Meeting Challenges for Modeling Brain Imaging Data: The Spatio-Temporal Perspective

• Activity Number: 486
• Type: Invited
• Date/Time: Wednesday, August 3, 2011 : 10:30 AM to 12:20 PM
• Sponsor: ENAR
• Author(s): DuBois Bowman*+ and Shuo Chen
• Companies: Emory University and Emory University
• Keywords: Imaging ; spatial modeling ; fMRI ; resting-state analysis ; mental health

Abstract

Analyzing resting-state fMRI data has revealed brain networks that exhibit consistent properties across subjects. Moreover, altered resting-state network properties are associated with mental illnesses, including major depressive disorder (MDD). Most resting-state fMRI studies conduct network analyses; however, important information may exist in localized "activation" properties. A key challenge in characterizing localized activation properties for resting-state fMRI data is that first-order moments are not interpretable as activation statistics. We extend a Bayesian spatial model (Bowman et al., 2008) that was primarily intended for the analysis of task-induced neural processing, which yields inferences for both localized activation patterns and network properties. We first construct frequency-domain descriptors of the resting-state activity profiles, then use a Bayesian hierarchical model to estimate localized patterns of this measure as well as a covariance matrix reflecting the functional connectivity between brain regions. We perform estimation using MCMC implemented via Gibbs sampler. We apply our Bayesian model to data from a resting-state fMRI study of MDD.