Functional MRI connectivity accurately distinguishes cases with psychotic disorders from healthy controls, based on cortical features associated with neurodevelopment

Background Machine learning (ML) can distinguish cases with psychotic disorder from healthy controls based on magnetic resonance imaging (MRI) data, with reported accuracy in the range 60-100%. It is not yet clear which MRI metrics are the most informative for case-control ML. Methods We analysed multi-modal MRI data from two independent case-control studies of patients with psychotic disorders (cases, N = 65, 28; controls, N = 59, 80) and compared ML accuracy across 5 MRI metrics. Cortical thickness, mean diffusivity and fractional anisotropy were estimated at each of 308 cortical regions, as well as functional and structural connectivity between each pair of regions. Functional connectivity data were also used to classify non-psychotic siblings of cases (N=64) and to distinguish cases from controls in a third independent study (cases, N=67; controls, N = 81). Results In both principal studies, the most diagnostic metric was fMRI connectivity: the areas under the receiver operating characteristic curve were 92% and 77%, respectively. The cortical map of diagnostic connectivity features was replicable between studies ( r = 0.31, P < 0.001); correlated with replicable case-control differences in fMRI degree centrality, and with prior cortical maps of aerobic glycolysis and adolescent development of functional connectivity; predicted intermediate probabilities of psychosis in siblings; and replicated in the third case-control study. Conclusions ML most accurately distinguished cases from controls by a replicable pattern of fMRI connectivity features, highlighting abnormal hubness of cortical nodes in an anatomical pattern consistent with the concept of psychosis as a disorder of network development.