Identifying autism biomarkers in default mode network using sparse representation of resting-state fMRI data

Exploring abnormal functional connectivities in neurodevelopmental disorders have received great attention in recent years. However, identifying functionally homogeneous brain regions as nodes in functional brain connectivity analysis is still challenging. In this paper, we adopt an effective data-driven framework of sparse representation to identify brain network nodes inspired by the nature of sparse population coding of the human brain. Using the default mode network (DMN) in patients with Autism as a test-bed, we evaluate sparse coding method and compared it with widely used group-wise independent components analysis (group-ICA). The experimental results demonstrate that the network nodes identified by sparse representation are more functionally homogeneous, which may explain the superiority of sparse representation in differentiating Autism and healthy controls by brain connectivity biomarkers.