AI diagnoses PTSD by analyzing veterans’ voices

Working with SRI International, the California tech lab that gave the world Siri, mental health specialists at NYU Langone have developed an AI-based tool that uses voice analysis to diagnose posttraumatic stress disorder—potentially via telemedicine.

Senior study author Charles Marmar, MD, and co-authors described their work in a study published online April 22 in Depression & Anxiety.

In introducing their findings, the authors pointed out that the usual means of diagnosing PTSD—clinical interviews and self-reporting by patients—are subjective and thus prone to inaccuracy.

They set out to create an algorithmic PTSD classifier that could objectively isolate numerous PTSD markers in everything from choice of words to tone of voice to rhythms of verbalization.  

The team drew from audio recordings of 52 Iraq and Afghanistan war veterans with known PTSD and 77 peers who were free of the condition.

Feeding the recordings into SRI software, they were able to identify more than 40,000 speech-based features for the algorithm to analyze.

Marmar and team found the tool distinguished the PTSD patients from the controls at an accuracy rate of nearly 90%.

In a press release sent by NYU Langone, co-lead author Adam Brown, PsyD, said 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 nonintrusively.”

Marmar added that, with further validation and fine-tuning, speech analysis algorithms may soon be in use in everyday clinical care for PTSD patients.

The study was supported by a U.S. Army research center that has telemedicine as an area of concentration.