According to a new study, a uniquely designed computer program can help in diagnosing post-traumatic stress disorder, or PTSD, in veterans by examining their voices.
The study, featured online in the journal Depression and Anxiety on April 22nd, 2019, revealed that an artificial intelligence tool has the ability to differentiate between the voices of veterans with or without the PTSD condition, with an accuracy of 89%.
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.
Charles R. Marmar, Study Senior Author, MD, and Chair, Department of Psychiatry, NYU School of Medicine
Marmar is also the Lucius N. Littauer Professor.
Globally, over 70% of adults experience a traumatic event at some stage in their lives, with around 12% of people suffering from PTSD in some struggling nations. When reminded of a triggering event, PTSD patients experience strong and persistent distress.
A PTSD diagnosis is typically established by a self-report assessment or clinical interview, both of which are innately prone to biases, stated the study authors. This spurred the efforts to create objective, physical, measurable markers of the progression of PTSD, just like lab values for medical conditions; however, progress has been rather slow.
Learning How to Learn
In the present analysis, a statistical/machine learning method, known as random forests, was used by the researchers. This technique has the potential to “learn” how to categorize individuals on the basis of examples. AI programs like these develop mathematical models and “decision” rules that allow decision-making with growing precision as the quantity of training data grows.
Initially, the team recorded typical, hours-long diagnostic interviews, referred to as Clinician-Administered PTSD Scale, or CAPS for short, of 53 Afghanistan and Iraq veterans suffering from PTSD related to military service, and also those of 78 veterans without this disorder. Subsequently, the recordings were fed into SRI International’s voice software to produce a total of 40,526 speech-based features that were recorded in short spurts of talk, which was sifted through by the team’s AI program for patterns. SRI International also invented Siri.
The arbitrary forest program linked patterns of particular voice features with PTSD, such as a lifeless, metallic tone and less clear speech, both of which had been traditionally reported anecdotally as useful in diagnosis.
Although the present analysis did not examine the disease mechanisms behind PTSD, the hypothesis is that brain circuits that process muscle tone and emotion are changed by traumatic events, influencing the voice of a person.
Going ahead, the researchers are planning to train the new AI voice tool with additional information, further verify it on an autonomous sample, and then apply for government sanction to use the AI voice tool in clinical settings.
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.
Adam Brown, Study Lead Author, PhD, Adjunct Assistant Professor, Department of Psychiatry, NYU School of Medicine
The speech analysis technology used in the current study on PTSD detection falls into the range of capabilities included in our speech analytics platform called SenSay Analytics™. The software analyzes words - in combination with frequency, rhythm, tone, and articulatory characteristics of speech - to infer the state of the speaker, including emotion, sentiment, cognition, health, mental health and communication quality. The technology has been involved in a series of industry applications visible in startups like Oto, Ambit and Decoded Health.
Dimitra Vergyri, Director, Speech Technology and Research (STAR) Laboratory, SRI International
Together with Brown and Marmar, study authors from the Department of Psychiatry were Meng Qian, Eugene Laska, Meng Li, Carole Siegel, and Duna Abu-Amara. Authors of the study from SRI International were Andreas Tsiartas, Colleen Richey, Dimitra Vergyri, Bruce Knoth, and Jennifer Smith. Brown is also an associate professor of psychology at the New School for Social Research.
The U.S. Army Medical Research & Acquisition Activity (USAMRAA), Telemedicine & Advanced Technology Research Center (TATRC) grant W81XWH- ll-C-0004, and the Steven and Alexandra Cohen Foundation supported the study.