GLOBAL SIGNAL REGRESSION AND BEHAVIORAL ASSOCIATION IN THE RESTING-STATE

Abstract

Global signal regression (GSR) is a controversial but widely used preprocessing step in studies of resting-state functional MRI (rs-fMRI) of human brain activity, due to the complex mixture of noise and neural information in global signal (GS). Given significant interest in the relationship between human behavior and functional brain architecture revealed by rs-fMRI, this thesis focused on (1) whether GSR strengthens or weakens rs-fMRI-behavior associations and (2) the relationship between GS and behaviors. First, the associations of rs-fMRI with a large range of behaviors were compared between preprocessing strategies with and without GSR. These rs-fMRI-behavior associations were quantified by two methods: variance component model and kernel regression. The results of both methods suggested that GSR robustly strengthened the rs-fMRI-behavior association. Furthermore, removing the GS had a larger impact on task-performance measures than on self-reported behavioral measures. Second, this thesis investigated the relationship between GS topography and an extended set of behavioral measures. Canonical correlation analysis revealed a significant mode relating GS topography to cognitive and life outcome measures, which suggests that the GS contains behaviorally relevant information. Nevertheless, the benefit of noise/artifact reduction from removing the GS appears to outweigh the loss of behavior-related information contained in the GS.