The ability to reliably predict the onset of psychosis in high-risk youth has, traditionally, been elusive. But new speech analysis software tools are offering specialists a window into the minds of young people who may be at risk.
“Computerized speech analysis generates biomarkers that we can add to our clinical toolbox. These biomarkers can be used to screen young people for language disturbances that might be predictors of schizophrenia and other psychoses,” said Cheryl Corcoran, MD, Director of Mount Sinai’s Psychosis Risk Program. “While clinicians routinely detect disorganized speech on the basis of clinical observation, artificial intelligence (AI), with its automated speech analysis algorithms, has the ability to pick up on subtle impairments we might miss. It adds to our assessment of whether a teen or young adult is at risk for psychosis and might be helped by early intervention.”
To achieve this goal, Dr. Corcoran has been working with Guillermo Cecchi, PhD, research scientist and manager of computational psychiatry and neuroimaging at IBM Corp. Dr. Cecchi is a pioneer in the use of computational linguistics to quantify psychiatric conditions such as mania, prodromal psychosis, and schizophrenia from short speech samples.
The collaboration between Drs. Corcoran and Cecchi and their respective research teams has led to the creation of algorithms that apply a combination of semantic and syntactic analyses to spoken language that has been recorded and transcribed from interviews with several cohorts of at-risk youth.
Critical to determining progression to psychosis is the detection of language disturbances—such as being tangential or going off track, and speaking simply with little content—that are early core indicators of schizophrenia. In a paper they published in 2018 in World Psychiatry, Drs. Corcoran and Cecchi found that computer-based speech analysis predicted with an accuracy of 83 percent which youths would go on to develop psychosis within two years, and cross-validated this language classifier in independent psychosis risk samples.
“Through this technology we’re able to identify patterns of language that precede onset of psychosis, including reduction in coherence and complexity,” said Dr. Corcoran, who was the lead author of the study. “And because speech is easy to collect and inexpensive to analyze through automated techniques, it could potentially be used across psychiatry to inform clinicians when individuals are at risk for suicide, for example, or Alzheimer’s disease, post-traumatic stress disorder, or other mental health disorders.” Working with colleagues, Dr. Corcoran has even begun to use computer-based language analysis to screen populations in other countries, including Brazil and China. Clinical data from those sites show some of the same speech patterns that are known to be early indicators of psychosis in English-speaking groups.
“We believe our work represents just the initial stage in the emerging science of advanced behavioral analysis through natural language processing,” said Dr. Corcoran. “And this could present psychiatry with a rich opportunity to move beyond reliance on self-report and clinical observation to more objective-and-effective measures of pathophysiology in patients.”
Cheryl Corcoran, MD
Director of Mount Sinai’s Psychosis Risk Program, and Associate Professor of Psychiatry