- Email: [email protected]
EP-1537: Iterative dataset optimization in automated planning: implementation for breast radiotherapy
S826 ESTRO 36 _______________________________________________________________________________________________
spared by the CAA, which decreased by 0.62% from the CA0. Minimizing the x-jaws significantly reduced the number of split field from 61 to 37. In every field tested the CAX optimization produced as good or superior results than the other three techniques. For aspherical PTVs, CAX on average reduced the number of split fields, the maximum dose, minimized the dose to the surrounding OAR, and reduced the MU all while achieving the same control of the PTV. Conclusion For aspherical lesions larger than 100 cc, rotating the collimator to minimize the x-jaw gap produced equal tumor control while reducing the toxicity to the organs at risk with lower monitor units and less split fields compared to keeping the collimator fixed at 0º or with using the Eclipse collimator optimization method. Compared to the fixed collimator angle, the monitor units decreased an average of 6% using a CAX approach, which was the lowest of the four methods tested. The maximum dose to the organs at risk also showed trends of decreasing, as well as evidence to decrease the peripheral dose. The number of split fields was highly controlled with CAX by optimizing the parameter that determines if a field will divide. Of the 20 cases studied, the number of split fields decreased by about 40% with the CAX from any other method used. The CAX optimization allowed for a rotation of the collimator between each field, which showed positive results in the overall dose shape of the delivered quality assurance tests on an electronic portal imaging device. EP-1537 Iterative dataset optimization in automated planning: implementation for breast radiotherapy J. Fan1, J. Wang1, Z. Zhang1, W. Hu1 1 Fudan University Cancer Hospital, Department of radiation oncology, Shanghai, China Purpose or Objective To develop a novel automated treatment planning solution for the breast radiotherapy. Material and Methods An automated treatment planning solution developed in this study includes selection of the optimal training dataset, dose volume histogram (DVH) prediction for the organs at risk (OARs) and automatically generation of the clinically acceptable treatment plans. The optimal training dataset was selected by using an iterative optimization strategy from 40 treatment plans for breast cancer patients who received radiation therapy. Firstly, the 2D KDE algorithm was applied to predict OAR DVH curves, including the heart and left lung, for patients in group A by considering the other group B as the training dataset. New plans in group A were automatically generated using the Pinnacle3 Auto-Planning (AP) module (version 9.10, Philips Medical Systems) based on the dose constrains derived from the predicted DVHs. Next the point-wise comparison, taking V5, V20 and mean value as the criteria, between the automatic plans and original clinical plans was performed both objectively and subjectively. Finally the preferred plans in group A got updated after the comparison and were used for the next iteration by considering itself as the training dataset instead. Above steps repeated until search and update for new preferred plans was exhausted. After selecting the optimal training dataset, additional 10 new breast treatment plans were re-planned using the AP module with the objective functions derived from the predicted DVH curves. These automatically generated re-optimized treatment plans were compared with the original manually optimized plans. Results The proposed new iterative optimization strategy, shown in Fig. 1, could effectively select the optimal training dataset and improve the accuracy of the DVH prediction.
