Covarying gray and white matter networks characterize schizophrenia and bipolar disorders on a continuum: A data fusion machine learning approach and a brain network analysis

Schizophrenia (SZ) and Bipolar disorder (BD) share genetic and cerebral abnormalities, supporting an expanded continuum hypothesis. In this paper, we aim to better characterize differences and commonalities of gray and white matter features between SZ and BD to clarify how they align or diverge on this continuum. We transposed independent vector analysis (tIVA), a data fusion technique, to the gray and white matter images of 128 individuals diagnosed with SZ, 128 with BD and 127 healthy controls (CTRL), matched for gender, age and IQ. Of the 18 tIVA networks detected, three differed between SZ and BD (tIV9,14,15), primarily involving fronto-temporal regions. These same networks plus two more (tIV3,4), differed between SZ and CTRL indicating a larger compromission, whereas only one network (tIV9) differed between BD and controls. Overall, SZ displayed the more pronounced GM-WM abnormalities in both extent and severity with BD lying in an intermediate position. Of note, one network differed among all three groups (SZ, BD, and CTRL). Random forest classification confirmed these results by indicating the tIV9 as the main predictors that separate the three groups. Moreover, to appreciate eventual differences between networks across the three groups a network analyses was performed. Individuals with SZ demonstrated a significantly different clustering coefficient and density compared to CTRL. While the comparison between individuals with BD and controls did not show marked differences. This study sheds new lights on the expanded continuum hypothesis according to which individuals with schizophrenia and bipolar disorder lay on the same continuum of neurological abnormalities.