Post-Traumatic Stress Disorder
AI Accurately Diagnoses PTSD Using Voice Analysis
NYU School of Medicine's post-traumatic stress disorder voice analysis tool
Posted April 22, 2019
Post-traumatic stress disorder (PTSD) may impact people of all ages, from all walks of life. According to the Anxiety and Depression Association of America (ADAA), 67 percent of people exposed to mass violence develop PTSD. Extreme situations such as exposure to mass gun shootings, combat, natural disasters, childhood abuse, rape, violence, crime, a severe accident, a life-threatening medical diagnosis, terrorist acts, and other traumatic incidents may give rise to the mental health condition.
PTSD symptoms may include intrusive memories and flashbacks, nightmares, changes in physical and emotional reactions, avoidance, and negative changes in thinking and mood that was caused by either witnessing or experiencing first-hand a traumatic or life-threatening event. Unlike acute stress disorder, sufferers of PTSD experience symptoms for longer than a month after the triggering event, and may continue to do so for years.
Traditional diagnosing of PTSD is performed by either a psychologist or psychiatrist through clinical interviews or a self-reporting assessment—an inherently biased methodology. In an effort to address the lack of an objective way to diagnose PTSD, a team at NYU School of Medicine announced today a revolutionary artificial intelligence (AI) tool that can analyze voices and identify those with PTSD with a high accuracy rate of 89 percent.
To create the AI solution, the researchers recorded standard diagnostic interviews known as the Clinician-Administered PTSD scale (CAPS) of 53 combat veterans suffering from PTSD and 78 veterans without PTSD. The voice recordings were uploaded into voice software from SRI International—creator of Siri, the voice-activated virtual assistant which was later acquired by Apple and incorporated into its smartphones.
“By combining advanced bioinformatics tools derived from artificial intelligence science with very high quality, high-fidelity audio recordings of the human voice spectrum, we were able to first partition the human voice spectrum into over 40,000 unique biophysical features,” said Lucius N. Littauer Professor Dr. Charles R. Marmar, senior study author and chair of the Department of Psychiatry at NYU School of Medicine in a NYU Langone Health video report. “And machine learning derived from AI, in particular, random forest machine learning (which is what we used in the study), is ideally suited to filter through and select down from tens of thousands of features—in our case, 18 features—that carry the signal of PTSD diagnosis.”
“Random forest emulates kind of the logic that a physician would use … going down a logic tree, excluding something, putting something in the possibility list, until the exhaust of the list of questions gets to what we call a ‘terminal node,’ said Dr. Eugene M. Laska in the same report. “And that node, you decide how many people enter that node. And based on the count of how many have PTSD, and how many don’t, you ascribe a possibility of PTSD.”
From the recordings, 40,526 speech features were extracted and then fed to a random forest (RF) algorithm to identify patterns. According to the research study published today in Depression and Anxiety, the likelihood of PTSD was higher for “markers that indicated slower, more monotonous speech, less change in tonality, and less activation” and the “overall correct classification rate was 89.1%.” The research team has shown that an artificial intelligence speech-based algorithm can objectively identify PTSD with a high rate of accuracy.
“Psychiatry is the only field of medicine that doesn’t rest on the foundation of laboratory medicine,” continued Marmar in the report. “So we don’t have yet, reliable blood tests for major psychiatric disorders. We’re working on them, and we think voice is one pathway into that.”
“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,” stated lead author Adam D. Brown, PsyD, adjunct assistant professor in the Department of Psychiatry at NYU School of Medicine in the report.
For the next step, the team plans to conduct additional training of the AI speech-based algorithm with more data and then testing on an independent sample. The technology is a natural fit for telemedicine and decentralized remote health care. The team envisions their AI algorithm as a clinical tool for mental health professionals to enable the diagnosing of PTSD quickly, and more objectively in the future.
Copyright © 2019 Cami Rosso All rights reserved.
References
Marmar, Charles R., Brown, Adam D., Qian, Meng, Laska, Eugene, Siegel, Carole, Li, Meng, Abu-Amara, Duna, Tsiartas, Andreas, Richey, Colleen, Smith, Jennifer, Knoth, Bruce, Vergyri, Dimitra. “Speech-based markers for posttraumatic stress disorder in US veterans.” Depression and Anxiety. 22 April 2019.
NYU Langone Health (2019, April 22). NYU School of Medicine Develops Tool that Diagnoses Post-Traumatic Stress Disorder by Voice Analysis [Press release]. Retrieved from ahttps://nyulangone.org/press-releases/nyu-school-of-medicine-develops-t…
Mayo Clinic. “Post-traumatic stress disorder (PTSD).” Retrieved 4-22-2019 from https://www.mayoclinic.org/diseases-conditions/post-traumatic-stress-di…