Neuroanatomical subtyping for schizophrenia with machine learning

Schizophrenia is a heterogeneous disorder with significant variability in neurobiological and clinical presentations. In this study, we aimed to investigate neuroanatomical subtypes of schizophrenia using a data-driven machine-learning algorithm. Structural MRI data from 222 participants (136 schizophrenia patients and 86 healthy controls) were analyzed. Subtypes were identified using HYDRA (Heterogeneity Through Discriminative Analysis), a semi-supervised machine learning algorithm designed to reveal disease-related patterns while minimizing the influence of normal anatomical variation followed by voxel-based morphometry (VBM) analysis to compare these subtypes with healthy controls. The study identified two subtypes among schizophrenia patients. Subtype 1 showed widespread lower grey matter volumes in several cortical regions, mainly in the insula, cingulate, frontal, and temporal regions. Subtype 2 demonstrated increased subcortical volumes, pallidal volumes relative to controls and thalamus, hippocampus relative to subtype 1. Despite significant neuroanatomical differences, the subtypes did not differ in demographic or clinical characteristics. These findings highlight the potential of machine learning to disentangle structural heterogeneity in schizophrenia, offering a refined framework for neuroanatomical subtyping. Identifying distinct subtypes may contribute to personalized treatment approaches and enhance the precision of future clinical and research efforts.