Exploring Functional Connectivity Biomarker in Autism Using Group-Wise Sparse Representation

Abstract

Exploring the brain as a complex, networked system and inferring the dysfunction of diseased brains by abnormal functional connectivity has received great attention in recent years. One critical problem in brain network analysis is how to identify functionally homogeneous brain regions as network nodes. Inspired by the nature of sparse population coding of the human brain, we propose a novel data-driven method to identify whole-brain network nodes based on group-wise sparse representation (gSR) algorithm. Using a publicly available autism dataset as test-bed, we evaluate our method and compare it with group-wise independent components analysis (gICA). The experimental results demonstrate that the brain ROIs identified by our method are more functionally homogeneous and thus may improve the sensitivity and accuracy of functional connectivity biomarkers in differentiating autism and healthy controls.