Imaging genetics posits a valuable strategy for elucidating genetic influences on brain abnormalities in psychiatric disorders. However, association analysis between 2D genetic data (subject × genetic variable) and 3D first-level functional magnetic resonance imaging (fMRI) data (subject × voxel × time) has been challenging given the asymmetry in data dimension. A summary feature needs to be derived for the imaging modality to compute inter-modality association at subject level. In this work, we propose to use variability in resting state networks (RSNs) and functional network connectivity (FNC) as potential features for purpose of association analysis. We conducted a pilot study to investigate the proposed features in a dataset of 171 healthy controls and 134 patients with schizophrenia (SZ). We computed variability in RSN and FNC in a group independent component analysis framework and tested three types of variability metrics, namely Euclidean distance, Pearson correlation and Kullback-Leibler (KL) divergence. Euclidean distance and Pearson correlation metrics more effectively discriminated controls from patients than KL divergence. The group differences observed with variability in RSN and FNC were highly consistent, indicating patients presenting increased deviation from the cohort-common pattern of RSN and FNC than controls. The variability in RSN and FNC showed significant associations with network global efficiency, the more the deviation, the lower the efficiency. Furthermore, the RSN and FNC variability were found to associate with individual SZ risk SNPs as well as cumulative polygenic risk score for SZ. Collectively the current findings provide preliminary evidence for variability in RSN and FNC being promising imaging features that may find applications as biomarkers and in imaging genetic association analysis.
Keywords: PGC; functional network connectivity; parallel ICA; resting state network; schizophrenia; variability.