Acquisition and analysis of snore signals for diagnosis of obstructive sleep apnea

Abstract

Obstructive sleep apnea (OSA) is a sleep-related breathing disorder that is common worldwide and potentially life-threatening; however, many affected individuals remain undiagnosed and untreated. This research aims to innovate on a simple, cost-saving, and reliable approach to diagnose OSA via the acquisition and analysis of snore signals, with an intention to mass screen for OSA. This thesis attempts to achieve the research aim through: (1) the implementation of a robust and user-friendly acquisition system for snore signals, along with recommendations for measurement standards; (2) the development of an advanced wavelet-driven preprocessing system that efficiently integrates both snore signal enhancement and snore activity detection; (3) the identification of effective snore-based OSA diagnostic markers, including formant frequencies (82.5-100% sensitivity, 82.0-95.0% specificity), wavelet bicoherence peaks (82.5-100% sensitivity, 83.3-100% specificity), and psychoacoustic metrics (72.0-78.0% sensitivity, 91.2-92.0% specificity), which accurately classify apneic and benign snores in same-and both-gender patient groups (p-value < 0.0001); (4) the formulation of regression models that are indicative of OSA severity; (5) the investigation of physiological-anatomical-acoustical relationships of snores via sourcefilter modeling; and (6) the successful generation of natural-sounding synthetic snores using a novel snore source flow model. Results consistently reveal that snore signals carry rich information for OSA detection; therefore, the use of snore properties to distinguish between patients with and without OSA is promising. Continued exploration in this research area will certainly unfold the clinical value of snore signals in diagnosing OSA in the near future.