms for dynamic arc treatments of head and neck tumors. Radiother Oncol 2002;64:85–95. 32. Verellen D, Linthout N, van den Berge D, et al. Initial experience with intensity-modulated conformal radiation therapy for treatment of the head and neck region. Int J Radiat Oncol Biol Phys 1997;39:99–114. 33. Martens C, Reynaert N, De Wagter C, et al. Underdosage of the upper-airway mucosa for small fields as used in intensitymodulated radiation therapy: A comparison between radiochromic film measurements, Monte Carlo simulations and collapsed cone convolution calculations. Med Phys 2002;57: 1528–1535.
doi:10.1016/j.ijrobp.2004.03.028
PHYSICS CONTRIBUTION
INFLUENCE OF CALCULATION MODEL ON DOSE DISTRIBUTION IN STEREOTACTIC RADIOTHERAPY FOR PULMONARY TARGETS ULRICH HAEDINGER, PH.D., THOMAS KRIEGER, M.S., MICHAEL FLENTJE, M.D.,
AND
JOERN WULF, M.D.
Department of Radiotherapy, University of Wuerzburg, Wuerzburg, Germany Purpose: To compare the pencil beam (PB) and collapsed cone (CC)-based three-dimensional dose calculation used for stereotactic irradiation of pulmonary targets. Methods and Materials: Three-dimensional conformal dose distributions (using 6-MV and 18-MV photon beams) were generated for 33 pulmonary targets using the PB algorithm implemented in the Helax-TMS treatment planning system and then recalculated with the CC algorithm of TMS using an identical beam setup and parameters. The differences were analyzed by evaluating the dose–volume histograms for the planning target volume (PTV) and clinical target volume (CTV) and evaluating the computed absolute monitor units (MUs). The influence of the photon energy was also studied. For three cases, the results were compared with Monte-Carlo calculations. Results: Use of the CC model typically showed increased dose inhomogeneity. Owing to a more accurate modeling of secondary charged particle disequilibrium at the tumor–lung interface, the beam penumbra is broadened. The median and mean target dose decreased by 13.9% and 11.2% for the PTV and 9.2% and 9.4% for the CTV, respectively, using the CC algorithm. Consequently, the average PTV dose coverage decreased by 7.1% (SD, 6.5%). On average, the MUs calculated to achieve the prescribed dose were 5.4% (SD, 5.8%) greater for the CC algorithm. The difference in MUs between the PB and CC increased with decreasing PTV size and high photon energy (18 MV; r ⴝ ⴚ0.68), reaching 26% at the maximum. Conclusion: The absorbed dose at the lung–tumor interface calculated by the PB algorithm was considerably greater than the dose calculated using the CC algorithm. In small targets (PTV <100 cm3) and for 18-MV photons, the MUs calculated with PB may lead to an insufficient dose to the target volume. © 2005 Elsevier Inc. Stereotactic radiotherapy, Lung tumors, Three-dimensional conformal radiotherapy, Pencil beam model, Collapsed cone algorithm, Monte-Carlo calculation, Target dose coverage.
INTRODUCTION Originally developed for tumors in the brain, stereotactic radiotherapy (RT) is now successfully applied to extracranial lesions. Targets in the lung, liver, abdomen, and pelvis are treated in either single fractions of 14 –26 Gy or using a hypofractionated treatment approach with two to eight fractions of 5–20 Gy (1–10). Various clinical groups have described their treatment approaches and presented their results with respect to local tumor control and treatmentrelated toxicity. A comparison of the results is difficult because of the different tumor doses, normalization procedures, and prescription options used. In addition, most of the publications did not provide any detailed information about the calculation model used to compute the dose distributions. For targets in the lung, the computed dose distributions strongly depend on the calculation model. Pulmonary tumors are embedded within lower density lung tissue (density, 0.1– 0.3 g/cm3), and, therefore, the charged secondary
particle equilibrium at the interface between the tumor and lung tissue is violated. The absorbed dose distribution in the patient can differ significantly from the calculated dose distribution if a calculation model that is too simplistic is used. This discrepancy might be of more importance in stereotactic RT for pulmonary targets. These tumors are usually very small, and the ratio of tumor surface and volume increases with decreasing target volumes, so that surface effects become more and more important. In this study, the influence of the calculation model on the relative dose distributions and computed absolute monitor units (MUs) was analyzed by comparing a pencil beam (PB) model with a collapsed cone (CC) implementation of the point kernel approach for a broad range of clinical setups. The evaluation was based on the comparison of the minimal, median, and mean relative doses in the planning target volume (PTV) and clinical target volume (CTV), as well as the dose coverage of both volumes. The influence of photon energy (6 MV vs. 18 MV) on these parameters was also
Reprint requests to: Ulrich Haedinger, Ph.D., Department of Radiotherapy, St. Vincentius Hospitals Karlsruhe, Steinhaeuserstrasse 18, D-76135 Karlsruhe, Germany. Tel: (⫹49) 721-8108-5170; Fax:
(⫹49) 721-8108-5152; E-mail: [email protected] Received Oct 28, 2003, and in revised form Mar 11, 2004. Accepted for publication Mar 11, 2004. 239
240
I. J. Radiation Oncology
● Biology ● Physics
analyzed. Monte-Carlo (MC) calculations were performed for three cases to compare the results with the most accurate dose calculation model available. METHODS AND MATERIALS The concept of extracranial stereotactic RT (ESRT) was introduced in our department in November 1997 on the basis of the experience of Blomgren et al. (1) and Lax et al. (11) of the Karolinska Hospital, Stockholm, Sweden. The treatment of pulmonary targets—together with irradiation of liver tumors—is the main application of this new therapy approach. A detailed description of the treatment procedure in our institution has been previously reported (10, 12–14). For patient immobilization and target motion minimization, a stereotactic body frame (SBF; Elekta Instruments AB, Stockholm, Sweden) was used. The SBF allows precise and reproducible positioning by noninvasive fixation of the patient using an individually shaped vacuum pillow mounted to the SBF. The breathing mobility of the target can be reduced mechanically to 5– 6 mm by increasing the abdominal pressure by pressing a SBF-attached template on the patient’s abdomen (11, 12).
