Schizophrenia is a common and disabling neuropsychiatric disorder whose symptoms can include hallucinationshallucinationsdelusions, confused thoughts, and behavioral changes (depending on National Health Service in the United Kingdom). At a clinical level, the disease is notably characterized by difficulties in abstract reasoning and less coherent language than in unaffected patients.

Researchers at the Institute of NeurologyUniversity of London (UCL) and Oxford used linguistic models ofartificial intelligenceartificial intelligence (AI) that can characterize signatures, even subtle ones, in the speech of patients diagnosed with schizophrenia. Published in PNAS, the study investigated how automated language analysis could help the medical profession diagnose and evaluate psychiatric disorders more accurately and sooner. In fact, the current psychiatric diagnosis of schizophrenia is based almost exclusively on interviews with patients and their loved ones.

More or less predictable words

Twenty-six patients with schizophrenia (with confirmed diagnosis) and twenty-six control participants performed verbal fluency tasks, consisting of quoting for five minutes as many words as possible belonging to the category “animals” or beginning with the letter “p.” This involves comparing the word choice and spontaneity of the two groups.

The researchers used natural language processing (or NLP for Natural language processing, in English), who were trained with large amounts of text to represent the meaning of words like humans. Result: The answers given by control participants were more predictable for the AI ​​model than those generated by people with schizophrenia, and this difference was greater in patients with more severe symptoms.

A possible explanation lies in the way in which brainbrain It encodes the relationships between memories and ideas and stores this information in “cognitive maps.” However, the clinical characteristics of schizophrenia – from thought disorder to delirium – would reflect a disorganization of these conceptual representations according to the authors.

Will artificial intelligence tools be used in clinics soon?

“ This work shows the potential of applying AI language models to psychiatry, a medical field closely linked to language, has declared Dr. Matthew Nour, lead author of the study. By combining these cutting-edge artificial intelligence models and brain scanning technology, we are beginning to uncover how meaning is constructed in the brain and how it can go wrong in psychiatric disorders. »

Scientists at UCL and Oxford now plan to use this technology in a larger sample of patients and in more varied speech contexts, to test whether it could be useful clinically. “ If these tools prove safe and robust, I believe they will begin to be implemented over the next decade. », Specified Dr. Nour.