Bayesian Tensor Factorized Mixed Effects Vector Autoregressive Processes for Inferring Granger Causality Patterns from High-Dimensional Neuroimage Data

Understanding the dynamics of functional brain connectivity patterns using noninvasive neuroimaging techniques is an important focus in human neuroscience. Vector autoregressive (VAR) processes and Granger causality analysis (GCA) have been extensively used to examine functional brain connectivity. While high-resolution neuroimage data are routinely collected now-a-days, the statistics literature on VAR models has remained heavily focused on small-to-moderate dimensional problems and single subject data. Motivated by these issues, we develop a novel Bayesian semiparametric VAR model that addresses the daunting dimensionality challenges by structuring the VAR coefficients matrices as a three-way tensor and then applying a tensor decomposition. A novel sparsity-inducing shrinkage prior allows data-adaptive dimension reduction, including automated lag selection. We also extend the approach to a novel mixed model for multi-subject neuroimaging data, capturing common brain connectivity patterns via shared fixed effects while also accommodating subject specific heterogeneity via random effects. Finally, GCA is performed via a posterior false discovery rate control procedure. We design a Markov chain Monte Carlo algorithm for posterior computation. We evaluate the methods' empirical performances through synthetic experiments. Applied to our motivating functional magnetic resonance imaging study, the proposed approach allows the directional connectivity of brain networks to be studied in fine detail, revealing meaningful but previously unsubstantiated cortical connectivity patterns.