Investigation of obstructive sleep apnea using nonlinear mode interactions in nonstationary snore signals.

Acoustic studies on snoring sounds have recently drawn attention as a potential alternative to polysomnography in the diagnosis of obstructive sleep apnea (OSA). This paper investigates the feasibility of using nonlinear coupling between frequency modes in snore signals via wavelet bicoherence (WBC) analysis for screening of OSA. Two novel markers (PF1 and PSF), which are frequency modes with high nonlinear coupling strength in their respective WBC spectrum, are proposed to differentiate between apneic and benign snores in same- or both-gender snorers. Snoring sounds were recorded from 40 subjects (30 apneic and 10 benign) by a hanging microphone, and subsequently preprocessed within a wavelet transform domain. Forty inspiratory snores (30 as training and 10 as test data) from each subject were examined. Results demonstrate that nonlinear mode interactions in apneic snores are less self-coupled and usually occupy higher and wider frequency ranges than that of benign snores. PF1 and PSF are indicative of apneic and benign snores (p < 0.0001), with optimal thresholds of PF1 = 285 Hz and PSF = 492 Hz (for both genders combined), as well as sensitivity and specificity values between 85.0 and 90.7%, respectively, outperforming the conventional diagnostic indicator (spectral peak frequency, PF = 243-275 Hz, sensitivity = 77.7-79.7%, specificity = 72.0-78.0%, p < 0.0001). Relationships between apnea-hypopnea index and the proposed markers could likely take the functional form of exponential or power. Perspectives on nonlinear dynamics analysis of snore signals are promising for further research and development of a reliable and inexpensive diagnostic tool for OSA.