Analysis of time-varying brain connectivity using nonparametric Bayesian model

The study of time-varying brain connectivity is essential for understanding the dynamic interactions between different brain regions, especially in the context of cognitive processes, neurological disorders, and brain network functioning. In this paper, we present a novel approach for analyzing effective brain connectivity using a nonparametric Bayesian model. Specifically, we apply a Hierarchical Dirichlet Process Auto-regressive Hidden Markov Model (HDP-AR-HMM) to capture the temporal evolution and structural patterns of connectivity between brain regions. The proposed model allows for flexible, data-driven clustering of brain states while incorporating both temporal dependencies and hidden states. We demonstrate the utility of this method in revealing the dynamic structure of brain networks and uncovering time-varying patterns of effective connectivity. Our approach is validated using Alzheimer fMRI data, showing that it the dynamic interaction among brain regions during a simple sensory-motor task experiment, providing new insights into the dynamic processes governing brain activity.