Spinal cord segmentation and disability prediction in multiple sclerosis using novel optimization and machine learning methods

Multiple sclerosis studies show a correlation between spinal cord atrophy and physical disability, indicating the potential use of atrophy as a biomarker to monitor disease progression. To quantify spinal cord atrophy, clinicians need to accurately measure the cord and determine which cord properties consistently capture tissue loss. We address these needs by making three contributions: (i) a novel algorithm to segment the spinal cord by finding the globally optimal minimal path in six dimensions; (ii) a machine learning spinal cord segmentation approach where we introduce the concept of global geometric features into decision forests to address the first algorithm's limitations; and (iii) novel morphological and appearance features extracted from magnetic resonance images (MRI) and corresponding spinal cord segmentations that are combined to predict the physical disability of patients with multiple sclerosis. Our results demonstrate improvements over state-of-the-art spinal cord segmentation methods and improved prediction of clinical disability from MRI data.