Generalisable functional imaging classifiers of schizophrenia have multifunctionality as trait, state, and staging biomarkers

Abstract

Schizophrenia spectrum disorder (SSD) is one of the top causes of disease burden; similar to other psychiatric disorders, SSD lacks widely applicable and objective biomarkers. This study aimed to introduce a novel resting-state functional connectivity (rs-FC) magnetic resonance imaging (MRI) biomarker for diagnosing SSD. It was developed using customised machine learning on an anterogradely and retrogradely harmonised dataset from multiple sites, including 617 healthy controls and 116 patients with SSD. Unlike previous rs-FC MRI biomarkers, this new biomarker demonstrated a notable accuracy rate of 77.3% in an independent validation cohort, including 404 healthy controls and 198 patients with SSD from seven different sites, effectively mitigating across-scan variability. Importantly, our biomarker specifically identified SSD, differentiating it from other psychiatric disorders. Our analysis identified 47 important FCs significant in SSD classification, several of which are involved in SSD pathophysiology. Beyond their potential as trait markers, we explored the utility of these FCs as both state and staging markers. First, based on aggregated FCs, we built prediction models for clinical scales of trait and/or state. Thus, we successfully predicted delusional inventory scores (r = 0.331, P = 0.0177), but not the overall symptom severity (r = 0.128, P = 0.178). Second, through comprehensive analysis, we uncovered associations between individual FCs and symptom scale scores or disease stages, presenting promising candidate FCs for state or staging markers. This study underscores the potential of rs-FC as a clinically applicable neural phenotype marker for SSD and provides actionable targets to neuromodulation therapies.