Spatio-Temporal Learning from Longitudinal Data for Multiple Sclerosis Lesion Segmentation

Segmentation of Multiple Sclerosis (MS) lesions in longitudinal brain MR scans is performed for monitoring the progression MS lesions. We hypothesize that spatio-temporal cues data can aid segmentation algorithm. Therefore, we propose a multi-task learning approach by defining an auxiliary self-supervised task deformable registration between two time-points to guide neural network toward from changes. show efficacy our method on clinical dataset comprised 70 patients with one follow-up study each patient. Our results information beneficial cue improving segmentation. improve result current state-of-the-art 2.6% terms overall score (p < 0.05). Code publicly available ( https://github.com/StefanDenn3r/Spatio-temporal-MS-Lesion-Segmentation ).