A Riemannian Framework For Functional Clustering Of Whole Brain White Matter Fibers

White matter consists of fibers that transmit signals between different brain regions, and there has been evidence that neural activity is encoded in the white matter BOLD signals. Functional clustering of fibers using white matter BOLD signals provides a novel perspective to explore the evolution of functional architecture in axonal fibers and the relationship between structure and function in the human brain. We develop a comprehensive Riemannian framework to integrate the physical characteristics of the fiber with the white matter BOLD signal along the fiber for whole brain fiber clustering. Specifically, we derive not only a novel metric that provides a cost function for registration and a geodesic distance for comparison but also a flexible weight to study the interrelationship of structure and function. The evaluation results demonstrate that our framework achieves functionally meaningful and consistent clustering results.