Computer-aided segmentation on MRI for prostate radiotherapy, part II: Comparing human and computer observer populations and the influence of annotator variability on algorithm variability

•Different loss functions for developing deep learning (DL) algorithms can change prostate and organs at risk (OAR) boundaries, particularly in anatomical regions with high interobserver variability.•DL-based automatic segmentation algorithms exhibit high variability in similar anatomical regions as the humans who annotated the images for the DL algorithm development.•Spatial entropy maps provide an intuitive characterization of voxel-wise segmentation variability.•DL-based automatic segmentation algorithms can be more consistent than human observers in delineating the prostate and OARs on MRIs for prostate radiotherapy.•Segmentation performance of T2-weighted planning MRIs was comparable to that of T2/T1-weighted postimplant MRIs.