IDENTIFICATION OF BRAIN NETWORK FEATURES SUPPORTING HUMAN BEHAVIOR

Abstract

A central question in systems neuroscience is how individual differences in brain network organization track behavioral variability. However, most studies focus on single behavioral traits, thus not capturing broader relationships across behaviors. In addition, the reliability of neuroimaging studies has been questioned for their small sample size and flexible analytical methods. In this thesis, brain network organization was utilized to predict a variety of behavioral measures across three behavioral domains: cognition, personality, and mental health. The results were replicated using different data samples and prediction models. Predictive network features were distinct across the behavioral domains but similar within each domain. This thesis also investigated the factors affecting the reliability of feature importance inferred from linear prediction models. Results showed that the Haufe inversion approach and large sample size lead to more reliable model interpretation. Furthermore, reliability of feature importance was also correlated with prediction accuracy.