Target characteristics For this study, 33 consecutive pulmonary targets were evaluated. Of these, 13 were primary non–small-cell lung cancer (cT1T3N0), 5 were local recurrences of lung cancer, and 15 were lung metastases. The CTV ranged from 2 to 256 cm3 (median, 58 cm3), and the PTV ranged from 12 to 384 cm3 (median, 122 cm3). Although 24 targets were located centrally in the lung, 5 were attached to, or very close to, the thoracic wall and another 4 were located close to hilar or mediastinal structures. Mechanical breathing control was used in 11 targets (33%).
Treatment planning and dose prescription The CTV was defined by adding a margin of 2–3 mm to the gross tumor volume as it appeared in the lung window of the CT scan for treatment planning. The PTV was defined by adding another 5-mm margin in the axial direction and a 5–10-mm margin in the longitudinal direction to the CTV. After institutional practice, the CTV was defined manually slice by slice, and the margins for the PTV definition were generated using the automatic threedimensional (3D) contour generation tool of the Helax-TMS system. The fractionation pattern for the patients evaluated in this study was 3 ⫻ 10 Gy, 3 ⫻ 12 Gy, or 3 ⫻ 12.5 Gy prescribed to the PTV enclosing the 100% isodose. To achieve equivalent dose prescriptions, treatment plans were optimized until the 100% isodose was the most conformal isodose relative to the PTV. The dose level was fixed to 150% at the isocenter to achieve an inhomogeneous dose distribution. In contrast to the International Commission on Radiation Units and Measurements Report 50 concept (15) and according to the concept of Lax (16), inhomogeneous dose distributions with a dose enhancement of about 50% in the center of the target are intended to increase the effectiveness of RT in potentially hypoxic central tumor areas. The Helax-TMS system, version 6.1A (Nucletron BV, Veenendaal, The Netherlands), was used for 3D conformal treatment planning, after acquisition of the planning CT study (Philips Tomoscan AV, slice thickness 3–5 mm). Pulmonary targets are often convex and approximately spherically or cylindrically shaped. Beam configurations of five to eight,
Volume 61, Number 1, 2005
equally spaced, coplanar, fixed photon beams (Siemens Primus) of 6 MV (n ⫽ 14) or 18 MV (n ⫽ 19) were used as the treatment technique of choice. The rationale for using 18 MV was the greater available dose rate (200 MU/min at 6 MV vs. 300 MU/min at 18 MV), which decreased the radiation and immobilization time by one-third. In some cases, rotational arcs were used for treatment of the patient. Because TMS does not allow the calculation of rotational beams with the CC algorithm, such treatment plans were remodeled using multiple static beams until the original dose distribution was reproduced. The accuracy of the remodeling was proven by comparing isodoses and dose–volume histograms (DVHs) for the CTV and PTV. Typically, one static beam replaced a 30° angular segment of a rotational field. Noncoplanar beams and the extensive use of wedges were avoided to decrease the irradiation time. Beam shapes were adapted to the target projections using the beam’s eye view tool of TMS. The treatment plans were generated using the interactive—PBbased— beam modeling tool of TMS (calculation grid size 4 –10 mm) to select the appropriate number of beams, gantry angles, beam shapes, and weights. The dose distributions were then recomputed using the same algorithm but with a finer calculation grid (maximal possible resolution 2 mm). The maximal number of calculation points is fixed in the TMS, and, therefore, the actual grid size depends on the patient’s outline contour (including the SBF). Consequently, calculation dose grid sizes of 2.0 – 4.8 mm were used in this study. No statistically significant differences were found in the DVH when the dose grid size was varied between 2 and 5 mm for 3 randomly selected patients (data not shown). The PB implementation in TMS has been described by Ahnesjo¨ (17) and Helax (18). An equivalent path length method was used for one-dimensional longitudinal inhomogeneity corrections. Separate correction factors for primary and scattered photons were applied, and lateral tissue inhomogeneities were not taken into account. An evaluation of the performance of the algorithm in the clinical environment has been reported by Kno¨o¨s et al. (19). In a second step, the treatment plans were recalculated using the collapsed cone option of TMS (20). In this approach, the dose deposition kernels are approximated as a set of discrete cones of a fixed solid angle. Energy transport within each cone is collapsed to the cone axis. During the transport along the cone axis, absorption and attenuation is scaled by local electron density as converted from Hounsfield numbers. To evaluate whether the CC calculation model in the TMS provides accurate results in the vicinity of large heterogeneities, ionizing chamber measurements in a phantom consisting of slabs of rigid water and Styrofoam were performed at our institution. The CC calculations agreed with the measurements within 3%. At Sion, Switzerland, comprehensive quality assurance tests of the CC algorithm in TMS have been performed using different phantom setups to simulate missing lateral and back scatter material and two-dimensional tissue heterogeneities. In addition, film and thermoluminiscent dosimetry (TLD) measurements were performed in a breast phantom and showed satisfactory agreement with the CC dose calculations (21).
