AI can predict psychosis risk in everyday language
18 June, 2019
People's language could reveal clues about their future risk of developing psychosis. Scientists concluded this after studying the subtle features of people's everyday speech.
Researchers at Emory University in Atlanta, GA, and Harvard University in Boston, MA, used a machine-learning technique to analyze language in a group of at-risk young people.
They found that they could predict which individuals would go on to develop psychosis with an accuracy of 93%.
A recent npj Schizophrenia study paper describes how the team developed and tested the method.
Senior study author Phillip Wolff, a professor of psychology at Emory University, explains that earlier research had already established that "subtle features of future psychosis are present in people's language." However, he noted, "we've used machine learning to actually uncover hidden details about those features."
He and his colleagues devised their machine-learning approach to measure two linguistic variables: semantic density and use of words relating to sound.
They concluded that "conversion to psychosis is signaled by low semantic density and talk about voices and sounds."
Low semantic density is a measure of what the team refers to as "poverty of content" or vagueness.
"This work," note the authors, "is a proof of concept study demonstrating that indicators of future mental health can be extracted from people's natural language using computational methods."
Machine learning and psychosis symptoms
Machine learning is a type of artificial intelligence in which computers "learn from experience" without scientists having to program the learning explicitly.
A machine-learning system looks for patterns in a known set of data and decides which patterns identify specific features. Having "learned" what these features are, it can then tirelessly identify them in a new set of data.
Machine learning can spot patterns in people's use of language that even doctors who have undergone training to diagnose and treat those at risk of psychosis may not notice.
"Trying to hear these subtleties in conversations with people is like trying to see microscopic germs with your eyes," explains first study author Neguine Rezaii, a fellow in the Department of Neurology at Harvard Medical School.
However, it is possible to use machine learning to find certain subtle patterns hiding in people's language. "It's like a microscope for warning signs of psychosis," she adds.
Rezaii began working on the study while she was a resident in the Department of Psychiatry and Behavioral Sciences at Emory University School of Medicine.
Psychosis is a state of mind in which it can be difficult to tell the difference between what is real and what is not.
When a person enters this state of mind, doctors call it a psychotic episode. During such an episode, people experience disturbed perceptions and thoughts. Delusions and hallucinations are common symptoms of psychosis.
During a psychotic episode, a person may display inappropriate behavior or talk incoherently. In addition, they may experience sleep disruption and become socially withdrawn, depressed, and anxious.
In the United States, about 3% of people will experience a period of psychosis during their lifetime, according to figures from the National Institute of Mental Health, which is one of the National Institutes of Health (NIH).
TAG(s):