B0207
Title: Bayesian inference of the brain's effective connectivity
Authors: Tingting Zhang - University of Virginia (United States) [presenting]
Abstract: A new high-dimensional dynamic model is built for the directional interaction, also called effective connectivity, among brain regions. The proposed model is based on a multivariate autoregressive model for time series measurements of brain activity. In order to distinguish strong connections from weak ones, we impose sparsity on the model parameters for the effective connectivity among every pair of brain regions. Specifically, the new model features a cluster structure, which consists of modules of densely connected brain regions. We show that the autoregressive model can outperform a high-dimensional ordinary differential equation model for the brain's effective connectivity. We develop a unified Bayesian framework to make inferences about the brain networks using a stochastic block prior for network structures. We apply the proposed method to time series data of brain activity, specifically, intracranial EEG data, and investigate brain network changes around time.
