A Bayesian Vector Autoregressive Approach to Joint Connectivity and Activation Analysis in fMRI

• Activity Number: 362
• Type: Contributed
• Date/Time: Tuesday, August 6, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Imaging
• Author(s): Zhe Yu*+ and Hernando Ombao and Wesley K Thompson and Robert E. Kass
• Companies: University of California, Irvine and University of California at Irvine and University of California, San Diego and Department of Statistics, Carnegie Mellon University
• Keywords: Connectivity ; Vector-Autoregressive Models ; Bayesian Statistics ; Granger Causality ; Hemodynamic Response Function ; Functional Magnetic Resonance Imaging

Abstract

To study connectivity between brain regions, we developed a modified Bayesian vector autoregressive model with exogenous variables (VARX) for fMRI data. The model simultaneously estimates hemodynamic response functions (HRF) whose convolutions with given stimulus functions make the exogenous variables. Existing methods either ignore the association between brain regions, or assume one known HRF across all brain regions, which could lead to huge bias in VAR components if the HRF is mis-specified. Our method addresses the limitations by including region-specific HRF, while modeling the regions together using VAR. Additionally, hyperparameters at subject level are used to capture the main characteristics and inter-subject variability. The model is able to study effective connectivity and functional connectivity between the brain regions by analyzing the VAR coefficient matrices and the partial correlation matrix, respectively. We investigated the statistical properties of the method through simulations, and also applied the method to data collected from subjects while they were performing BART task to determine the effect of alcohol on restraint for risky behavior.