“…While some previous studies have leveraged such longitudinal data for patient-specific fine-tuning, resulting in tailored networks for individual patients [8] , [9] , [10] , [11] , [12] , [13] , a deeper examination of the potential advantage of incorporating multiple images per patient into a general auto-contouring network remains somewhat unexplored. Longitudinal images have found their place in some auto-contouring studies for MR-Linacs [14] , [15] and other treatments that involve daily imaging and contouring, such as cervical brachytherapy [16] , [17] , [18] . However, the question of whether multiple images from the same patient can provide sufficient diversity to substantially enhance the performance of an auto-contouring network in comparison to single-image approaches is not investigated.…”