S1621
Physics - Autosegmentation
ESTRO 2026
applied to expand CTV to PTV.Dosimetric evaluation was performed fraction-by-fraction using the dose guidance:PTV coverage: D98% ≥ 95% Rectum: V41Gy < 10% (most critical), V38Gy < 15%, V33Gy < 30%Bladder: mean dose Dmean < 34 GyStatistical analysis used the Wilcoxon signed-rank test for paired comparisons. AI- based plans were assessed for non-inferiority relative to GT plans for OOI constraints. Results: The interquartile range of Dice for AI segmentations was 83–89% (CTV), 92–95% (bladder), and 84–87% (rectum). Dosimetric analysis showed an apparent trend: plans based on AI segmentations with Dice ≥ 91% for PTV met coverage requirements, see figure 1. AI-based plans had significantly lower PTV coverage than GT plans but were non-inferior for OAR doses, with all AI plans meeting all OOI criteria. AI-derived CTV volumes were generally smaller than GT ones, explaining reduced PTV coverage and consistent OOI compliance.
Conclusion: AI-based segmentation can support clinically
acceptable automated online adaptive planning for MR-linac prostate workflows if accuracy thresholds are met. A Dice coefficient >91% for PTV appears necessary for adequate coverage while OOI criteria were always met. These findings highlight AI segmentation’s potential to streamline MR-linac workflows, reduce treatment time, and advance toward fully automated adaptive radiotherapy. Keywords: Adaptive, Segmentation, automation
Made with FlippingBook - Share PDF online