Bayesian Factor Analysis for fMRI Data
Lin
Sun

PhD Candidate, Statistics

Monday, July 15, 2013 - 3:00pm

Functional magnetic resonance imaging (fMRI) provides high quality visualization of the location of activity in the brain resulting from sensory stimulation or cognitive function. BOLD fMRI takes advantage of the fact that local blood flow increases following an increase in neuronal activity. Among the various fMRI analysis techniques, the Bayesian factor analysis has been widely used for assessment of multivariate dependence and codependence, which solve the problem that the parameters can't be uniquely determined from the likelihood alone in classical factor analysis. In this study, BOLD measurement of the acute effect, a substance shown previously to engage multiple sites within the orbitofrontal cortex, was processed with the Bayesian factor model for comparative group. The flexibility of the Bayesian factor analysis was shown by choosing different modeling strategies to form the prior reference functions, including approximating simultaneously measured behavioral data and the effect of transformed stimulus in semiparametric Bayesian factor model. Bayesian factor analysis provides a powerful approach to understand BOLD response. Several factors have been determined to explain most of variance within the data. It is demonstrated that Bayesian factor analysis successfully associates the activated activity with the Bayesian factors.

 

Food Sciences Building, Room 131
Major Professor(s): 
Dr. Nicole Lazar