AI for the earliest possible diagnosis of psychosis

An algorithm helps in the early detection of psychoses

Artificial intelligence (AI) could make a significant contribution to the early detection of mental illnesses. By combining artificial and human intelligence, better prevention of psychoses in young people could be achieved, reports the Max Planck Institute of the results of a current study.

For the study, the research team led by Professor Dr. Nikolaos Koutsouleris from the Max Planck Institute for Psychiatry used machine learning models that analyze clinical and biological data with the assessments of the treating physicians. This led to a significant improvement in early detection compared to the prediction by doctors alone. The study results were published in the journal JAMA Psychiatry.

What is a psychosis? Psychosis is a condition that affects the way your brain processes information. It makes you lose touch with reality. You can see, hear or believe things that are not real. Psychosis is a symptom, not a disease. It can be caused by mental or physical illness, substance abuse, extreme stress, or trauma. Psychotic disorders, such as schizophrenia, include a psychosis that usually first appears in late teenage years or early adulthood. Young people are particularly affected, but doctors don’t know why. Even before the first psychosis episode (FEP), you can show slight changes in your behavior or thinking. This is called the prodromal period and can last for days, weeks, months, or even years.

The risk of a bad outcome is often underestimated

Although experts make very precise predictions about positive disease courses, the frequency of bad courses in which relapses occur is often underestimated, reports the Max Planck Institute. Researchers therefore asked whether machine learning could improve the transition to psychosis in patients with clinically high-grade or recent depression.

Improved prediction of psychosis

Using 334 people with clinical high-risk or recent depression and 334 people as a control group, the research team investigated whether the combination of specialist assessment and computer-aided evaluation of all clinical, neurocognitive, imaging and genetic information could improve the prediction of psychosis.

“The lack of prognostic sensitivity of the clinicians, measured by a false-negative rate of 38.5%, was reduced to 15.4% by the sequential prognostic model,” the researchers report on their results. It was the combination of AI and and specialist assessment that optimized the prediction, the researchers said. “This allows us to improve the prevention of psychoses, especially in young patients at high risk or with emerging depression, and to manage interventions in good time,” emphasizes Professor Koutsouleris.

Decision aid for practice

“The algorithm does not replace the treatment by the medical staff, but rather aids in decision-making and gives recommendations as to whether further examinations are useful in individual cases,” the Max Planck Institute continues. On this basis, it can be decided at an early stage, for example, which patients require therapeutic intervention and which are not.

Integration into the clinical workflow

“The results of our study can help drive a two-way and interactive process of clinical validation and refinement of prognostic tools in real-life early intervention services,” concluded Professor Koutsouleris. In predicting psychosis, an individualized prognostic workflow that “sequentially integrates risk assessments from algorithms and clinicians” could lead to significant improvements, the researchers write. However, the proposed workflow would have to be fully validated before clinical implementation.


Leave a Comment

This site uses Akismet to reduce spam. Learn how your comment data is processed.