Research Insights

Obstructive sleep apnea (OSA) is a major health problem and can lead to or increase the risk of cardiovascular disease, stroke, metabolic disease, daytime sleepiness, workplace errors, traffic accidents and death, if it is left undetected. Worldwide it is estimated one billion people, one in seven adults, have OSA. As sleep occurs primarily in the bedroom, monitoring sleep quality at home, instead of in sleep labs, would significantly advance the self-management, and potentially the clinical management, of OSA and other sleep disorders. Thereafter, an approach that noninvasively monitors sleep quality at home would have significant societal and health benefits. This project brings together leading researchers from informatics and health disciplines to create a contactless sensor system for OSA monitoring and characterization, which integrates advanced Artificial Intelligence (Al) and Data Science (OS) into smart sensors and home care. The key research challenges are to convert the information-rich sensor signals to clinically meaningful vital signs and behavioral patterns that are linked with OSA. This project addresses three primary themes of the SCH solicitation: Automating Health, Transformative Data Science, and Effective Usability. The proposed research makes fundamental contributions to computer, data and biomedical science and engineering and will create the first contactless Internet of Things (loT) system for real-time and engagement-free sleep apnea monitoring and characterization. The main technological innovation is a set of novel stream data Al/OS for sleep events and vitals monitoring: a robust signal quality control and segmentation process based on a moving-sum statistic and recursive binary segmentation; a novel factor auto-regressive recurrent neural network framework to characterize key sleep events; a new approach of monitoring vital signs and their variations based on an innovative panel data model and the structural changes in regression coefficients; and a flexible and distributional robust feature assessment method to enable out-of-distribution (000) generalization. The proposed interdisciplinary research takes a coordinated approach that balances theory with evidence-based analysis and systematic advances. The project will conduct empirical validation of new concepts through research prototypes, ranging from specific components to entire systems, and lead to new fundamental insights and effective usability.

Funder: National Institutes of Health

Amount: $1,171,628

PI: Wenzhan Song, College of Engineering