Evaluating the Accuracy of White-Matter Connectomes

Franco Pestilli¹, Jason Yeatman¹, Ariel Rokem¹, Kendrick Kay¹, Brian Wandell²

We propose a method to evaluate the accuracy of white-matter connectomes generated using diffusion weighted MRI and fiber tractography. The method predicts the diffusion data in each voxel as a weighted sum of the contributions from all the fascicles in that voxel. We derive several connectomes using different tractography algorithms (TEND, FACT and CSD). We evaluate each connectome using cross-validation with respect to a second set of diffusion measurements in the same subject. Connectomes generated using the CSD algorithm with an intermediate order of spherical harmonic basis set (4>lmax<16) predicts diffusion measurements better than data reliability.