A Novel Paradigm For Automated Segmentation of Very Large Whole-Brain Probabilistic Tractography Data Sets

Robert Elton Smith^1,2, Jacques-Donald Tournier^1,2, Fernando Calamante^1,2, Alan Connelly^1,2

Conventional clustering based upon pair wise similarities has proven inadequate for the task of meaningful segmentation of whole brain probabilistic tractography. A new fully-automated algorithm has been developed based upon the identification of bound coherent bundles of tracks; fibers are segmented based upon their traversal through a common structure, rather than similarity along their entire lengths. It identifies anatomically-meaningful structures at a wide range of physical scales, and intrinsically captures the structural connectivity of each region. We demonstrate the technique on a 10,000,000 probabilistic streamlines data set.