A Bayesian Spatiotemporal Model for Joint Estimation of Brain Activation and Connectivity in fMRI Studies

Task-based functional magnetic resonance imaging (fMRI) experiments play a crucial role in modern data-driven neuroscience research. These studies often aim to understand how external stimuli trigger activation in specific brain regions and to explore functional connectivity patterns among predefined regions, commonly referred to as regions of interest (ROIs). Accurately estimating both brain activation and inter-regional connectivity is challenging due to complex spatiotemporal correlations and low signal-to-noise ratios inherent in fMRI data. This paper introduces a joint spatiotemporal Bayesian framework that simultaneously models activation and connectivity across multiple subjects while estimating the hemodynamic response function (HRF) for each region. Spatial dependencies are captured via an unweighted graph-Laplacian prior on regression and autoregressive coefficients, and region-specific random effects are modeled using a Bayesian Gaussian graphical model to reflect connectivity among ROIs. We evaluate the performance of the model through simulation studies, demonstrating robust estimation under realistic low signal-to-noise conditions. The approach is then applied to a multisubject motor task dataset from the Human Connectome Project (HCP), mapping brain motor areas associated with specific movements (e.g., finger, toe, tongue) and assessing their activation and lateralization in response to visual cues. The model is further evaluated on the Individual Brain Charting (IBC) dataset, a high-resolution 3T fMRI dataset designed for fine-grained cognitive mapping across multiple tasks. Finally, the model is validated through comparisons with the classical general linear model (GLM), which is commonly used in the neuroscience community, highlighting the advantages of our Bayesian approach in jointly capturing activation and connectivity patterns.