The average of mean dose of the OARs in the iterative process for each group, group A and group B are illustrated in Fig. 2. The dose differences, between the real and prediction, decreased with iterations which indicated the convergence of our proposed technique. As can be seen from Tab. 1, the automatically AP generated treatment plans using the dose constrains derived from the predicted DVHs could achieve better dose sparing for some OARs with the other comparable plan qualities. Conclusion The proposed novel automated treatment planning solution can be used to efficiently evaluate and improve the quality and consistency of the treatment plans for modulated breast radiation therapy. EP-1538 VMAT craniospinal radiotherapy, planning strategy and results in twenty pediatric and adult patients F. Lliso1, V. Carmona1, J. Gimeno1, B. Ibañez1, J. Bautista1, J. Bonaque1, R. Chicas1, J. Burgos1, J. Perez-Calatayud1 1 Hospital Universitario La Fe, Radiotherapy Department, Valencia, Spain Purpose or Objective To describe our VMAT craniospinal radiotherapy planning strategy, to report the dosimetric results for the first 20 patients treated with RapidArc and to compare with previously published data. Material and Methods Patients were treated in supine position in Varian Clinac iX linacs (Millennium 120 MLC, OBI) and 6 MV RapidArc, prescription doses (Dprescr) were 23.4, 30.6 and 36 Gy. Twelve patients were children (3-13 years, avg.: 7.2) and nine adults (23-71 years; avg.: 38.8), the resulting PTV avg. length was 64.8 cm (47-82 cm). The treatment planning was performed with Eclipse® (V 13.0). Depending on the PTV length, 2 or 3 isocenters were used with an overlapping region of about 4cm, all coordinates being equal except the longitudinal one. Two complete arcs were applied in the cranial isocentre, one of them encompassing the cranium plus the superior part of the spinal region, and the other one intended to improve conformity and optic sparing, only encompassing cranium. For the spine, one full arc per isocentre was employed except in 36 Gy cases for which a partial posterior arc was added at lungs level. Collimator angles were set to ±5º except for the second cranial arc (40º). Plans were optimized using Progressive Resolution Optimizer (PROII) and calculated with AAA algorithm, the main were goals that at least 95% of the PTV received Dprescr and also the OAR sparing. The planning objectives were defined at the first step of the optimization. Firstly, optimization weights to PTV, Normal Tissue, lenses and lungs were assigned and, once DVH values were close to the desired ones, the rest of the surrounding OARs were sequentially included (mean doses were employed); in addition, for pediatric patients, homogeneous irradiation of the vertebrae was required; finally, weights to maximum dose to OAR located inside the brain PTV were set to avoid hot spots. After that, the rest of the optimization was carried out using the automatic “Intermediate Dose Calculation” option. Various dosimetric parameters and indices were employed: PTV mean dose; D1cc, D2% and D98%; mean and maximum doses for OAR; V5 and V20 for lungs; conformity (CI=Vol95%/VolPTV) and homogeneity (HI=D1cc/Dprescr) indices; Normal Tissue mean and V5 dose. Results We found avg. values for CI, 1.02 (0.98-1.09) and for HI, 1.08 (1.05-1.10) regardless of Dprescr. OAR mean dose values along with average values reported by different authors are shown in Fig. 1. A great reduction is observed for almost all OAR for 23.4 Gy, while for 36 Gy our results are favorable for eyes and lenses and similar or slightly
spared by the CAA, which decreased by 0.62% from the CA0. Minimizing the x-jaws significantly reduced the number of split field from 61 to 37. In every field tested the CAX optimization produced as good or superior results than the other three techniques. For aspherical PTVs, CAX on average reduced the number of split fields, the maximum dose, minimized the dose to the surrounding OAR, and reduced the MU all while achieving the same control of the PTV. Conclusion For aspherical lesions larger than 100 cc, rotating the collimator to minimize the x-jaw gap produced equal tumor control while reducing the toxicity to the organs at risk with lower monitor units and less split fields compared to keeping the collimator fixed at 0º or with using the Eclipse collimator optimization method. Compared to the fixed collimator angle, the monitor units decreased an average of 6% using a CAX approach, which was the lowest of the four methods tested. The maximum dose to the organs at risk also showed trends of decreasing, as well as evidence to decrease the peripheral dose. The number of split fields was highly controlled with CAX by optimizing the parameter that determines if a field will divide. Of the 20 cases studied, the number of split fields decreased by about 40% with the CAX from any other method used. The CAX optimization allowed for a rotation of the collimator between each field, which showed positive results in the overall dose shape of the delivered quality assurance tests on an electronic portal imaging device. EP-1537 Iterative dataset optimization in automated planning: implementation for breast radiotherapy J. Fan1, J. Wang1, Z. Zhang1, W. Hu1 1 Fudan University Cancer Hospital, Department of radiation oncology, Shanghai, China Purpose or Objective To develop a novel automated treatment planning solution for the breast radiotherapy. Material and Methods An automated treatment planning solution developed in this study includes selection of the optimal training dataset, dose volume histogram (DVH) prediction for the organs at risk (OARs) and automatically generation of the clinically acceptable treatment plans. The optimal training dataset was selected by using an iterative optimization strategy from 40 treatment plans for breast cancer patients who received radiation therapy. Firstly, the 2D KDE algorithm was applied to predict OAR DVH curves, including the heart and left lung, for patients in group A by considering the other group B as the training dataset. New plans in group A were automatically generated using the Pinnacle3 Auto-Planning (AP) module (version 9.10, Philips Medical Systems) based on the dose constrains derived from the predicted DVHs. Next the point-wise comparison, taking V5, V20 and mean value as the criteria, between the automatic plans and original clinical plans was performed both objectively and subjectively. Finally the preferred plans in group A got updated after the comparison and were used for the next iteration by considering itself as the training dataset instead. Above steps repeated until search and update for new preferred plans was exhausted. After selecting the optimal training dataset, additional 10 new breast treatment plans were re-planned using the AP module with the objective functions derived from the predicted DVH curves. These automatically generated re-optimized treatment plans were compared with the original manually optimized plans. Results The proposed new iterative optimization strategy, shown in Fig. 1, could effectively select the optimal training dataset and improve the accuracy of the DVH prediction.
