An explainable AI latent space of brain dynamics reveals a cerebello-prefrontal signature of schizophrenia symptoms

A novel artificial intelligence framework has been developed to explain the neural dynamics of schizophrenia symptoms in a way that can be readily understood by researchers and clinicians. The study reveals a specific pattern of brain activity in schizophrenia that is centered in the cerebellum and prefrontal cortex, areas that are distinct from the patterns seen in healthy individuals.

This framework uses a technique called self-supervised contrastive learning to map the complex resting-state functional magnetic resonance imaging (rs-fMRI) data onto a lower-dimensional space, allowing for easier analysis. A new attribution method was also introduced to pinpoint the exact regions in the brain that are most influential in encoding the symptoms of schizophrenia.

When applied to two separate cohorts of patients with schizophrenia, the approach was able to accurately predict individual symptom severity and cognitive function. The attribution maps showed that the schizophrenia-specific neural signature is shaped by activity in the prefrontal, posterior cerebellar, and temporal regions, which deviate from the patterns typically seen in the brains of healthy individuals.

These patterns are dominated by neural activity in auditory, limbic, and ventral-striatal circuits. By establishing an interpretable latent space, this research provides a valuable tool for characterizing the distributed neural substrates of schizophrenia symptoms at the individual patient level, and offers a promising path toward more precise decoding of symptom severity at the brain level.