Intra-subject reliability of the high-resolution whole-brain structural connectome

Recent advances in diffusion weighted image acquisition and processing allow for the construction of anatomically highly precise structural connectomes. In this study, we introduce a method to compute high-resolution whole-brain structural connectome. Our method relies on cortical and subcortical triangulated surface models, and on a large number of fiber tracts generated using a probabilistic tractography algorithm. Each surface triangle is a node of the structural connectivity graph while edges are fiber tract densities across pairs of nodes. Surface-based registration and downsampling to a common surface space are introduced for group analysis whereas connectome surface smoothing aimed at improving whole-brain network estimate reliability. Based on 10 datasets acquired from a single healthy subject, we evaluated the effects of repeated probabilistic tractography, surface smoothing, surface registration and downsampling to the common surface space. We show that, provided enough fiber tracts and surface smoothing, good to excellent intra-acquisition reliability could be achieved. Surface registration and downsampling efficiently established triangle-to-triangle correspondence across acquisitions and high inter-acquisition reliability was obtained. Computational time and disk/memory usages were monitored throughout the steps. Although further testing on large cohort of subjects is required, our method presents the potential to accurately model whole-brain structural connectivity at high-resolution.