Obstructive Sleep Apnea Severity Multiclass Classification Using Analysis of Snoring Sounds

Abstract

The current gold standard for diagnosing obstructive sleep apnea (OSA) is an overnight multi-channel polysomnography (PSG), an expensive, labour-intensive, and uncomfortable procedure. Accordingly, it would be beneficial to have a screening method to promptly determine the severity of a patient, prior to a referral for PSG. This paper intends to distinguish the severity of OSA patients. We show that an accurate multiclass classification of snoring subjects with four classes of OSA, can be achieved on the sound spectrum of snoring without any information requirement on the number of apneas. 33 Snoring sounds with different degrees of obstructive sleep apnea and non-OSA were analyzed. The proposed technique uses K-Means clustering to cluster the sound spectrum and reconstruct features. Support vector machine (SVM) has been used for the classification. The multiclass snore sounds classification approves early stratification of subjects according to their severity. A classification accuracy of 75.76% was reported using the proposed method. The experimental results also demonstrate that the proposed method can provide diagnostic suggestions for OSA screening.