In multiple sclerosis (MS), some brain white matter lesions are more destructive and feature an edge of chronic/smoldering inflammation, which corresponds to a paramagnetic rim on susceptibility-based MRI. 1 Higher paramagnetic rim lesion (PRL) burden is associated with more aggressive disease course, 2, 3 however the longitudinal effect of available MS disease modifying treatments (DMT) on PRL is still largely unknown. Considering that manual rim classification is tedious and associated with interobserver variability, a multimodal convolutional neural network for automated detection (RimNet) has been recently develope d 4. In this work, we tested RimNet (i) on data acquired in two centers using MRI scanners from two different vendors (including one not used for training in the original paper); and (ii) longitudinally on MRI data acquired in primary and secondary progressive MS patients (PMS) before and after DMT.