Post traumatic stress disorder (PTSD) can be a very serious problem, and it unfortunately often goes undiagnosed. New technology could help, however, as it uses artificial intelligence to determine if someone has PTSD – based on their speech.
Currently, the disorder is typically diagnosed either through patient self-reports or interviews conducted in clinics. Both of these approaches are somewhat subjective, as they're influenced by the bias of the patient or physician. With that in mind, researchers at the New York University School of Medicine set out to develop a more objective system.
Led by Dr. Charles R. Marmar, they started by recording diagnostic interviews conducted with 53 Iraq and Afghanistan war veterans who had already been diagnosed with PTSD, along with interviews conducted with 78 veterans who did not have the disorder. All of the recordings were then processed using voice software developed by SRI International, resulting in a collection of 40,526 speech-based features that were captured in "short spurts of talk."
A statistical machine-learning technique known as a random forest algorithm was then used to analyze all those short spurts, teaching itself which speech features were associated with PTSD. It ultimately determined that 18 factors, such as unclear speech and a "lifeless, metallic tone," were strong indicators of the disorder – this could be due to traumatic events changing the brain circuitry that processes emotion and muscle tone.
When the system subsequently used what it had learned to guess whether or not interviewees had PTSD, it was 89 percent accurate at doing so. The scientists now plan on using additional data to train the system further, thus raising that rate of accuracy.
"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," says Asst. Prof. Adam D. Brown, lead author of a paper on the research. That paper was published this Monday in the journal Depression and Anxiety.
Source: New York University