Characterizing task-evoked and intrinsic functional networks from task-based fMRI data via two-stage sparse dictionary learning

Recently, increasing studies suggest that task-evoked brain networks and intrinsic connectivity networks are concurrent during task performance in the architecture of functional brain organization. However, it remains challenging to identify and quantitatively characterize these mixtures of networks from task-based functional magnetic resonance imaging (fMRI) data. In this paper, we propose a two-stage sparse dictionary learning method by establishing spatial correspondences among the brain networks of multiple subjects and automatically categorize these networks into task-evoked, intrinsic and uncertain ones. The proposed framework is applied to Human Connectome Project (HCP) task-based fMRI data. Experimental results demonstrate that this method can effectively identify group-wise task-evoked and intrinsic networks simultaneously. Our study provides a novel data-driven method to facilitate a comprehensive understanding of the functional brain architecture.