Treatment plan evaluation The difference in dose distributions calculated by the PB and CC algorithms was studied by analyzing the DVHs for the CTV and PTV. For each treatment plan, the minimal, maximal, median, and mean relative doses in the CTV and PTV were compared and the differences calculated. Because the minimal target dose alone does not contain any information about the underdosed volume,
Influence of calculation model on dose in pulmonary SRT
the dose coverage of the CTV and PTV by the reference isodose was calculated as:
TC共TV兲 ⫽
V TV.ref , V TV
(1)
where TV denotes either the CTV or PTV (13, 14), VTVref specifies the volume of the target (CTV or PTV) covered by the reference isodose, VTV specifies the target volume itself, and TC (target coverage) refers to those fractions of the CTV or PTV circumscribed by the reference isodose surface (throughout this analysis, the volume included by the 100% isodose surface was defined as the reference volume). The TC was 0 in the case of a total geographic miss of the target and 1 if the CTV/PTV was completely surrounded by the reference isodose. The numbers of MUs calculated by both algorithms for delivery of the prescribed dose to the reference isodose line were also compared. To address the influence of photon energy and the target volume size on the results, the analysis was differentiated by 6-MV vs. 18-MV treatment plans and PTV size. For statistical analysis, Statistica software, version 6.0 (StatSoft), was used. All results were evaluated relative to the PB algorithm. A decrease in dose and TC, together with an increase in MUs, when calculated by the CC-based algorithm, reflects the overestimation tendencies of the PB algorithm. To compare the dose distributions achieved by the PB and CC algorithms to the standard of dose-calculation, MC– based treatment planning was performed for 3 cases. An intermediate target (CTV 21 cm3) located centrally in the lung with its base close to the diaphragm, and the smallest (2 cm3) and largest (256 cm3) CTV were chosen. MC calculations were performed using the Voxel Monte Carlo for photon and electron beams (XVMC) algorithm developed by Fippel et al. from the University of Tu¨bingen, Germany (22–24). The XVMC and its beam model have been demonstrated to reproduce measured (with a diamond detector) relative dose distributions with an accuracy of better than ⫾2% in various homogeneous and inhomogeneous phantoms (23). Because the program version implemented at our department did not allow for DVH computation, the PB, CC, and MC calculated dose distributions (6-MV photons, Siemens Primus) were compared at three representative axial slices at different target levels: the cranial edge of the target, the isocenter level in the central part of the target where the tumor was surrounded by lung tissue only laterally, and the caudal edge of the target. For quantitative comparison of the dose distributions achieved by the different calculation algorithms, dose profiles intersecting the isocenter in the craniocaudal, lateral, and AP directions were generated (see Fig. 6).
RESULTS Comparison of target dose calculated by PB and CC algorithms The comparison of dose distributions by the DVHs computed by the PB and CC algorithms revealed increased inhomogeneity for the CC calculation because of charged particle disequilibrium at the tumor–lung tissue interface. Typically, the volumes covered by the high isodoses levels (100 –140%) shrunk significantly if calculated by the CC algorithm, resulting in reduced target doses (Fig. 1). Although the maximal dose remained relatively unaffected
● U. HAEDINGER et al.
241
(average difference ⫺3.7%; SD 4.7%), the average median (PTV, ⫺13.9% and CTV, ⫺9.2%), mean (PTV, ⫺11.2% and CTV, ⫺9.4%), and minimal (PTV, ⫺5.5% and CTV, ⫺14.9%) doses decreased significantly using the CC algorithm (Table 1). Evaluation of the differences between both algorithms with respect to the PTV size and photon energy revealed that the decrease in the median relative dose was generally greater for 18 MV (mean difference, ⫺16.4%; SD, 5.1%) compared with 6 MV (mean difference, ⫺10.4%; SD, 5.4%). The difference was greater for smaller targets but without a statistically significant correlation to PTV size (18 MV r ⫽ 0.14; 6 MV r ⫽ 0.34; Fig. 2). The initial treatment plans, designed on the basis of the PB algorithm, yielded large values for TC (PTV, average, 96.1%,; SD, 3.1%; CTV, average, 99.9%; SD, 0.3%) independent of the target volume size or photon energy. Minimal values for TC were 84.7% for the PTV and 98.3% for the CTV. Recomputing the treatment plans using the CC model reduced the average PTV coverage by approximately 7%. The differentiation of this effect by photon energy revealed a volume independent (r ⫽ 0.06) decrease in the average PTV coverage of 3.3% for 6 MV (SD 3%). If 18-MV photons were used, dose calculation by the CC algorithm led to a decrease of the average PTV coverage by 9.9% (SD 7%). This effect was most evident in smaller targets (⬍200 cm3; r ⫽ 0.31). Nevertheless, the CTV coverage remained basically unaffected for both algorithms and both photon energies. Large TC values can be achieved at the expense of conformality if the dose distributions are too generous. In a recent analysis, we found that in ESRT for lung targets, highly conformal treatment techniques could be applied, primarily as a result of the relatively simple geometric shape of most pulmonary targets (13). Even as the PTV coverage dropped to a minimal value of 69.5%, the minimal CTV coverage found with CC was 93.2%. This indicates that in the worst case, approximately 7% of the CTV received less than the reference dose because of “shrinkage” of the volume covered by the reference isodose. The results are detailed in Table 1 and Fig. 3.
