The machine learning technique is more accurate as the amount of training data grows.
Photo and thumbnail have been modified. U.S. Air Force photo/Tech. Sgt. Nadine Barclay.
A new study conducted by researchers at NYU School of Medicine found that an artificial intelligence (AI) tool can distinguish between the voices of those with or without post-traumatic stress disorder (PTSD) with 89% accuracy.
The research team, led by senior study author Charles R. Marmar, M.D., the Lucius N. Littauer professor and chair of the department of psychiatry at NYU School of Medicine, used a statistical/machine learning technique called random forests, which can “learn” to classify individuals based on examples.
These AI programs build “decision” rules and mathematical models that enable decision-making with increasing accuracy as the amount of training data grows.
Standard, hours-long diagnostic interviews — called Clinician-Administered PTSD Scale — were taken of 52 Iraq and Afghanistan veterans with military-service-related PTSD. There were an additional 78 interviews conducted on those without the disease.
The interview recordings were then fed into voice software from SRI International’s Speech Technology and Research Laboratory to capture more than 40,700 speech-based features.
“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,” said Dimitra Vergyri, director of SRI International’s Speech Technology and Research Laboratory.
Speech features were extracted using audio quality control, audio segmentation, extraction of fame level features, computation of spurt-level features and computation of speaker-level features.
The random forest program linked patterns of 18 specific voice features with PTSD, such as clear speech and a lifeless, metallic tone. It is said that traumatic events change brain circuits that process emotion and muscle tone, changing the person’s voice.
The AI program was 89.1% accurate and the probability of PTSD was higher for markers that indicated slower, more monotonous speech, less change in tonality and less activation.
The random forest program also had an area under the curve of 0.954.
“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,” Marmar said.
The team wants to train the AI program with more data, further validate it on an independent sample and apply for government approval to use the tool clinically.
The study, “Artificial intelligence can diagnose PTSD by analyzing voices,” was published on April 22 in the journal Depression and Anxiety.
Get the best insights in digital health directly to your inbox.