A Riemannian Fiber Similarity Measure and its Application to White Matter Fiber Clustering
A. Demir, G. Unal
brain, white matter, fiber clustering, Riemannian geometry
Clustering of reconstructed brain white matter fibers into meaningful anatomical bundles becomes an important tool for detailed analysis of brain white matter diseases via diffusion MRI. In this paper we present a Riemannian geometry based geodesic distance measure between fiber pairs, which can directly be utilized in fiber clustering. Second contribution is a fiber selection algorithm, which compresses and down-samples the dataset. We demonstrate performance of proposed methods on several synthetic fiber datasets, and validate on a brain white matter atlas. We also show the clustering results of real clinical diffusion MRI cases.