S1565
Physics - Autosegmentation
ESTRO 2026
Results: The TL models improved performance across all metrics and observers versus the baseline UNet, making predictions more representative of each observer. The 14-100% band resulting from the probabilistic maps were larger at cranial-caudal and lateral borders, consistent with expected IOV, with average volumetric difference of 193±103 cm3. Predictions applied to training data reached average ECDF of 0.9 at 9.3%±3.4 (under segmentation residuals), 10.4%±3.5 (over segmentation residuals) and 16.5%±1.3 (residual sum). Expert contours fell well within the predicted 14-100% band (Fig. 2A), while residents’ contours deviated significantly, mainly at the cranial-caudal axis, confirming model’s sensitivity to contouring expertise (Fig. 1B). Planning results showed clinical lung and heart DVHs always falling within the 14-100% variability band (Fig. 2 B).
Conclusion: This TL approach enables the generation of robust and reliable probabilistic segmentation maps, offering direct IOV impact estimate, together with the mean auto-contour. Current study confirms that the model well captures IOV in a test cohort with multiple observers delineating the same patients. Applications include clinical trials QA, contour quality assessment, tutoring and education. Supported by CCM 2024 (Ministry of Health). References: [1] Ubeira-Gabellini M. G., Palazzo G., et al. “Development and external multicentric validation of a deep learning-based clinical target volume segmentation model for whole-breast radiotherapy”. Physics and Imaging in Radiation Oncology 34 (Apr. 2025), p. 100749. doi: 10.1016/j.phro.2025.100749. Keywords: breast, radiotherapy, CTV probability map Digital Poster 2230 Pancreatic tumour auto-segmentation for online MRI-guided radiotherapy Emilie Helgesen Karlsson 1,2 , Smith Khare 3 , Uffe Bernchou 4,2 , Faisal Mahmood 1,2 1 Laboratory of Radiation Physics, Department of Oncology, Odense University Hospital, Odense, Denmark. 2 Department of Clinical Research, University of Southern Denmark, Odense, Denmark. 3 Applied AI and Data Science, University of Southern Denmark, Odense, Denmark. 4 Laboratory of Radiation Physics, Department of Oncology, Odense, Odense, Denmark
Made with FlippingBook - Share PDF online