AI Language Models Aid in Diagnosing Schizophrenia

[baoshi.rao]

Scientists at University College London's Institute of Neurology have developed new tools based on AI language models that can characterize subtle features in the speech of patients diagnosed with schizophrenia. This study, published in the Proceedings of the National Academy of Sciences (PNAS), aims to understand how automated language analysis could help doctors and scientists diagnose and assess mental disorders.

Currently, the diagnosis of mental disorders relies almost entirely on conversations with patients and their close contacts, with tests like blood work and brain scans playing minimal roles. However, this diagnostic imprecision hinders deeper understanding of mental illness causes and monitoring of treatment effectiveness.

Image credit: AI-generated image licensed from Midjourney

The researchers had 26 participants with schizophrenia and 26 control participants complete two verbal fluency tasks, asking them to name as many words as possible from the "animal" category or starting with "p" within five minutes. To analyze responses, the team used an AI language model trained on vast internet text to represent word meanings similarly to human cognition. They tested whether spontaneously recalled words could be predicted by the AI model and whether predictability was reduced in schizophrenia patients.

Results showed control participants' responses were indeed more predictable by the AI model than those of schizophrenia patients, with this difference being most pronounced in patients with more severe symptoms. Researchers suggest this difference may relate to how the brain learns relationships between memories and thoughts, and stores this information in "cognitive maps." In the study's second part, brain scans measured activity in regions involved in learning and storing these cognitive maps, supporting this theory.

Lead author Dr. Matthew Nour stated that with the emergence of AI language models like ChatGPT, automated language analysis has become available to doctors and scientists. This work demonstrates the potential of applying AI language models to psychiatry, which is closely related to language and meaning. The team plans to extensively test this technology in larger patient samples for clinical applications. If proven safe and reliable, these tools could begin clinical use within the next decade.