Diagnosis of Obstructive Sleep Apnea using Formant Features of Snore Signals

Abstract

Obstructive sleep apnea (OSA) is a sleep disordered breathing that affects about 4% of adult men and 2% of adult women in the world. It is caused by a collapse of the upper airway during sleep, which could lead to many serious consequences, such as high risk of cardiovascular morbidity and even mortality. The gold standard for OSA diagnosis is the Polysomnography (PSG), which requires patients to stay overnight in a sleep laboratory, and attached to numerous physiologic sensors. Therefore, a full PSG diagnosis can be time consuming, inconvenient, and costly.

Snoring is generated by the vibrating soft tissue in the upper airway, and it is the earliest symptom of OSA. Thus, snore signals may provide an excellent framework for non-invasive diagnosis of OSA. In this paper, the use of formant features of snore signals for OSA diagnosis is presented. The raw snore signals were preprocessed using a modified Normalized Least-Mean-Square (NLMS) adaptive filter for noise cancellation, and subsequently modeled using Linear Predictive Coding (LPC) for spectra analysis. Acoustical changes due to the collapsing upper airways can be reflected on the formant features in the frequency spectrum, which were extracted to discriminate between benign and apneic snores.

Results show that the formant features of snore signals carry useful information for OSA diagnosis, and demonstrate that the use of snore signals can be a convenient, inexpensive, and reliable diagnostic approach for mass screening of OSA.