The average of mean dose of the OARs in the iterative process for each group, group A and group B are illustrated in Fig. 2. The dose differences, between the real and prediction, decreased with iterations which indicated the convergence of our proposed technique. As can be seen from Tab. 1, the automatically AP generated treatment plans using the dose constrains derived from the predicted DVHs could achieve better dose sparing for some OARs with the other comparable plan qualities. Conclusion The proposed novel automated treatment planning solution can be used to efficiently evaluate and improve the quality and consistency of the treatment plans for modulated breast radiation therapy. EP-1538 VMAT craniospinal radiotherapy, planning strategy and results in twenty pediatric and adult patients F. Lliso1, V. Carmona1, J. Gimeno1, B. Ibañez1, J. Bautista1, J. Bonaque1, R. Chicas1, J. Burgos1, J. Perez-Calatayud1 1 Hospital Universitario La Fe, Radiotherapy Department, Valencia, Spain Purpose or Objective To describe our VMAT craniospinal radiotherapy planning strategy, to report the dosimetric results for the first 20 patients treated with RapidArc and to compare with previously published data. Material and Methods Patients were treated in supine position in Varian Clinac iX linacs (Millennium 120 MLC, OBI) and 6 MV RapidArc, prescription doses (Dprescr) were 23.4, 30.6 and 36 Gy. Twelve patients were children (3-13 years, avg.: 7.2) and nine adults (23-71 years; avg.: 38.8), the resulting PTV avg. length was 64.8 cm (47-82 cm). The treatment planning was performed with Eclipse® (V 13.0). Depending on the PTV length, 2 or 3 isocenters were used with an overlapping region of about 4cm, all coordinates being equal except the longitudinal one. Two complete arcs were applied in the cranial isocentre, one of them encompassing the cranium plus the superior part of the spinal region, and the other one intended to improve conformity and optic sparing, only encompassing cranium. For the spine, one full arc per isocentre was employed except in 36 Gy cases for which a partial posterior arc was added at lungs level. Collimator angles were set to ±5º except for the second cranial arc (40º). Plans were optimized using Progressive Resolution Optimizer (PROII) and calculated with AAA algorithm, the main were goals that at least 95% of the PTV received Dprescr and also the OAR sparing. The planning objectives were defined at the first step of the optimization. Firstly, optimization weights to PTV, Normal Tissue, lenses and lungs were assigned and, once DVH values were close to the desired ones, the rest of the surrounding OARs were sequentially included (mean doses were employed); in addition, for pediatric patients, homogeneous irradiation of the vertebrae was required; finally, weights to maximum dose to OAR located inside the brain PTV were set to avoid hot spots. After that, the rest of the optimization was carried out using the automatic “Intermediate Dose Calculation” option. Various dosimetric parameters and indices were employed: PTV mean dose; D1cc, D2% and D98%; mean and maximum doses for OAR; V5 and V20 for lungs; conformity (CI=Vol95%/VolPTV) and homogeneity (HI=D1cc/Dprescr) indices; Normal Tissue mean and V5 dose. Results We found avg. values for CI, 1.02 (0.98-1.09) and for HI, 1.08 (1.05-1.10) regardless of Dprescr. OAR mean dose values along with average values reported by different authors are shown in Fig. 1. A great reduction is observed for almost all OAR for 23.4 Gy, while for 36 Gy our results are favorable for eyes and lenses and similar or slightly