Anatomical parcellation of human brain using structural covariance

Abstract

A reliable human brain atlas is critical for brain network analysis at macro-scale. Most studies employed existing anatomical brain atlases or randomly parcellated the whole brain into discrete regions. However, these anatomical atlases had a large variation in region sizes, and the random parcellation procedure was lack of explicit biological significance. In this study, we proposed a new brain parcellation framework which could automatically construct anatomical brain atlases for a specific group of subjects based on the structural covariance patterns. The changes of the modulated grey and white matter densities across individuals were used as features and sparse representation was employed to calculate the similarities. The results showed that our method achieved high consistency in brain parcellation on two independent datasets with well correspondence to existing anatomical atlases. Validation experiments on specific brain regions presented consistent parcellation patterns with anatomical and functional connectivity. These results implied that our method could generate biologically meaningful parcellations for the human brain.