AI can tell if a patient is approaching a psychotic episode through analyzing what the individual says, potentially better than humans can. Automation offers great standardization to a medical field where thoughts and state of mind have to be intuited through speech.
Researchers from the Columbia University Medical Center, IBM T. J. Watson Research Center, and New York State Psychiatric Institute published their findings in which their speech analysis system was able to correctly determine the patients who later developed psychosis. Out of the ‘clinical high risk’ patients studied over a span of two and half years, the analysis system was able to differentiate the cases in which psychosis later became onset.
Individuals considered to be at ‘clinical high risk’ for psychosis (estimated to be about 1% of the country’s 14 to 27 year olds) typically experience symptoms like unusual or tangential thinking, perceptual changes and suspiciousness. 20% of those at clinical high risk are expected to actually have a psychotic episode at some point.
Predicting those who are headed that way before the episode enables intervention measures that could delay or even stop the psychosis from fully developing. Clinical professionals attempt to intuit those cases from the speech of individuals but this novel, automated method is definitively more standardized. From examining the transcriptions of narrative interviews with patients, the analysis system was able to flag the five patients, out of the 34 studied, who had later developed full psychosis. The analysis recognized sentence meaning and structure markers (for example phrase length and frequency of determiners like “which”) that accurately predicted further psychosis development.
Red denotes those who transitioned to psychosis, blue denotes those who did not