Data-driven Modeling and Prediction of Obstructive Sleep Apnea based on Physics-guided Pathophysiological Understanding

Obstructive sleep apnea (OSA) is the most common sleep orientated breathing disorder. OSA is characterized by upper airway obstruction, which is highly body position dependent. Specifically, head and head-neck position are critical in determining airway collapse, and adjusting head and head-neck positioning can be useful therapy for OSA. Thus, a real time monitoring and intervention of the head neck position and the corresponding airway collapsibility during sleep would allow an individualized adaptive positioning therapy. The main challenges of developing real-time OSA treatment systems include the unclear pathophysiological understanding of the collapse of pharyngeal walls and thus real-time cyber-physical system (CPS) based OSA therapy is underdeveloped. In this paper, we establish a data-driven method for modelling and predicting the occurrence of OSA in real time based on physics-guided pathophysiological understanding, which provides the basis for real-time CPS-based OSA therapy towards personalized and automated treatment.