Decoding the white matter geometrical structure by encoding connectomes in multidimensional spaces

FRANCO PESTILLI; CESAR F. CAIAFA

San Diego

Conferencia; Society for Neuroscience 2016; 2016

SfN

A growing body of scientific evidence suggests the inner workings of the brain cannot be simply understood at the level of single neurons or isolated brain areas; rather, modern efforts must focus on brain networks and circuit behavior. The full network of brain connections, commonly referred to as the connectome, is comprised of both grey matter regions representing neural units of information processing, and white-matter tracts serving as structural communication pathways. To date, large-scale human brain networks can be mapped in-vivo by using diffusion- weighted magnetic resonance imaging and fiber tracking methods (Bullmore and Sporns 2009). Standard methods for mapping connectomes operate on data based on the naturally occurring geometry of the brain white matter. This geometry is complex and difficult to handle computationally. As a result a majority of algorithms for mapping connectomes rely on sets of ad-hoc rules and heuristics that identify white matter pathways one fascicle at the time. This is a computationally suboptimal process prone to failure because no rule is expected to be optimal in all brain locations or across multiple brains (Takemura et al., 2016). Computational limitations restrict our ability to routinely measure the accuracy of connectome models (Pestilli et al., 2014).We introduce an approach that uses multidimensional arrays to encode anatomical properties of connectomes in a computationally-efficient way. The method allows performing neuroanatomical operations on the geometric organization of the brain?s white matter. We analyzed 1,040 brain connectomes generated using multiple tractography algorithms on publicly available datasets (Human Connectome Project, purl.stanford.edu/ng782rw8378). Results show that the framework: (1) Can be used to establish approaches for precision connectomics by allowing building highly reliable within-brain and highly discriminable between-brain connectomes; (2) Allow efficient estimation of distributions of crossing angles between white matter fascicles (Van Wedeen et al., 2012; Catani et al., 2012). In sum, we show a connectome encoding framework with important computational advantages for decoding fundamental features of the human brain.