Imagine this scenario: a patient in a primary care office is due for a mental health screening, but the nursing staff aren’t trained in assessment of psychotic disorders, so psychosis is not evaluated. 

What if AI had the capability to be part of the screenings, helping to detect psychosis? 

Dr. Roman Kotov of the Department of Psychiatry in Stony Brook University’s Renaissance School of Medicine and Dr. Sean Clouston from the Department of Family Population & Preventative Medicine and the Program in Public Health in Stony Brook University’s Renaissance School of Medicine are among the nine researchers that conducted a study, “Behavioral Markers of Psychotic Disorders: Using Artificial Intelligence to Detect Nonverbal Expressions in Video.”

The foundation of this study was the use of artificial intelligence measures of non-verbal expressions, or “NveAI.” NveAI was tested to see if it had the capability to detect the differences in facial expressions and head movement that patients with schizophrenia have in comparison to those with other psychotic disorders, and those with no history of psychotic disorders.

“What we’re trying to do is identify objective measures that would tell us when things are not quite right,” says Clouston. 

Participant video data was obtained from the 25-year follow-up of the Suffolk County Mental Health Project, a project that follows patients who were admitted to a psychiatric hospital and experienced psychosis for the first time in the early 90s. Half of the project’s participants were diagnosed with schizophrenia. Dr. Kotov currently runs the project. 

“25 years later, we know a lot about the illness’s course, so we have a very precise clinical description. Now we’ve collected video during an interview about the patients’ symptoms, occupation, relationships, family, etc.,” says Kotov. 

The video data participants were split into three groups: those with schizophrenia/schizoaffective disorder, those with other psychotic disorders, and those that were never psychotic. Using FaceReader, a facial expression analysis software, the participants’ non-verbal expressions were assessed. 

Ultimately, nveAI was able to detect the abnormalities in facial expressions and head movement of the schizophrenia/schizoaffective disorder group. In consideration of nveAI’s developments, as research expands, it may be given a larger role in the medical office. 

“Eventually we may have a diagnostic biomarker that outperforms symptoms, and that would be great for people,” says Clouston.

“At this point, the research has left computer science laboratories, and is ready for validation in clinical research,” says Kotov. “We hope that it will prove itself in research and be translated into clinical care.” 

-Sara Giarnieri, Communications Assistant