fMRI Analysis with Structured Bayesian Variable Selection

• Activity Number: 327
• Type: Topic Contributed
• Date/Time: Tuesday, August 3, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Feng Liang*+
• Companies: University of Illinois

Abstract

Functional magnetic resonance imaging (fMRI), one of the recently developed forms of neuroimaging technology, has radically improved our understanding of the structure, functional organization and pathology of the human brain. One main aim of fMRI is to detect regions that respond to certain stimuli, i.e., activation regions. One challenge in fMRI analysis is that the fMRI response is not instantaneous, but lagged and damped by an unknown hemodynamic response function (HRF). We present a novel Bayesian method for simultaneous HRF estimation and activation detection for fMRI data. A Bayesian variable selection approach is used to induce shrinkage and sparsity; a spatial prior on latent variables is used to pool information from neighboring voxels; the activation map is generated from the full spectrum of posterior inference through a Markov chain Monte Carlo scheme.