ESTRO 38 Abstract book

S222 ESTRO 38

LRPM generated plans were compared. To steer towards relevant trade-offs between all criteria for all training patients, the user specifies preferences for each criterion. E.g., for prostate cancer, the preferences encourage reducing high rectum dose and mean bladder dose, while retaining similar PTV coverage and allowing slight degradation in mean anus dose. The quality of the automatic LRPM configurations was tested by comparing database plans with LRPM generated plans for 197 prostate cancer evaluation patients (not used for training) and 75 HN evaluation patients. Results With the automatically generated LRPM configurations, all automatically LRPM generated plans had sufficient PTV coverage (V95%≥99% for prostate, V95%≥98% for HN). For the evaluation patients, differences in OAR doses between database plans and LRPM generated plans (both Pareto- optimal) are shown in fig. 1 (prostate) and fig. 2 (HN) using boxplots. Medians of criteria differences marked with * were statistically significantly different from zero (paired two-sided Wilcoxon signed-rank test, p<0.05). For prostate cancer, the LRPM generated plans had on average a decreased high rectum dose at the cost of slight increases in mean anus dose and maximum dose to the femoral heads (while still within constraints). For HN, we found more variation: generally, the LRPM generated plans had large improvements for some criteria at the cost of slight degradations for other criteria.

Figure 1. DVH comparison between the deliverable plan (solid lines) and the navigated dose (dotted lines). Conclusion A form of multi-criteria navigation for VMAT that gives only minor or negligible errors between the deliverable plan and the navigated dose has been demonstrated. A highly realistic navigated dose has the potential to simplify clinical decision making and improve planning efficiency thanks to a decreased or eliminated need for manual tuning after the conversion of the navigated dose into a deliverable plan. PV-0425 Automated configuration of an algorithm for fully automated Pareto-optimal treatment planning R. Van Haveren 1 , B. Heijmen 1 , S. Breedveld 1 1 Erasmus Medical Center Rotterdam Daniel den Hoed Cancer Center, Radiation Oncology, Rotterdam, The Netherlands Purpose or Objective The Lexicographic Reference Point Method (LRPM) is a novel algorithm for automated, fast and consistent generation of clinically favourable Pareto-optimal treatment plans. Current configuration of the LRPM requires manual tweaking of the algorithm’s parameters. This is a labour-intensive and time-consuming trial-and- error procedure. We propose a novel configuration method to automatically tune the algorithm’s parameters based on historical treatment plans. We demonstrate the method for prostate, and head and neck (HN) cancer. Material and Methods The LRPM for fast automated multi-criterial treatment planning has previously been introduced for generating Pareto-optimal plans with clinically favourable trade-offs between all plan criteria. The proposed novel configuration method generates a single LRPM configuration per treatment site. Validation was performed with patient/plan databases for prostate cancer (287 plans) and HN cancer (105 plans). All database plans were generated with an in-house developed clinically applied algorithm for automated multi-criterial planning (other than LRPM). Configuration of the LRPM was based on 90 training plans for prostate cancer and 30 plans for HN cancer. For each criterion, differences between the training plans and