A specially designed computer program can help diagnose posttraumatic stress disorder (PTSD) in veterans by analyzing their voices, revealing a new study.
Published online April 22 in the journal Depression and anxietyThe research has shown that the artificial intelligence tool can differentiate the voices of those with or without PTSD, with 89 percent accuracy.
"Our findings suggest that speech characteristics can be used to diagnose this disease, and with further training and confirmation, they can be used in the clinic in the near future," says senior author Charles R. Marmar, MD, Lucius N Littauer Professor and President of the Department of Psychiatry at the Medical School in New York.
More than 70 percent of adults in the world experience a traumatic event at some point in their lives, with about 12 percent of people in some countries suffering from PTSD. Those who have a state experience a strong, upset agitation when they remind themselves of the trigger event.
The authors of the study say that the diagnosis of PTSD is most commonly determined by clinical interviews or self-assessment, both inherently prone to bias. This led to the effort to develop objective, measurable, physical indicators of PTSD progression, similar to laboratory values for medical conditions, but progress was slow.
Learning how to learn
In the current study, the research team used the technique of statistical / machine learning, which is called random forest, which has the ability to "learn" how to classify individuals on the basis of examples. Such AI programs build decision-making rules and mathematical models that allow decision-making with increased accuracy as the amount of training data increases.
The researchers first recorded standard long-term diagnostic interviews, called PTSD Scales under Clinician's Checks, or CAPS, 53 veterans from Iraq and Afghanistan with PTSP-related military service, and 78 non-ill veterans. The recordings were then inserted into the voice software of the SRI International Institute, which also invented Siri – to gain a total of 40,526 short speech voices recorded by the AI program team through the samples.
The random forest program linked patterns of specific voice features with PTSP, including less clear speech and lifeless, metal tone, which have long been anecdotal reported as useful in diagnosis. Although current study has not investigated PTSD disease mechanisms, the theory is that traumatic events alter brain assemblies that process emotions and tone of muscle, affecting the person's voice.
Moving forward, the research team is planning to train a multi-data voice intelligence tool, further verifying it on an independent sample, and requesting government approval for the clinical use of the tool.
"Speech is an attractive candidate for use in an automated diagnostic system, perhaps as part of a future PTSD smartphone application, because it can be measured cheaply, remotely and non-intrusive," says Dr. Adam Brown, Assistant Professor at the Department of Psychiatry at the NYU Medical School.
"The speech analysis technology used in the current PTSD detection study falls within the range of capabilities included in our speech analysis platform called SenSay Analytics ™," says Dimitra Vergyri, director of speech research and research labs (STAR). "The software analyzes the words – combined with frequency, rhythm, tone, and articulate speech characteristics – to figure out the speaker's state of mind, including emotions, feelings, knowledge, health, mental health, and communication quality, a range of industrial applications visible in startups such as Oto, Ambit and Decoded Health. "
Along with Marmar and Brown, the authors of the psychiatry department were Meng Qian, Eugene Laska, Carole Siegel, Meng Li, and Duna Abu-Amara. The authors of the SRI International research were Andreas Tsiartas, Dimitra Vergyri, Colleen Richey, Jennifer Smith and Bruce Knoth. Brown is also an associate professor of psychology at New School for Social Research.
The research was supported by American Medical Research and Acquisition Activities (USAMRAA) and the Telemedicine and Advanced Technology Research Center (TATRC), issued by W81XWH-ll-C-0004, as well as the Steven and Alexandra Cohen Foundation.
Waiver: AAAS and EurekAlert! are not responsible for the accuracy of the information published in EurekAlert! by the contributing institutions or the use of any information through the EurekAlert system.