A stringent fiber distance measure for dMRI tractography clustering and segmentation

Most analysis and segmentation methods for diffusion MRI tractography datasets require a fiber distance measure able to determine the similarity between a pair of fibers. We present a stringent fiber distance measure able to perform a good discrimination between fiber shapes and lengths. It uses three terms: (i) a fiber maximum Euclidean distance, (ii) a fiber shape distance, and (iii) a fiber length distance. The distance was evaluated applying a hierarchical clustering of fibers connecting the pre-and post-central gyri of a subject. Results where compared with other known fiber distance measures. A better sensitivity to differences in fiber shape and length was found for the proposed distance. This will be very useful for the detailed study and description of white matter bundles. Known bundles will be better decomposed into sub-bundles, with more precision on the bundle shape and on the regions connected by the fibers. For short association bundles, this distance will be a real improvement, as even the most stringent distance used until now shows some limitations when evaluating the similarity of these fibers.