Structural MRI-Based classification of NMDAR encephalitis and its expression in schizophrenia

This project investigates the neuroanatomical overlap between anti-N-methyl-D-aspartate receptor (NMDAR) antibody encephalitis and schizophrenia using structural magnetic resonance imaging (MRI) and machine learning. Anti-NMDAR encephalitis is an autoimmune disorder that frequently presents with prominent psychiatric symptoms, including psychosis. While NMDAR antibodies have been detected in a subset of individuals with first-episode psychosis, the extent to which immune-related mechanisms contribute to structural brain alterations in schizophrenia remains unclear. Clarifying whether schizophrenia shares a neuroanatomical signature with NMDAR encephalitis may help identify biologically meaningful subgroups and inform treatment strategies. The primary aim of this study is to determine whether patients with anti-NMDAR encephalitis can be distinguished from healthy controls based on regional structural MRI measures using supervised machine learning. We will train a linear Support Vector Machine (SVM) classifier using Neuromorphometrics atlas-derived grey matter, white matter, and cerebrospinal fluid volumes that have been corrected for total intracranial volume. Feature preprocessing, harmonisation (age, sex, protocol/site), dimensionality reduction, and model optimisation will be implemented within a nested cross-validation framework in NeuroMiner. Statistical inference will be performed using permutation testing. The second aim is to investigate whether individuals with schizophrenia express an NMDAR-like structural brain signature. To address this, the trained NMDAR vs healthy control classifier will be applied, without retraining, to an independent schizophrenia cohort (Australian Schizophrenia Research Bank). The resulting SVM decision values will be interpreted as continuous “NMDAR-expression scores,” reflecting the degree to which an individual’s neuroanatomical pattern resembles the NMDAR signature. We will quantify the proportion of schizophrenia participants classified on the NMDAR side of the decision boundary and examine the distribution of expression scores at the group level. The third, exploratory aim is to examine whether schizophrenia individuals with stronger NMDAR-like structural signatures exhibit elevated peripheral inflammation. Circulating cytokines and C-reactive protein will be analysed in relation to continuous decision values and subgroup classification (NMDAR-like vs non-NMDAR-like). These analyses are hypothesis-generating and will include correction for multiple testing. All data used in this project have been previously collected. The anti-NMDAR encephalitis cohort (n = 192; 96 patients and 96 healthy controls) was obtained from Charité – Universitätsmedizin Berlin. The independent schizophrenia cohort (n = 286; 185 patients and 101 healthy controls) was drawn from the Australian Schizophrenia Research Bank. The study is cross-sectional and observational, and no new data collection will be conducted. Overall, this project aims to determine whether a biologically defined autoimmune encephalitis signature can be detected in structural MRI data and whether a subset of schizophrenia patients shares this neuroanatomical pattern. Identifying such overlap could contribute to improved stratification of psychotic disorders and advance understanding of immune-related mechanisms in severe mental illness.