Medical imaging technology has allowed for unparalleled insight into the structure and function of the human brain, giving clinicians powerful new tools for disease diagnosis and monitoring. Yet the complex and high-dimensional nature of imaging data makes computational analysis challenging. In multiple sclerosis (MS), this complexity is typically simplified by identifying regions of visible tissue damage and measuring spatial extent. However, many common radiological measures have been shown to be only weakly associated with clinical outcomes (a discovery that has been referred to as “the clinico-radiological paradox”). We attempt to bridge this gap by developing statistical methods capable of extracting clinically relevant information from MRI scans in MS. Here, we discuss three such techniques: a texture modeling approach to improve research on lesion dynamics; a biomarker detection algorithm to support diagnostic decision-making; and a flexible multi-modal group differences test to facilitate exploration of subtle disease processes. The performance of these methods is illustrated using simulated and real data, and the opportunities and obstacles for their clinical use are discussed.