The Case for Optimized Edge-Centric Tractography at Scale

• Published in: Frontiers in neuroinformatics Vol. 16; p. 752471
• Main Authors: Moon, Joseph Y, Mukherjee, Pratik, Madduri, Ravi K, Markowitz, Amy J, Cai, Lanya T, Palacios, Eva M, Manley, Geoffrey T, Bremer, Peer-Timo
• Format: Journal Article
• Language: English
• Published: Switzerland Frontiers Research Foundation 16.05.2022; Frontiers Media SA; Frontiers Media S.A
• Subjects: Algorithms; Brain research; connectomes; diffusion MRI; EDI; Embedding; identifiability; Neuroimaging; Neuroscience; optimization; Orthopedics; tractography; Traumatic brain injury; edge-centric; tractography; optimization; identifiability; connectomes; diffusion MRI; EDI
• ISSN: 1662-5196; 1662-5196
• DOI: 10.3389/fninf.2022.752471

The anatomic validity of structural connectomes remains a significant uncertainty in neuroimaging. Edge-centric tractography reconstructs streamlines in bundles between each pair of cortical or subcortical regions. Although edge bundles provides a stronger anatomic embedding than traditional connectomes, calculating them for each region-pair requires exponentially greater computation. We observe that major speedup can be achieved by reducing the number of streamlines used by probabilistic tractography algorithms. To ensure this does not degrade connectome quality, we calculate the identifiability of edge-centric connectomes between test and re-test sessions as a proxy for information content. We find that running PROBTRACKX2 with as few as 1 streamline per voxel per region-pair has no significant impact on identifiability. Variation in identifiability caused by streamline count is overshadowed by variation due to subject demographics. This finding even holds true in an entirely different tractography algorithm using MRTrix. Incidentally, we observe that Jaccard similarity is more effective than Pearson correlation in calculating identifiability for our subject population.