Comparison of MUs calculated by PB and CC algorithms The computed MUs to achieve the reference dose were generally greater for the CC algorithm. On average, the MUs increased by 5.4% (SD 5.8%) compared with the PB-based calculation. The difference was most prominent for 18-MV photons (mean, ⫹8.4%; SD, 5.5%) compared with 6-MV photons (mean, ⫹1.5%; SD, 3.4%). The dependence of the MU ratio on the PTV size is shown in Fig. 4. Although no correlation was found between PTV size and the number of MUs for 6-MV photons (r ⫽ ⫺0.14), the number of MUs calculated by the CC method for 18-MV treatment plans were significantly greater for smaller targets (e.g., PTV ⬍100 cm3; r ⫽ ⫺0.68). For example, the smallest target in the sample (CTV 2 cm3, PTV 12 cm3) irradiated by an 18-MV dose plan showed the largest difference in calculated MUs (21%).
242
I. J. Radiation Oncology
● Biology ● Physics
Volume 61, Number 1, 2005
Fig. 1. Difference in dose–volume histograms (DVHs) from dose plans (6 MV) calculated by pencil beam (PB) and collapsed cone (CC) algorithms for lung target in right lower lobe (clinical target volume [CTV] 21 cm3; planning target volume [PTV] 42 cm3). Left curves in each DVH from CC calculations; right curves show overestimation of target dose calculated by PB algorithm. In this example (dose distributions shown in Fig. 5), minimal dose in CTV and PTV decreased from 124% and 90% to 103% and 81% (change, ⫺21% and ⫺9%) and median dose decreased from 145% and 137% to 137% and 123% (change, ⫺8% and ⫺14%), respectively, when calculated with CC algorithm.
Comparison of PB and CC algorithms to MC calculations To support the reliability of CC-based dose calculations, MC calculations were performed on 3 cases using 6-MV photons. An example of the axial dose distributions for the intermediate-size target (CTV 21 cm3, PTV 42 cm3) at the isocenter level and 1.5 cm cranial and 1.5 cm caudal of the
isocenter is shown in Fig. 5. The differences are most clearly visible in the dose profiles displayed in Fig. 6. All profiles demonstrated a decrease in the high-dose volume (isodose levels ⬎100%) from the PB to CC to MC algorithm. Although only three examples were calculated, the comparison with the MC calculation indicated that the CC
Fig. 2. Differences in median relative planning target volume (PTV) dose as calculated by pencil beam (PB) and collapsed cone (CC) algorithms. On average, median dose decreased by ⫺13.9% if CC algorithm was used. Differences increased with smaller targets and higher photon energy.
Influence of calculation model on dose in pulmonary SRT
Table 1. Comparison of dosimetric data for CTV and PTV for 33 pulmonary targets as calculated with Helax-TMS using pencil beam model and collapsed cone algorithm
Data PTV coverage Average/SD Range PTV dose (average/ SD) Minimum Maximum Median Mean SD CTV coverage Average/SD Range CTV dose (average/ SD) Minimum Maximum Median Mean SD Monitor units (average/ SD) All targets 6 MV 18 MV
Pencil beam calculation
Collapsed cone calculation
96.1/3.1 84.7/100.0
89.1/8.2 69.5/99.5
80.6/11.4 157.8/5.7 141.3/5.6 136.6/5.2 15.9/2.9 99.9/0.3 98.3/100.0
75.0/9.9 154.1/4.6 127.4/9.7 125.4/7.9 16.7/2.2 99.5/1.3 93.2/100.0
⌬ CC-PB ⫺7.1/6.5
⫺5.5/5.6 ⫺3.7/4.7 ⫺13.9/5.9 ⫺11.2/4.1 0.9/1.8 ⫺0.4/1.0
120.7/12.3 157.8/5.7 148.6/2.5 147.5/2.9 6.2/2.7
105.8/3.3 154.1/4.6 139.4/6.1 138.1/5.8 9.7/2.6
⫺14.9/9.4 ⫺3.7/4.7 ⫺9.2/4.9 ⫺9.4/4.0 3.5/2.5
100 fixed 100 fixed 100 fixed
105.4/5.8 101.5/3.4 108.4/5.5
5.4/5.8 1.5/3.4 8.4/5.5
Abbreviations: CC ⫽ collapsed cone; PB ⫽ pencil beam; PTV ⫽ planning target volume; CTV ⫽ clinical target volume; SD ⫽ standard deviation. Last column shows average difference between both calculation algorithms. Although the median and mean PTV dose decreased 13.9% and 11.2%, respectively, on average, the minimal CTV dose decreased by 14.9% and mean/median CTV doses by almost 10%. The SD of the dose in the CTV increased by 3.5%, on average, indicating a more inhomogeneous dose distribution with the CC model. The monitor units to achieve the reference dose were, on average, 5.4% greater with the CC algorithm (8.4% for targets irradiated with 18-MV photons).
algorithm is more accurate than the PB calculation in the high-dose region and that the PB-based dose calculation might overestimate the therapeutic target dose.
