A0150
Title: Bayesian modeling in neuroimaging: Brain networks dynamics
Authors: Michele Guindani - University of California Los Angeles (United States) [presenting]
Abstract: The critical role that statistical approaches play in analyzing brain imaging data will be first highlighted, particularly for functional magnetic resonance imaging (fMRI) data. Appropriate statistical methods are necessary to handle the complexity of spatial and temporal correlations typical of brain data. More specifically, we will discuss approaches to studying dynamic brain connectivity, which seeks to understand the changing interactions between different brain regions over time. We will present two novel Bayesian approaches to capture these dynamic relationships within multivariate time series data. First, we will present a scalable Bayesian time-varying tensor vector autoregressive (TV-VAR) model, aimed at efficiently capturing evolving connectivity patterns. This model leverages a tensor decomposition of the VAR coefficient matrices at different lags and sparsity-inducing priors to capture dynamic connectivity patterns. Next, we will introduce a Bayesian framework for sparse Gaussian graphical modeling, which employs discrete autoregressive switching processes. This method improves the estimation of dynamic connectivity by modeling state-specific precision matrices, using novel prior structures to account for temporal and spatial dependencies. Throughout the talk, we will illustrate the performance of these Bayesian methods with examples from simulation studies and real-world fMRI data.
