A1576
Title: A Bayesian approach to simultaneous estimation of neural functional connectivity and sub-network structure
Authors: Julia Fisher - University of Arizona (United States) [presenting]
Edward Bedrick - University of Arizona (United States)
Abstract: Over the past couple of decades, a great deal of research has examined how distinct regions of the brain co-activate, typically using functional magnetic resonance imaging data gathered from subjects scanned while awake but at rest. A key observation of such studies has been that the functional network of the human brain can be partitioned into sets of regions (e.g., modules or sub-networks such as the well-studied default mode network) that are more closely related to each other than to regions outside the set. Most approaches to analyzing such data focus either on the estimation of region-to-region connectivity (often via partial or full correlation) or sub-network structure, but not both. Simultaneous estimation of the partial correlation between regions of interest and the partitioning of the network into modules is proposed via a Bayesian multivariate normal-Wishart model with an embedded Dirichlet process mixture model for clustering. The model on a range of simulated data is evaluated, and it is compared to other analytical approaches.