DISCUSSION At different centers, primary and/or metastatic lung tumors are treated using the ESRT approach (1, 2, 4 –10). Dose-escalation studies have been performed to define the appropriate doses for achieving tumor eradication without severe treatment-related side effects (8). It is difficult to compare the results because a variety of dose fractionation schedules and different procedures for dose normalization
● U. HAEDINGER et al.
243
and prescription have been used. Furthermore, the aspect of dose calculation accuracy was often neglected. The presented results showed the largest differences between calculation models in small targets. The median CTV in this study was 58 cm3, considerably larger than the CTV in the reports of ESRT for lung tumors by other authors, which ranged from 4.0 to 22.5 cm3 (1, 2, 4, 6, 8). Under certain circumstances, the use of 18-MV photons in this study was desirable for ESRT to lung tumors, because the higher photon energy was delivered at a higher dose rate by the Siemens Primus linear accelerators at our institution. This factor becomes even more relevant if breathing-triggered RT is used (increase of RT time owing to restriction of radiation to a defined breathing phase). Because of the very small target volumes and the attraction of the higher photon energy, evaluation of the accuracy of dose-calculation algorithms becomes extremely important in ESRT for lung tumors. PB-based algorithms are widely used in computerized treatment planning systems. The limitations of this approach for dose calculation in the case of lung tumors have been reported by Kno¨o¨s et al. (19). Electronic equilibrium is generally disturbed at density inhomogeneities such as at the tumor–lung interface. This results in incorrect predictions of the calculated dose if energy deposition kernels, such as derived from homogeneous media, are used. PB models apply corrections for tissue inhomogeneities in one dimension by scaling the PB kernels in a longitudinal direction. Because MC approaches are exact mathematical models of fundamental physical interaction processes, they can, in principle, predict the dose with arbitrary accuracy. However, these approaches remain slow and are not yet widely accessible clinically. As an alternative, superposition/convolution models have been developed, which take into account tissue inhomogeneities by scaling the “pointlike” energy deposition kernels according to the local density distribution in all three dimensions (a list of references can be found in Butson et al. [25]). To speed up computation time, such a model has been implemented into the TMS in the so-called collapsed cone approach. The energy deposition kernels are approximated as a set of discrete cones, and the energy attenuation and transport is restricted to the cones axis or “collapsed to the cones axis” (the CC implementation in Helax-TMS is extensively described by Ahnesjo¨ [20]). This procedure promises a more reliable prediction of doses in heterogeneous media. Various examinations have shown satisfying agreement between the CC calculations and measurements in phantoms in which slabs of low-density material simulate lung tissue. Butson et al. (25) obtained adequate accuracy within an anthropomorphic lung phantom for 6-MV and 10-MV photon energies. Francescon et al. (26) compared a CC model with the MC code BEAM for mediastinal and breast treatment. Performing DVH analysis in real patient CT data sets, they concluded that CC and MC give absolute comparable results for common 3D-RT planning. Butts and Foster (27) concluded that the differences between diode measurements and the
244
I. J. Radiation Oncology
● Biology ● Physics
Volume 61, Number 1, 2005
Fig. 3. Planning target volume (PTV) coverage by 100% reference isodose plotted vs. PTV size using pencil beam (PB) (open circles, open triangles) and collapsed cone (CC) (solid circles, solid triangles), differentiated by photon energy of 6 MV (triangles) and 18 MV (circles). Although a volume-independent decrease in TC of 3.3% indicated at mean by CC algorithm for 6-MV photons, TC decreased by 9.9% at mean if dose was calculated by CC with 18-MV photons. This effect increased with decreasing PTV size. Difference between photon energies and its relation to target volume was not detected with PB algorithm.
Fig. 4. Ratio of monitor units (MUs) as calculated by collapsed cone (CC) and pencil beam (PB) algorithms. Amount of MUs necessary to achieve planned reference dose was significantly greater when calculated by CC algorithm, leading to ratio of ⬎100% in most cases. Results plotted vs. planning target volume (PTV) size and differentiated by photon energy, showing that MU ratio increased significantly for small targets and 18-MV photon radiation and indicating large dose overestimations by PB algorithm in these cases.
Influence of calculation model on dose in pulmonary SRT
● U. HAEDINGER et al.
245
Fig. 5. Dose distributions (6-MV photons) for target in central lung located above diaphragm (clinical target volume [CTV] 21 cm3; planning target volume [PTV] 42 cm3) as calculated with (a) Monte Carlo (MC) algorithm, (b) collapsed cone (CC) algorithm, and (c) pencil beam (PB) algorithm. Slices located at cranial edge of target (1.5 cm cranial of isocenter, left column), at isocenter level (central column), and base of target (1.5 cm caudal of isocenter, right column). Dose normalized to 150% level at isocenter of treatment plan. Volumes covered by 140%, 120%, and 100% isodoses decreased from PB to CC to MC, indicating that even CC-based calculations might overestimate doses compared with MC calculations. Dose profiles in three main directions shown in Fig. 6.
