Heterogeneous structural brain signatures of methamphetamine and ketamine use disorders revealed by normative modeling.

Marked interindividual heterogeneity in substance use disorder (SUD) impedes the identification of consistent neuroanatomical alterations, limiting the development of robust markers for clinical risk stratification. We applied a normative modeling framework to explore individual neuroanatomical deviations in individuals with SUD across two distinct substance classes, and assessed their potential for classification and risk prediction. We established a developmental normative model of gray matter volume (GMV) using T1-weighted images from 1209 healthy individuals aged 18-65 years. Deviations from this normative model were then estimated for 184 individuals with methamphetamine use disorder (MUD), 94 with ketamine use disorder (KUD), and 220 healthy controls (HC) collected from two independent sites. Based on these individual deviation scores, a machine learning model was constructed to identify individuals at risk for SUD. The normative model revealed high interindividual heterogeneity in GMV deviations among individuals with MUD and KUD. Cross-substance extreme negative deviations were primarily located in regions of the default mode network (DMN). Notably, individual-level deviations could successfully distinguish both MUD and KUD from HC, and replicated in an external dataset. We further identified significant associations between abnormal GMV trajectories and neurobiological features, particularly myelination and neurotransmitter systems. These findings enhance our understanding of the heterogeneous neurobiology underlying SUD, providing potential neurobiological biomarkers for clinical diagnosis, risk stratification, and prevention.