The machine learning reveals that the words "sound" predict psychosis

A machine learning method discovered a hidden clue in people's language, predictive of the later appearance of a psychosis – the frequent use of words associated with sound. An article published by the journal npj schizophrenia published the findings of scientists from Emory University and Harvard University.

Researchers have also developed a new method of machine learning to more accurately quantify the semantic richness of the conversation language, a known indicator of psychosis.

Their results show that an automated analysis of the two linguistic variables – more frequent use of sound-related words and speaking with low or unspecified semantic density – can predict whether a person at risk will develop psychosis with 93% accuracy. .

Even trained clinicians did not notice how people at risk for psychosis used more words associated with sound than the average, although abnormal hearing perception is a preclinical symptom.

"Trying to hear these subtleties in conversations with people, it's like trying to see microscopic germs with the eyes," says Neguine Rezaii, first author of the article. "The automated technique we have developed is a very sensitive tool for detecting these hidden patterns.It's like a microscope to signal the warning signs of a psychosis."

Rezaii began working on paper while she was a resident of the Department of Psychiatry and Behavioral Sciences at Emory School of Medicine. She is currently a member of the Department of Neurology at Harvard Medical School.

"We previously knew that people's language exhibited subtle features of future psychosis, but we used machine learning to uncover hidden details about these characteristics," said lead author Phillip Wolff. professor of psychology at Emory. Wolff's lab focuses on linguistic semantics and machine learning to predict decision-making and mental health.

"Our discovery is new and adds to the evidence showing the potential of using machine learning to identify linguistic abnormalities associated with mental illness," says co-author Elaine Walker, a professor of psychology and psychology. Neuroscience at Emory, who studies the evolution of schizophrenia and other psychotic disorders. .

Schizophrenia and other psychotic disorders generally appear in their early twenties, with precursor signs, called prodromal syndrome, beginning around 17 years of age. About 25 to 30% of young people who meet the criteria for a prodromal syndrome will develop schizophrenia or another psychotic disorder. .

With the aid of structured interviews and cognitive tests, trained clinicians can predict psychosis with an accuracy of nearly 80% in those who suffer from a prodromal syndrome. Research in machine learning is one of many ongoing efforts to streamline diagnostic methods, identify new variables, and improve forecast accuracy.

Currently, there is no cure for psychosis.

"If we can identify people at risk earlier and use preventative interventions, we may be able to address deficits," says Walker. "There is good data showing that treatments such as cognitive behavioral therapy can delay the onset and perhaps even reduce the onset of psychosis."

For this article, researchers first used machine learning to establish "standards" for conversational language. They have powered a computer program with the online conversations of 30,000 users of Reddit, a social media platform where people have informal discussions on various topics. The software, called Word2Vec, uses an algorithm to change individual words into vectors, assigning each one a location in a semantic space according to its meaning. Those with similar meanings are closer to each other than those whose meanings are very different.

The Wolff laboratory has also developed a computer program that allows researchers to describe what is known as "vector decompression" or semantic density analysis of word usage. Previous works measured the semantic coherence between sentences. Vector decompression allowed researchers to quantify the amount of information contained in each sentence.

After generating a "normal" database, the researchers applied the same techniques to the diagnostic interviews of 40 participants led by qualified clinicians, as part of the multi-site longitudinal study on the North American prodrome (NAPLS). ), funded by the National Institutes of Health. NAPLS is focused on young people at high clinical risk of psychosis. Walker is the principal investigator of NAPLS at Emory, one of nine universities involved in this 14-year project.

Automated analyzes of participant samples were then compared to the normal core sample and longitudinal data on eventual conversion of participants to psychosis.

The results showed that a higher than normal use of sound-related words, combined with a higher rate of use of words with a similar meaning, meant that psychosis was probably at the same time. horizon.

The strengths of the study include the ease of use of two variables – both based on sound theoretical foundations – the replication of results in a reserve dataset, and the high accuracy of its predictions, superior to 90%.

"In the clinical field, we often lack precision," says Rezaii. "We need more quantified and objective ways to measure subtle variables, such as those hidden in the use of language."

Rezaii and Wolff are now assembling larger data sets and testing the application of their methods to various neuropsychiatric diseases, including dementia.

"This research is interesting not only for its potential to say more about mental illness, but also to understand how the mind works – how it associates ideas," Wolff said. "The machine learning technology is progressing so rapidly that it gives us tools to analyze data from the human mind."