Spatial-Temporal Convolutional Attention for Mapping Functional Brain Networks

Using functional magnetic resonance imaging (fMRI) and deep learning to explore functional brain networks (FBNs) has attracted many researchers. However, most of these studies still rely on temporal correlation between sources and voxel signals, lacking exploration of the dynamics of brain function. Due to the prevalent local correlations in volumes, FBNs can be directly generated in the spatial domain using spatial-wise attention (SA) in a self-supervised manner, resulting in higher spatial similarity with templates compared to classical methods. Therefore, we propose a novel Spatial-Temporal Convolutional Attention (STCA) model to dynamically discover FBNs using sliding windows. We validate the performance of our proposed method on the HCP-rest dataset, showing that STCA can be used to dynamically discover FBNs, offering a novel approach to better understand the human brain.