Whole Brain Functional Connectivity Using Multi-scale Spatio-Spectral Random Effects Model

Abstract

Functional brain networks produce connected low frequency patterns of activity when the brain is at rest which can be analyzed with resting state functional MRI (rs-fMRI) by fitting general linear models for signals acquired at a pre-defined seed region and other regions of interest (ROIs). However, typical rs-fMRI analysis tends to ignore spatial correlations in rs-fMRI data, hence biases the standard errors of estimated parameters and leads to incorrect inference. Spatio-temporal or spatio-spectral models can incorporate the spatial correlations in fMRI data. To date, these models have not targeted rs-fMRI connectivity analysis. Herein, we expand a spatio-spectral model from fMRI analysis based on several ROIs to whole brain rs-fMRI connectivity analysis. Our model captures distance-dependent local correlation (within an ROI), distance-independent global correlation (between ROIs), and temporal correlations for whole brain rs-fMRI connectivity analysis with or without confounders. Simulated and empirical experiments demonstrate that this spatio-spectral model yields valid inference for whole brain rs-fMRI connectivity analysis.