Identifying Heterogeneous Disorders Through Normative Modeling of the Resting State BOLD Signal

Abstract

One of the recent developments of interest in neuroscience is the detailed study intoresting state data. This data allows us to examine the way our brains behave in an awakestate with no tasks presented. When coupled with non-imaging data, resting state data cangreatly assist in the classification of disorders and speed patients along the road to recovery.The method of classification examined in this thesis is normative modeling through ScalableMulti-Task Gaussian Process Regression (S-MTGPR). Two distinct and unrelated datasetsare used - a large dataset (n=172) and a small dataset (n=27) - to prove the effectiveness ofS-MTGPR is unrelated to sample size. In this thesis we examine how normative models arebuilt and how they classify subjects based on neurological activity.