CC monitor unit calculation in an anthropomorphic phantom were within acceptable limits. In this study, the influence of the dose-calculation algorithms (PB vs. CC implemented in the Helax-TMS treat-
ment planning system) for a significant number (n ⫽ 33) of clinical cases in ESRT for lung tumors using 6-MV (n ⫽ 14) or 18-MV (n ⫽ 19) photons was evaluated. The analysis revealed that the dose at the interface of the tumor and
246
I. J. Radiation Oncology
● Biology ● Physics
Volume 61, Number 1, 2005
Influence of calculation model on dose in pulmonary SRT
● U. HAEDINGER et al.
247
Fig. 6. Comparison of dose profiles (6-MV photons) in craniocaudal (left), lateral (middle), and AP (right) directions calculated by Monte Carlo (MC), collapsed cone (CC), and pencil beam (PB) algorithms for three examples. (a) Largest clinical target volume (CTV) (256 cm3). (b) Intermediate-size target (dose distribution shown in Fig. 5) with clinical target volume (CTV) of 21 cm3. (c) Smallest CTV (2 cm3). Profiles intersect at isocenter (0 cm). Dose was prescribed to PTV enclosing 100% isodose with normalization to 150% at isocenter. Profiles revealed that CC-calculated dose corresponded better to MC-based calculations than did PB calculations, especially in therapeutic dose area of 100 –150%. Proportional effect of PB algorithm dose overestimation increased with decreases in target volume.
low-density lung tissue is considerably overestimated by the PB algorithm, especially in small targets ⱕ100 cm3 irradiated with 18-MV photons. On average, the CC calculations resulted in a 7% reduced PTV coverage. Differentiated by photon energy, the average PTV coverage decreased by only 3.3% for 6-MV but 9.9% for 18-MV photons. The comparison with MC calculations showed that the CC algorithm is more reliable than the PB algorithm, especially in the high-dose area between the 100% and the 150% isodose level. In relation to the total target volume, this effect increased with a decreasing target volume because of the larger interface to lung tissue. Therefore, a statistically significant overestimation of dose (average deviation of median dose ⫺13.9%) and PTV TC (average difference ⫺9.9%) and underestimation of MUs (average difference 8.4%) using the PB calculation must be considered if small target volumes are irradiated and dose prescription is performed to a PTV enclosing isodose instead of to the isocenter. The effect is further increased with the use of highenergy photons (e.g., 18 MV). No decrease in average dose coverage to the CTV was found. This indicated that the margins applied in the PTV definition to allow for target positioning/movement uncertainties—not for dose-calculation uncertainties—were sufficient to ensure CTV TC under the condition of high geometric accuracy of treatment delivery. In a recent analysis of the impact of target reproducibility on tumor dose in ESRT (calculations based on a PB model) it was shown, by matching studies from CT simulation and treatment planning, that in 91% of the lung targets CTV dose coverage ⬎95% was achieved (14). In the present study, a statistically significant reduction of the median and mean relative doses to the CTV (average reduction median ⫺9.2% and mean ⫺9.4%) and PTV (median ⫺13.9% and mean ⫺11.2%) in the CC calculation was found, indicating that about 10% less dose was delivered to the CTV than intended, even under optimal geometric conditions. Our results expand on the work of other authors, which was based on a range of clinical situations comparing different dose-calculation approaches and measurements in phantoms. Engelsman et al. (28) examined the impact of simple tissue-inhomogeneity correction algorithms on conformal RT for lung tumors in a specially designed lung phantom with an AP irradiation geometry. Using film measurements, they found that the field margins will be underestimated by PB algorithms. The actual energy-dependent broadening of the beam penumbra is not correctly described by the PB algorithm of TMS. They reported values for PTV
underdosage (with respect to the planned 95% isodose level) of between 10% and 15% for low energy beams (6 MV) and ⬎20% for higher energies. The authors stated that the underdosage found for their PTV did not necessarily lead to an underdosage of the gross tumor volume or CTV. This was also reflected in our data. Even if the PTV dose coverage drops when using the CC method, the difference in the planned CTV coverage was not statistically significant. From this point of view, there seems to be no strong influence of the dose calculation model because neither the average nor the minimally detected CTV coverage changed dramatically in our evaluation. Nevertheless, the PTV is defined to cover uncertainties in the CTV reproducibility due to setup inaccuracy and breathing mobility. It is clinically reasonable to expect the CTV to be located at the edge of the PTV during treatment. In this case, a decrease in PTV dose coverage would correspond to an equally reduced dose to the CTV. Also, although the average CTV coverage did not decrease within this study, the median and mean doses in the CTV decreased by approximately 10%, reducing the intended dose to the CTV. The increased influence of this effect was clearly demonstrated with decreasing target size. Although the clinical impact of the additional target dose achieved by the inhomogeneous dose distribution is still unclear, a considerable decrease of the dose to the CTV must be taken into account when the PB algorithm is used. Extracranial stereotactic radiotherapy is not only performed as hypofractionated treatment—such as in the present study— but also as single-dose treatment (2, 4, 6). In contrast to 3D-conformal RT in multiple fractions, underdosage of a certain proportion of the target because of setup inaccuracy or target mobility will not be compensated for by random error during the treatment course. Reliable dose calculation is then of extreme importance, primarily to avoid local failures and secondarily to allow for interpretation of failures (e.g., due to target miss or insufficient setup accuracy, target definition, or target dose). Several studies have been started to establish the most appropriate dose for ESRT. If dose calculation is possibly inaccurate, as demonstrated in this evaluation, the interpretation of doses from these studies will be difficult. Therefore, dose-calculation algorithms, as well as the photon energy used for treatment planning, should be reported. The influence of photon energy on the target dose has been evaluated by other authors using MC calculations. Saitoh et al. (29) examined the dose distribution for narrow beam irradiation of a lung phantom by comparing Electron Gamma Shower Code Version 4 (EGS4) MC simulation
