Detecting Brain Vulnerability using Network Neuroscience Methods

Abstract

In this thesis, I present a sophisticated pipeline for functional subnetwork detection that applies to individualized brain data analysis. The pipeline is presented as part of a research toolkit for network neuroscience research that includes both analysis and multiple neuroimaging exports, enabling better communication in an interdisciplinary field. The pipeline and analytical tools presented were used to evaluate the brain's health status and its risk of dysfunction, an area of research with limited data-driven methodologies due to its complexity. I evaluated how brain vulnerability manifests in two at-risk cohorts; the first study investigates how ovarian hormones affect functional brain networks resulting in a window of vulnerability to affective disorders during the menstrual cycle. Using the toolkit's efficient functional subnetwork detection pipeline, we tracked each participant's brain organization changes across the menstrual cycle. We found that ovarian hormones influence regions in the brain responsible for mood regulation and emotion, starting earlier in the menstrual cycle than expected. The second study opens a new avenue for assessing brain vulnerability to cognitive impairment in those who experienced a transient ischemic attack, a mini-stroke with symptoms subsiding within 24 hours. Traditional cognitive evaluation methods such as clinical assessments found no longitudinal changes within this dataset, although there are significant cross-sectional differences between the afflicted group versus the control group. Using controllability metrics from network control theory, we found that individuals who had a transient ischemic attack showed subtle yet significant control changes within the brain's sub-structures that are important for executive and perceptual functions. This demonstrates the sensitivity of controllability metrics and how they can be used to detect brain vulnerability to cognitive decline. The results of this thesis shed light on how vulnerability is expressed in different brain network models. This highlights the comprehensiveness and versatility of the code base and its computational efficiency to support the analysis of cohorts at the individual level. Given the trend towards large neuroimaging datasets to support statistical power, and the pressing need to understand the great diversity in human brain architecture, the toolkit offers a powerful and timely tool for researchers of brain health.