AI tool diagnoses veterans with PTSD by analyzing speech

A statistical machine learning technique has been shown to be 89 percent accurate in distinguishing between the voices of veterans with post-traumatic stress disorder and those without PTSD.

Researchers at the NYU School of Medicine—in collaboration with SRI International—have developed an artificial intelligence tool that is able to link patterns of specific voice features with PTSD, a medical diagnosis defining symptoms that last at least a month after experiencing a traumatic event.

“The classifier assigns higher probabilities of PTSD to those with features indicating speech that is slower, more monotonous, and less change in tonality and activation,” finds a study published this week in the journal Depression and Anxiety.

Voice recordings of veterans with PTSD, as well as those without the condition, were fed into software from SRI International, which generated speech-based features that the AI tool analyzed for patterns.

Symptoms of PTSD include remembering or reliving the trauma, feeling numb and withdrawn, and having forms of anxiety that interfere with daily life. According to the study’s authors, the diagnosis of PTSD is typically based on clinical interviews or self‐report measures. However, researchers contend that both approaches are subject to under‐ and over‐reporting of symptoms.

“An objective test is lacking,” states their study. “We have developed a classifier of PTSD based on objective speech‐marker features that discriminate PTSD cases from controls.”

NYU Hospital

Also See: VA online tool helps veterans compare PTSD treatments

“Our findings suggest that speech-based characteristics can be used to diagnose this disease, and with further refinement and validation, may be employed in the clinic in the near future,” says senior study author Charles Marmar, MD, the Lucius N. Littauer Professor and chair of the NYU School of Medicine’s Department of Psychiatry.

As a next step, researchers plan to apply for government approval to use the AI tool in a clinical environment.

“Speech is an attractive candidate for use in an automated diagnostic system, perhaps as part of a future PTSD smartphone app, because it can be measured cheaply, remotely, and non-intrusively,” adds lead author Adam Brown, adjunct assistant professor in NYU’s Department of Psychiatry.

For reprint and licensing requests for this article, click here.