RI: Small: Speech-Centered Robust and Generalizable Measurements of "In the Wild" Behavior for Mental Health Symptom Severity Tracking

点击翻译按钮获取中文摘要

Bipolar disorder is a common and chronic illness characterized by pathological swings from euthymia (healthy) to mania (heightened energy) and depression (lowered energy). Mood transitions are associated with profound consequences to one's personal, social, vocational, and financial well-being. Current management is clinic-based and dependent on provider-patient interactions. Yet, increased demand for services has surpassed capacity, calling for radical changes in the delivery of care. This project will create new algorithms that can process speech data naturally collected from smartphone use to measure behavior and changes in behaviors and to associate these measurements with the severity of the symptoms of bipolar disorder. This will lead to the creation of new early warning signs, indications that clinical intervention is needed. Natural behavior provides a wealth of information about the health an individual. However, when assessing health, clinicians typically access cross-sectional medical data at point-of-care that is based on traditional medical methods (exams, labs, and surveys). Next generation 'precision health' depends on an inclusive and holistic approach that captures changes in health as people live their lives. This is highly relevant as 130 million Americans live with chronic disease and need efficient monitoring strategies. Speech is a promising medium for monitoring mood. Clinicians subjectively assess both form and content of speech when evaluating human disease, as speech is altered by changes in mood and health states. Yet, while speech is easy to record, speech-centered mobile monitoring solutions are not currently publicly available. The technology is neither sufficiently accurate nor robust. The central challenge is the signal itself: speech is inherently variable and complex. Existing techniques are insufficient to handle this complexity, limiting the accuracy and robustness of speech-centered mood monitoring technologies. This project will create novel and robust approaches to extracting mood symptom severity measures from speech. Mood is clinically quantified via the Hamilton Depression Rating Scale (HamD) and the Young Mania Rating Scale (YMRS). The technology focuses on the creation of methods that accurately extract symptom-focused measures, whose variation lies between that of speech and mood severity, and that are robust to conditions, both environmental and social, in which the data were recorded. The methods will be validated on an existing natural speech dataset at the University of Michigan. The unification will provide critical steps towards speech-centered mHealth solutions.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.