Abstract
Resting-state functional connectivity is an important and widely used measure of individual and group differences. These differences are typically attributed to various demographic and/or clinical factors. Yet, extant statistical methods are limited to linking covariates with variations in functional connectivity across subjects, especially at the voxel-wise level of the whole brain. This paper introduces a generalized linear model method that regresses whole-brain functional connectivity on covariates. Our approach builds on two methodological components. We first employ whole-brain group ICA to reduce the dimensionality of functional connectivity matrices, and then search for matrix variations associated with covariates using covariate assisted principal regression, a recently introduced covariance matrix regression method. We demonstrate the efficacy of this approach using a resting-state fMRI dataset of a medium-sized cohort of subjects obtained from the Human Connectome Project. The results show that the approach enjoys improved statistical power in detecting interaction effects of sex and alcohol on whole-brain functional connectivity, and in identifying the brain areas contributing significantly to the covariate-related differences in functional connectivity.
