SAIL, which has long applied AI and machine learning to identify and classify video, audio and physiological data, partnered with researchers at UCLA to analyze voice data from patients being treated for serious mental illnesses, including bipolar disorder, schizophrenia and major depressive disorders. These individuals and their treating clinicians used the MyCoachConnect interactive voice and mobile tool, created and hosted on the Chorus platform at UCLA, to provide voice diaries related to their mental health states. SAIL then collaborated with UCLA researchers to apply AI to listen to hundreds of voicemails using custom software to detect changes in patients’ clinical states. The SAIL AI was able to match clinicians’ ratings of their patients.
“Machine learning allowed us to illuminate the various clinically-meaningful dimensions of language use and vocal patterns of the patients over time and personalized at each individual level,” said senior author Dr. Shri Narayanan, Niki and Max Nikias Chair in Engineering and Director of SAIL at the USC Viterbi School of Engineering.
Tracking changes in clinical states is important, say the researchers, to detect if there is a change that shows that condition has improved or worsened that would warrant the need for changing treatment.
“Listening to people has always been at the core of psychiatry. Our approach builds on that fundamental technique to hear what people are saying using modern AI. We hope this will help us better understand how our patients are doing and transform mental health care to be more personalized and proactive to what an individual needs,” said lead author of the study Dr. Armen Arevian, Director of the Jane and Terry Semel Institute Innovation Lab which focuses on the interdisciplinary intersection of mental health, neuroscience and technology.
This project builds on SAIL’s body of work in behavioral machine intelligence to analyze psychotherapy sessions to detect empathy of addiction counselors-in- training in order to improve their chances of better outcomes, in addition to the Lab’s work analyzing language for cognitive diagnoses and legal processes.
The next step in this joint research is to scale up this individualized approach to a larger population over longer time periods of observation.