Application Of Simultaneous Uncertainty Quantification And Segmentation For Oropharyngeal Cancer Use-case With Bayesian Deep Learning

This article discusses the use of uncertainty in deep learning for 3D medical imaging, specifically in the context of auto-segmentation of GTVp. The study proposes probabilistic deep learning models and evaluates various uncertainty measures for patient-level uncertainty. Three novel measures are also introduced and the results are evaluated using established segmentation performance measures. The study also considers the utility of uncertainty information and its link to performance measures.