248
I. J. Radiation Oncology
● Biology ● Physics
and dosimetric films and proved the broadening of dose profiles at the edges of the tumor owing to the lack of scattering material. They concluded that the variations in relative dose distributions and beam profiles change as a function of incident photon energy. Because the dispersion of absorbed energy in tumors increases with increased photon energy, they recommended the use of low-photon energies for stereotactic treatment of lung tumors. Papiez (30) studied the influence of photon energy, target size and location, tissue density, and complexity of the treatment plan (number of beams) in a mathematical model of the human chest using the PENELOPE MC code. He reported a significant dose reduction at the tumor edge (15–20% for 6 MV X-rays and 25–50% for 15 MV X-rays) that was not correctly described by commercial treatment planning systems with semiempiric beam attenuation correction or firstorder scatter corrections. Linthout et al. (31) studied the effect of the dose-calculation model for dynamic arc treatments in a humanoid head-and-neck phantom. They found discrepancies between the calculated dose at the interface of tumor–air cavities and measurements with dosimetric film, as well as TLD. In their study, the PB algorithm overestimated the dose in the PTV at the tissue–air boundaries by approximately 10%. Differences were smaller for a CC algorithm. Verellen et al. (32) reported similar results. Martens et al. (33) compared radiochromic film measurements, MC simulations, and CC calculations for intensity-modulated RT in a phantom simulating the air cavity of the trachea. They obtained very good agreement between the film measurement and MC, but the CC algorithms were not able to predict the interface dose correctly. The CC calculations showed considerable improvements compared with the PB models. In addition to the differences in relative dose distributions, large discrepancies in the numbers of MUs calculated with PB or CC algorithms can emerge. This usually happens if the dose is prescribed to a PTV-enclosing isodose (i.e., in a region in which secondary charged particle equilibrium is not maintained). In other methods that prescribe the dose to the International Commission on Radiation Units and Measurements reference point in the center of the PTV ade-
Volume 61, Number 1, 2005
quately surrounded by tumor tissue, the accuracy of the PB calculation of MUs is sufficiently high (28). As a result of this evaluation and the experiences of other authors, we now use only the CC algorithm of the HelaxTMS treatment planning system and 6-MV photons for stereotactic RT to pulmonary targets. The dose is delivered to the patient using the absolute MUs computed by the CC algorithm. To compensate for the reduction in the median and mean relative target dose, the beam widths have to be enlarged by adding an extra “penumbra broadening margin” of 3–5 mm to the beams originally designed with the interactive PB-based algorithm. Another approach is an additional, small-size beam designed to increase the dose in the “underdosed” part of the target. Alternatively, the dose normalization level could be raised from 150% to 160%, because the maximal dose is located at the center of the tumor and will not harm the patient as long as the dose is delivered precisely to the target. CONCLUSION The issue of dose-calculation is important for ESRT for targets of the lung, especially when various clinical studies are compared. Because of their intrinsic limitations, the present, widely used PB algorithms overestimate the absorbed dose at the interface of tumor and low-density lung tissue. Therefore, the more reliable CC models should be used in these cases, as long as MC calculation is not available in treatment planning systems for routine clinical use. When the dose is prescribed to the edge of the target instead of the International Commission on Radiation Units and Measurements reference point in the center of the tumor, the numbers of MUs calculated by PB algorithms can be considerably too low. This is especially the case for small targets with a PTV ⱕ100 cm3 and high-energy photons. In these cases, use of a PB algorithm can result in an overestimation of the doses at the reference isodose level of up to 20%. In a 3 ⫻ 10-Gy concept, this would be equivalent to a decrease in the total dose of 6 Gy. Additional systematic studies exploring the influence of photon energy, the specific target location within the lung, and the potential of MC-based dose calculation algorithms are needed.
REFERENCES 1. Blomgren H, Lax I, Go¨ranson H, et al. Radiosurgery for tumors in the body: Clinical experience using a new method. J Radiosurg 1998;1:63–74. 2. Hara R, Itami J, Kondo T, et al. Stereotactic single high dose irradiation of lung tumors under respiratory gating. Radiother Oncol 2002;63:159–163. 3. Herfarth KK, Debus J, Lohr F, et al. Stereotactic single dose radiation therapy of liver tumors: Results of a phase I/II trial. J Clin Oncol 2001;19:164–170. 4. Hof H, Herfahrt KK, Munter M, et al. Stereotactic single-dose radiotherapy of stage I non-small-cell lung cancer (NSCLC). Int J Radiat Oncol Biol Phys 2003;56:335–341. 5. Nagata Y, Negoro Y, Aoki T. Clinical outcomes of 3D conformal hypofractionated single high-dose radiotherapy for one
or two lung tumors using a stereotactic body frame. Int J Radiat Oncol Biol Phys 2002;52:1041–1046. 6. Nakagawa Y, Aoki Y, Tago M, et al. Megavoltage CTassisted stereotactic radiosurgery for thoracic tumors: Original research in the treatment of thoracic neoplasms. Int J Radiat Oncol Biol Phys 2000;48:449–457. 7. Onimaru R, Shirato H, Shimizu S, et al. Tolerance of organs at risk in small-volume, hypofractionated, image-guided radiotherapy for primary and metastatic lung. Int J Radiat Oncol Biol Phys 2003;56:126–135. 8. Timmerman RD, Papiez L, McGarry R. Extracranial stereotactic radioablation: Results of a phase I study in medically inoperable stage I non-small cell lung cancer. Chest 2003;124: 1946 –1955.
