Psychiatrists know that they shouldn't just listen to what their patients say, but also how they say it. And now researchers at the University of Michigan have created a smartphone app that mimics that listening behavior. By analyzing the phone conversations of people with bipolar disorder, researchers say they've detected the speech patterns associated with manic and depressive episodes.
The small study included only six patients, and was intended as a proof of concept. Next the researchers want to develop an app that detects early signals of mood swings in people with bipolar disorder, allowing for prompt medical intervention. The researchers presented their paper this week at the IEEE International Conference on Acoustics, Speech, and Signal Processing.
There's growing interest in using smartphone apps in psychiatry, both to help people manage their own illnesses and to let clinicians keep a closer eye on patients with ailments like depression, schizophrenia, and PTSD. In one experiment that's rather similar to this bipolar study, researchers with the Boston company Cogito tested an app that analyzed vocal and social activity to look for symptoms of PTSD.
The University of Michigan researchers designed their tool, called Priori, as a smartphone app that unobtrusively records and analyzes the user's outgoing speech during phone calls (incoming speech is not recorded or analyzed). Because the app is recording the user's personal conversations, patient privacy is a key issue. The recordings are therefore encrypted and off-limits to the researchers, who only have access to the results of the computer analyses.
The Priori system tracked the user's patterns of speech and silence, the pitch of the user's voice, and other acoustic features. To teach the system which features were associated with mania, depression, or the calm of a normal mood, the researchers made a weekly phone call to each user, in which they assessed the user's mood and labeled it. Then the researchers tested their model by having the Priori system analyze the user's other conversations on the day of the assessment and the days immediately before and after, when the user's mood was assumed to be about the same. While the signals of mania or depression were more subtle in those "unstructured" conversations, they were detectable.
This project will run for a few more years, and researchers are seeking volunteers.
Eliza Strickland is a senior editor at IEEE Spectrum, where she covers AI, biomedical engineering, and other topics. She holds a master’s degree in journalism from Columbia University.