Identification of Sleep Apnea Syndrome by Analyzing Sleep Sound Data Using a Clustering Method

Abstract

Sleep Apnea Syndrome (SAS) has attracted considerable attention in recent years because it is known to cause catastrophes when people suffering from it fall asleep while driving. In addition, approximately 70% of SAS patients have other lifestyle-related diseases, which are further complicated owing to SAS; therefore, its early detection is important. Moreover, the detection of pre-SAS, which is likely to turn into SAS in the future, may help in the early detection of SAS. In this study, a method to identify SAS and pre-SAS using only sound data has been proposed. In the proposed method, peculiar breathing sound patterns in SAS are analyzed using the k-means clustering method and perform discriminant analysis using quantification theory based on the results. Because respiratory sound data is extensive and non-uniform during sleep, an efficient analysis method is required. The presented analysis method focuses on the features of SAS sounds, and performs the analysis almost automatically; therefore, the time required for analysis is reduced and the uniformity of the analysis results is guaranteed. Using this method, it is possible to distinguish between non-SAS and SAS symptoms including pre-SAS symptoms, which is difficult to perform using the Apnea-Hypopnea Index, which is a measure of SAS that employs breathing sound data only. Thus, this method, which also analyzes pre-SAS, is convenient for the early detection and treatment of SAS and can also be expected to contribute toward the early detection of preliminary lifestyle-related diseases.