UCL researchers recommend AI language models for schizophrenia diagnosis

Automated analysis of language could help diagnose and assess psychiatric conditions

The University College London’s (UCL) Queen Square Institute for Neurology researchers have suggested that artificial intelligence (AI) language models could help diagnose schizophrenia.

Verbal fluency tasks were undertaken by 26 participants with schizophrenia and 26 control participants to determine how the automated analysis of language could assist doctors and scientists when diagnosing and assessing psychiatric conditions.

Affecting around 24 million people worldwide, schizophrenia is a debilitating and common psychiatric disorder that can cause hallucinations, delusions and behaviour changes.

Psychiatric diagnoses are currently based on talking to patients and those close to them and using limited tests, including blood tests and brain scans.

The verbal fluency tasks required participants to name as many words as possible in five minutes, in the context of ‘animals’ or words starting with the letter ‘p’.

After analysing the participants answers, researchers tested whether their responses could be predicted by the AI model, which was trained on internet text to represent the meaning of words in human context, and whether predictability was reduced in schizophrenic patients.

Researchers discovered that control participant answers were more predictable by the AI model than those generated by participants with schizophrenia.

The team believe this difference could be linked to the way the brain learns relationships between memories and ideas by storing this information in cognitive maps.

Researchers supported this theory after using brain scans to measure brain activity in parts of the brain that were involved in learning and storing cognitive maps.

Dr Matthew Nour, UCL Queen Square Institute of Neurology and University of Oxford, said: “By combining state-of-the-art AI language models and brain scanning technology, we are beginning to uncover how meaning is constructed in the brain, and how this might go awry in psychiatric disorders.”

The UCL team, along with the University of Oxford, will use the AI model across more diverse speech settings, in a larger sample of patients, to determine whether it would be useful for clinical use.