ESTRO 2025 - Abstract Book

S2430

Physics - Autosegmentation

ESTRO 2025

Conclusion: The proposed post-training uncertainty calibration method outputs geometrically accurate DLAS models with well calibrated uncertainty estimates at the individual patient level, essential for the clinical use of uncertainty maps. This method is applicable to other ART scenarios requiring contour adaptation, such as CBCT-based ART.

Keywords: uncertainty, Monte Carlo Dropout, MR-linac

References: [1] Jungo, A., et al. (2020). Analyzing the Quality and Challenges of Uncertainty Estimations for Brain Tumor Segmentation. Frontiers in Neuroscience , 14 , 282. [2] Rousseau, A. J., et al. (2021). Post training uncertainty calibration of deep networks for medical image segmentation. IEEE 2021 , pp. 1052-1056. [3] Nixon, J., et al. (2019). Measuring Calibration in Deep Learning. CVPR workshops, 2, 7.