Influence of calculation model on dose in pulmonary SRT
9. Uematsu M, Shioda M, Suda A, et al. Computed tomographyguided frameless stereotactic radiotherapy for stage I nonsmall cell lung cancer: A 5-year experience. Int J Radiat Oncol Biol Phys 2001;51:666–670. 10. Wulf J, Haedinger U, Oppitz U, et al. Stereotactic radiotherapy of targets in the lung and liver. Strahlenther Onkol 2001; 177:645–655. 11. Lax I, Blomgren H, Larson D, et al. Extracranial stereotactic radiosurgery of localized targets. J Radiosurg 1998;1:135– 148. 12. Wulf J, Haedinger U, Oppitz U, et al. Stereotactic radiotherapy of extracranial targets: CT-simulation and accuracy of treatment in the stereotactic body frame. Radiother Oncol 2000;57:225–236. 13. Haedinger U, Thiele W, Wulf J. Extracranial stereotactic radiotherapy: Evaluation of PTV coverage and dose conformity. Z Med Phys 2002;12:221–229. 14. Wulf J, Haedinger U, Oppitz U, et al. Impact of target reproducibility on tumor dose in stereotactic radiotherapy of targets in the lung and liver. Radiother Oncol 2003;66:141–150. 15. International Commission on Radiation Units and Measurements. ICRU report 50: Prescribing, recording and reporting photon beam therapy. Bethesda: International Commission on Radiation Units and Measurements; 1993. 16. Lax I. Target dose versus extra-target dose in stereotactic radiosurgery. Acta Oncol 1993;32:453–457. 17. Ahnesjo¨ A. A pencil beam model for photon dose calculation. Med Phys 1992;19:263–273. 18. Helax AB. Dose formalism and methods in Helax-TMS: Manual. 1998. 19. Kno¨o¨s T, Ahnesjo¨ A, Nilsson P, et al. Limitation of a pencil beam approach to photon dose calculation in lung tissue. Phys Med Biol 1995;40:1411–1420. 20. Ahnesjo¨ A. Collapsed cone convolution of radiant energy for photon dose calculation in heterogeneous media. Med Phys 1989;16:577–592. 21. Corminboeuf F, Ray JY, Bieri S. Performance evaluation of the collapsed-cone calculation on Helax-TMS. Kerzers: Verlag; Max Huber 2001, p. 128 –134. 22. Fippel M. Fast Monte Carlo dose calculation for photon beams based on the VMC electron algorithm. Med Phys 1999;26: 1466–1475.
● U. HAEDINGER et al.
249
23. Fippel M, Laub W, Huber B, et al. Experimental investigation of a fast Monte Carlo photon beam dose calculation algorithm. Phys Med Biol 1999;44:3039–3054. 24. Fippel M, Nuesslin F. Verification of a fast Monte Carlo dose calculation algorithm by EGSnrc using the statistical separation method. Z Med Phys 2001;11:152–160. 25. Butson MJ, Elferink R, Cheung T, et al. Verification of lung dose in an anthropomorphic phantom calculated by the collapse cone convolution method. Phys Med Biol 2000;45:143– 149. 26. Francescon P, Cavedon C, Reccanello S, et al. Photon dose calculation of a three-dimensional treatment planning system compared to the Monte Carlo code BEAM. Med Phys 2000; 27:1579–1587. 27. Butts JR, Foster AE. Comparison of commercially available three-dimensional treatment planning algorithms for monitor unit calculations in the presence of heterogeneities. J Appl Clin Med Phys 2001;2:32–41. 28. Engelsman M, Damen EMF, Koken PW, et al. Impact of simple tissue inhomogeneity correction algorithms on conformal radiotherapy of lung tumors. Radiother Oncol 2001;60: 299–309. 29. Saitoh H, Fujisaki T, Sakai R, et al. Dose distribution of narrow beam irradiation for small lung tumor. Int J Radiat Oncol Biol Phys 2002;53:1380–1387. 30. Papiez L. Dosimetry of solid tumors in the lung and radioablative therapy. In: Proceedings of the Second Annual Extracranial Stereotactic Radioablation: Future Directions, June 6 – 8, 2003. Halifax;2003. p. 32. 31. Linthout N, Verellen D, van Acker S, et al. Evaluation of dose calculation algorithms for dynamic arc treatments of head and neck tumors. Radiother Oncol 2002;64:85–95. 32. Verellen D, Linthout N, van den Berge D, et al. Initial experience with intensity-modulated conformal radiation therapy for treatment of the head and neck region. Int J Radiat Oncol Biol Phys 1997;39:99–114. 33. Martens C, Reynaert N, De Wagter C, et al. Underdosage of the upper-airway mucosa for small fields as used in intensitymodulated radiation therapy: A comparison between radiochromic film measurements, Monte Carlo simulations and collapsed cone convolution calculations. Med Phys 2002;57: 1528–1535.