B1524
Title: A novel Bayesian covariance regression approach to unveil functional connectivity in resting-state fMRI data
Authors: Tianwen Ma - Emory University (United States) [presenting]
Benjamin Risk - Emory University (United States)
Abstract: Functional magnetic resonance imaging (fMRI) is a non-invasive tool to measure correlations among brain regions and help characterize neurological disorders, such as autism spectrum disorder (ASD). ASD is thought to be associated with changes in communication between brain regions. Motion creates large artifacts in fMRI, which is particularly challenging for children with autism spectrum disorder because they tend to move more. Existing preprocessing pipelines only model the first-order (mean) effects of motion and other nuisance confounders on the fMRI time series and then calculate the covariance between the residuals from different brain locations. Such a procedure may not be sufficient for motion quality control. The goal is to improve the estimation of the covariance matrix of brain activity by additionally considering the impact of nuisance confounders on the covariance structure. A Bayesian covariance regression model is proposed to capture the first-order mean and second-order covariance effects of nuisance confounders. The approach models the covariance matrix conditional on motion and other nuisance confounders. Bayesian covariance regression may characterize the underlying neurological activity more accurately, especially with small sample sizes.
