AI spots early psychosis risk in youth

Singapore researchers have developed an AI framework that analyzes casual conversations to identify young people at high risk of developing psychosis. The system, developed by Nanyang Technological University, Singapore, and the Institute of Mental Health, combines emotional signals from large language models with traditional machine learning. In a study of 400 participants, the tool achieved up to 93% sensitivity.

How the AI Works

The framework, named LEMon-AID, uses LLMs to assess a person’s emotional state from speech transcripts. It extracts subtle emotional profiles, including joy, trust, fear, and sadness. The system generates risk assessments and explanations in natural language.

The research received funding from the Ministry of Health through the National Medical Research Council Office. It is described as the first of its kind in Singapore. The study analyzed transcripts from 429 participants aged 14 to 29 who were part of a longitudinal study on youth psychosis risk.

Study Findings

Researchers first tested if standard LLMs could predict psychosis risk directly from speech. Each participant provided a 45-second speech sample in response to a neutral prompt. The study found that standard AI models performed no better than chance. They failed to identify most at-risk individuals.

The team then developed LEMon-AID. This hybrid approach successfully identified participants at high risk of psychosis with up to 93% sensitivity. “This potentially allows for timely clinical assessment, intervention and support to improve long-term outcomes,” said first author Matheus Calvin Lokadjaja, a research assistant at NTU’s Lee Kong Chian School of Medicine.

Dr. Wilson Goh, an assistant professor at the same school and study co-lead investigator, highlighted a key strength: privacy. The framework uses inferred emotional vectors, not the original speech. However, Lokadjaja noted the framework’s overall accuracy was 67%, indicating limitations like the risk of false positives.

Still, the findings suggest LEMon-AID has “strong potential” as an early screening tool. The researchers believe it could help clinicians triage patients and identify those who need closer monitoring.

Why It Matters

Early identification of psychosis risk is difficult because teenagers “rarely articulate mental health struggles directly,” the research team stated. They sought to see if AI could detect risk from casual conversation. LEMon-AID “connects emotional patterns with mental health risks while keeping patient data private and providing clear, understandable explanations,” Dr. Goh explained.