Informed Segmentation Approaches for Studying Time-Varying Functional Connectivity in Resting State fMRI

Abstract

The brain is a complex dynamical system that is never truly “at rest”. Even in the absence of explicit task demands, the brain still manifests a stream of conscious thought, varying levels of vigilance and arousal, as well as a number of postulated ongoing “under the hood” functions such as memory consolidation. Over the past decade, the field of time-varying functional connectivity (TVFC) has emerged as a means of detecting dynamic reconfigurations of the network structure in the resting brain, as well as uncovering the relevance of these changing connectivity patterns with respect to cognition, behavior, and psychopathology. Since the nature and timescales of the underlying resting dynamics are unknown, methodologies that can detect changing temporal patterns in connectivity without imposing arbitrary timescales are required. Moreover, as the study of TVFC is still in its infancy, rigorous evaluation of new and existing methodologies is critical to better understand their behavior when applied in resting data, which lacks ground truth temporal landmarks against which accuracy can be assessed. In this dissertation, I contribute to the methodological component of the TVFC discourse. I propose two distinct, yet related, approaches for identifying TVFC using an informed segmentation framework. This data-driven framework bridges instantaneous and windowed approaches for studying TVFC, in an attempt to mitigate the limitations of each while simultaneously leveraging the advantages of both. I also present a comprehensive, head-to-head comparative analysis of several of the most promising TVFC methodologies proposed to date, which does not exist in the current body of literature.