ESTRO 2020 Abstract book

S816 ESTRO 2020

interest located within the target (mean dose difference of ±3%). Outside the target, a maximum dose difference of 19% was observed for RS-PB in the presence of the bone- lung inserts (Figure 1). The RS-MC revealed much better agreement outside the target for the same configuration, with a maximum dose difference of 5%. For the oblique beam, maximum dose differences of 12.2% and 5.8% were observed for the RS-PB and RS-MC, respectively. Both dose calculations performed well, in the presence of the range shifter, with the maximum dose difference of 2%. A summary of all results is given in Figure 2.

PO-1441 A novel, fast and accurate 3D-dose-based MLC segmentation for robotic radiotherapy B. Schipaanboord 1 , S. Breedveld 1 , B. Heijmen 1 1 Erasmus MC Cancer Center, Radiation Oncology, Rotterdam, The Netherlands Purpose or Objective A novel, fast segmentation method was developed for converting fluence-map optimized (FMO) dose distributions into deliverable treatment plans. Segment dose calculations fully accounted for MLC scatter effects. Material and Methods The applied column generation algorithm converts an input dose distribution into a deliverable plan by iteratively identifying segments and adding them to the treatment plan. For accurate dose modeling, the clinical dose engine (CDE) from the CyberKnife (Accuray Inc.) was integrated into the segmentation. Two ways for integrating the CDE were investigated. In the FAST method, segment doses were calculated as sums of the individual pencil-beams during segmentation, followed by a full segment dose re-calculation with the CDE at the end of the segmentation and re-optimization of the segment weights. The SLOW method uses the CDE in each iteration of the segmentation to accurately model the segment dose. A configurable MU penalty term was implemented for balancing between plan quality and treatment time. Pareto-optimal FMO plans were created and subsequently segmented with the proposed methods. Segmentation performance was evaluated for 32 patients (12 with liver cancer, 20 with prostate cancer) treated with SBRT. To visualize trade-offs between plan quality and treatment time, segmentations were performed for increasing MU penalty weights (w=0.1, 0.25, 0.5, 0.75, 1.0). Plan quality was in part evaluated with the Plan Quality Index (PQI, Fig. 2), defined as the mean difference in achieved criterion values between an FMO plan and its corresponding segmented plan. Results Fig. 1 shows dosimetric plan parameter values of the segmented plans for the liver patients, plotted against the values for the corresponding FMO plans (w=0.5). All plans (FMO, FAST and SLOW) met all constraints. Differences between FAST and SLOW plans were not statistically significant (p>0.05) or were clinically irrelevant (<0.5 Gy). Similar results were obtained for prostate cancer. Fig. 2 shows clear trade-offs between the PQI and the number of segments or MU. Differences between FAST and SLOW in PQI were again clinically irrelevant. Identifications of the segments for the liver patients with the FAST method took on average 0.7±0.1 min., while an extra 11.5±3.0 min. was needed for the final CDE dose calculations. The SLOW method took for liver on average 49.7±14.0 min. For the prostate patients, these times were on average 1.1±0.1 min. and an extra 11.8±2.0 min. for CDE dose calculations for the FAST method and 43.3±8.3 min on average for the SLOW method. Since segments can be iteratively replaced by better ones during segmentation, more segments are generated than used in the final plan. Therefore, the SLOW method has a larger overhead for CDE dose calculations.

Conclusion In the treated volume, the performance of the two dose calculation algorithms (RS-PB and RS-MC) was comparable for different depths and in the presence of the investigated tissue interfaces. However, significant dose differences were observed outside the target where the RS-PB under-predicted the dose. This might lead to an increased risk of side effects for closeby organs or even to higher toxicities than anticipated. Consequently, for the TPS involved in the study, the RS-MC is recommended when dealing with complex treatment geometries.