Scientists at the National Center for Supercomputing Applications and the University of Illinois College of Medicine Peoria (UICOMP) were authors of a research paper published in the Journal of Acoustical Society of America Express Letters that demonstrates improved, automated screening methods for anxiety and major depressive disorders.
In the project titled, “Automated acoustic voice screening techniques for comorbid depression and anxiety disorders,” Mary Pietrowicz, along with colleagues from the University of Illinois Urbana-Champaign and UICOMP, explored how machine learning could effectively distinguish individuals with comorbid depression and anxiety disorders from healthy controls using acoustic and phonemic analysis of semantic verbal fluency data.
More and more people are being diagnosed with these disorders, yet many suffering remain undiagnosed due to known perceptual, attitudinal and structural barriers. Anxiety affects 19.1% of adults in the United States and major depression 8.3% while being the leading cause of disability in individuals under 40 years old. Despite this high prevalence, treatment rates are low and, if left untreated, can lead to decreased productivity, poor functioning in society, erosion of cognitive abilities, strained relationships and suicide.
New methods, tools and technologies – such as automated acoustic voice analysis – are needed to overcome these barriers and improve screening rates.
This research demonstrates that analysis of short samples of acoustic voice, specifically one-minute verbal fluency tests, can be used in screening for anxiety and depression disorders, and can function online, at any time, addressing many of the barriers to screening and treatment. In addition, our AI models provide explainability, and therefore insight, into the impact that depression and anxiety have on speech and language. This work enables the development of clinical screening and tracking systems at scale.
Mary Pietrowicz, NCSA Senior Research Scientist
Researchers tested a custom dataset curated specifically for this study that included both healthy people and people with comorbid depression and anxiety across the spectrum of severity. People with other comorbid conditions known to affect speech and language were excluded from the study. Acoustic models using only data from one-minute verbal fluency tests discerned the presence of comorbid disorders at a highly successful rate.
“The data for this study were collected by multiple medical students at the University of Illinois College of Medicine Peoria,” said UICOMP Director of Research Services Sarah Donohue. “These students interviewed each of the participants, recorded the interviews and did an animal naming task at the end of the interviews with the participants.”
A primary benefit of these acoustic tests is that they’re accessible. They may be administered either online, in-app or in-clinic, which directly addresses known barriers to screening, including factors such as stigma, low self-perception of need, costs, transportation issues and limited access to healthcare.
“The development of an efficient, accurate and easy-to-use method for screening patients who may be suffering from depression or anxiety offers tremendous promise,” said UICOMP Chair and Professor of Clinical Psychiatry Ryan Finkenbine. “The application of advanced machine learning models to the clinical setting provides a remarkable path for clinicians to screen for signs of mental illness in an adaptive and practical way. Patients and clinicians alike will benefit from improved methods for comprehensive medical and mental health care.”
