Researchers from the University of Tokyo, as part of an international consortium, have developed a machine-learning tool that can predict the onset of psychosis by classifying MRI brain scans. The tool was trained on scans from over 2,000 people across 21 global locations, with approximately half of the participants identified as being at high risk of developing psychosis. The classifier demonstrated 85% accuracy in differentiating between those at risk and those who later experienced overt symptoms using training data, and 73% accuracy using new data.
Psychosis, which includes symptoms such as delusions, hallucinations, and disorganized thinking, can be triggered by various factors including illness, injury, trauma, substance use, medication, or genetic predisposition. Although treatable and with a high chance of recovery, it can be challenging to identify individuals in need of help, particularly during the adolescence and early adulthood years when the brain and body are undergoing significant changes.
According to Associate Professor Shinsuke Koike, identifying individuals who will go on to develop psychotic symptoms is crucial for early intervention. Only about 30% of clinically high-risk individuals eventually experience overt symptoms, while the remaining 70% do not. Therefore, there is a need for a tool that can detect not only subclinical signs but also biological markers.
This machine-learning tool developed by the consortium uses brain MRI scans to identify individuals at high risk of psychosis before it manifests. Previous studies using MRI scans have shown structural differences in the brains of individuals after the onset of psychosis, but this is the first time that such differences have been identified in those at high risk but without symptoms.
The research team assembled a large and diverse group of adolescent and young adult participants from various institutions and countries. Challenges in MRI research for psychotic disorders arise from differences in brain development and MRI machines, making it difficult to obtain accurate and comparable results. Additionally, distinguishing between typical developmental changes and those due to mental illness in young people can be challenging.
By training a machine-learning algorithm on the brain anatomy patterns of the participants, the researchers were able to classify them into healthy controls and those at high risk of psychosis. The tool achieved 85% accuracy in training and 73% accuracy in predicting high-risk individuals in the final test using new data. Based on these results, providing brain MRI scans for clinically high-risk individuals could aid in predicting future onset of psychosis.
To make the classifier more robust and applicable to real-life clinical settings, the team plans to test it with new data from different sites and machines. If successful, this could enhance the classification of MRIs and improve early intervention efforts.
The findings from this research have significant implications for identifying individuals at risk of psychosis and facilitating timely interventions. By detecting early signs of psychosis, healthcare professionals can provide appropriate support and treatment, leading to better outcomes and minimizing negative impacts on individuals’ lives.
