Tractometric Coherence of Fiber Bundles in DTI

Based on a diffusion tensor image (DTI) and a tentative tractogram of a fiber bundle we propose a filtering method for operationally defining and removing outliers using tractometry. To this end we assign to each track a set of K invariants, i.e. scalars invariant under rigid transformations. The cluster of K-tuples of all tracks in a bundle may be pruned using outlier detection methods in RK, after which backprojection of the remaining K-tuples produces a filtered tractogram with enhanced coherence. This intrinsic pruning method is blind to the relative spatial organization of tracks in a bundle. We consider two types of invariants, one capturing local diffusion properties and one representing differential properties averaged along tracks. Our experiments indicate that our tractometric filtering is complementary to extrinsic methods based on the relative spatial configuration of tracks.