Multiple sclerosis (MS) is an autoimmune disease that typically manifests in relapses. Clinical trials typically focus on disability, but this is difficult to measure and has high heterogeneity, hence the search for useful image-based markers derived from magnetic resonance imaging (MRI). In this talk, I will outline two collaborative research efforts using machine learning with MRI data to improve differential diagnosis (using a convolutional neural network) and survival prediction (using graph-based representations of lesion distributions).