Short Summary/Abstract:

Schizophrenia (SCZ) is a heterogeneous psychiatric disorder with substantial variability in clinical presentation, treatment response, and neurobiological profiles. Traditional diagnostic approaches treat SCZ as a single entity, potentially obscuring meaningful biological subtypes that could inform personalized treatment strategies. Recent advances in deep learning, particularly variational autoencoders (VAEs), offer powerful tools for discovering latent patterns in high-dimensional neuroimaging data that may correspond to distinct disease subtypes.

This study aims to identify neuroanatomically-defined subtypes of schizophrenia using cortical thickness measurements analyzed through VAE-based dimensionality reduction. By training VAE on combined healthy control and SCZ patient data, we will learn latent representations that capture the full spectrum of neuroanatomical variation, and expect to identify distinct subgroups within the patient population. This data-driven approach may reveal novel subtypes that cut across traditional diagnostic boundaries and provide insights into the neurobiological heterogeneity underlying schizophrenia.