NYU School of Medicine Research Reveals AI Can Diagnose PTSD with Voice Analysis


New York University School of Medicine researchers are in hot pursuit of demonstrating how Artificial Intelligence (AI) can detect post-traumatic stress disorder (PTSD) in veterans by analyzing their voices. A recent study published in Depression and Anxiety and picked up by EurekAlert reveals that an AI-based tool can distinguish, with 89% accuracy, between the voices of those with or without PTSD.

The article notes up to 12% of people in some struggling countries suffer PTSD. Moreover, up to 70% of adults worldwide go through some form of traumatic experience or another that can form the basis for PTSD.  Traditional ways to diagnose PTSD include a clinical interview or a self-reported assessment. Both of these approaches are inherently prone to biases. Researchers seek to develop an objective, measurable—physical markers of PTSD progression. The team reports in the press release that progress on this front has been slow.

The NYU team used the random forests technique.  The AI essentially infuses into the tool the ability to learn how to classify individuals based on examples and leveraged for the study involving 53 Iraq and Afghanistan veterans with military-service- related PTSD.  The researchers leveraged voice software from SRI international (inventors of Siri) to produce 40,526 speech-based features captured in short spurts of talk—the AI program sifted through the data for patterns. The random forest program “linked patterns of specific features with PTSD.

As we contemplate what Dr. Marmar and team are undertaking, along with hundreds if not thousands of other AI advancements, the implications are thought provoking, to say the least. With ongoing move to all things digital vast pools of data accumulate for systematic and ongoing “training” by various AI tools and systems. Of course we are in the early stages of this trajectory, but we believe at TrialSite News the inevitable march toward AI-driven diagnostics represents an inevitable future. As this science and associated technologies are matured and hardened imagine how it will change how health care providers see individuals; how they will be diagnosed and cared for in locations short of clinical or medical staff (think about the clinical research implications).

Consider already we face shortages of behavioral health professionals out in the field.  Imagine when this type of tool and associated bundle (people, process and technology) are deployed in a focused  therapeutic, indication or specific treatment context—efficiently and effectively collecting data, learning, diagnosing and treating.  AI directed treatment nodes can be situated in health centers or even in remote locations; nodes in an integrated network that is run by AI– continuously learning and advancing its ability to systematically and accurately detect and identify specific ailments or conditions—such as PTSD. The very term “health system” triggers a connotation with broader, bolder and deeper and more radical implications and depending on one’s point of view the topic becomes incredibly exciting or scary—or both.

Lead Research/Investigator

Charles R. Marmar, MD, the Lucius N. Littauer Professor and chair of the Department of Psychiatry, NYU School of Medicine