Anatomical structural network analysis of human brain using partial correlations of gray matter volumes

Structural connectivity in human brain has been studied by modeling the statistical dependence between features of cortical regions, such as gray matter thickness. Statistical correlations between gray matter thickness have been mainly used as a metric to study this dependence. In this paper, we propose the use of partial correlations instead of Pearson correlation for inferring the brain structural connectivity using gray matter volumes from a large population of 466 subjects. We argue that partial-correlation is a better measure for extracting connectivity matrix from multivariate data because it removes the effects of confounding correlations that get introduced due to canonical dependence between data. Our experimental results on gray-matter volumes from a large population of brains compare and contrast the connectivities obtained by applying both correlation and partial correlation analysis.