Real-Time Portable Obstructive Sleep Apnea Monitoring Using Electrical Impedance Tomography With a Dual-Branch Compensation Network
Abstract
Real-time and convenient monitoring of obstructive sleep apnea (OSA) can capture dynamic changes in upper airway obstruction during sleep, providing valuable information for early screening, disease assessment, and long-term management. In this study, electrical impedance tomography (EIT) was employed to assess upper airway collapse for real-time OSA monitoring. To improve the spatial resolution and generalization of the EIT reconstruction, we developed a dual-branch compensation network (DBCNet). By integrating sensitivity compensation with main and auxiliary branches that jointly impose physical constraints, DBCNet accelerates model convergence and improves learning stability when trained with limited samples. A 3 D neck model was constructed using generalized procrustes analysis to simulate various upper airway collapse states. Phantom experiments showed that the error in assessing upper airway collapse was less than 5.2%. Human-subject experiments involving ten participants demonstrated the feasibility of DBCNet for dynamically monitoring apnea-related upper airway changes, while synchronous polysomnography in five participants provided preliminary validation of its event-detection performance. Furthermore, deploying the lightweight DBCNet on mobile devices enabled real-time assessment of apnea-related events, demonstrating its potential for dynamic OSA monitoring in nonclinical settings.