Obstructive sleep apnea detection using discrete wavelet transform-based statistical features

Abstract Motivation and objective Obstructive sleep apnea (OSA) is a disorder identified in nearly 10% of middle-aged people, which deteriorates the normal functioning human organs, notably that heart. Furthermore, untreated OSA associated with increased hypertension, diabetes, stroke, cardiovascular diseases, thereby increasing mortality risk. Therefore, early identification significant interest. Method In this paper, an automated approach for diagnosis using single-lead electrocardiogram (ECG) has been reported. Three sets features, namely moments power spectrum density (PSD), waveform complexity measures, higher-order moments, are extracted from 1-min segmented ECG subbands obtained discrete wavelet transform (DWT). Later, correlation-based feature selection particle swarm optimization (PSO) search strategy employed getting optimum vector. This process retained 18 features initially computed 32 features. Finally, acquired set fed to different classifiers including, linear discriminant analysis, nearest neighbors, support vector machine, random forest perform per segment classification. Results Experiments on publicly available physionet dataset show proposed classifier effectively discriminates signals. Specifically, our method achieved accuracy 89% 90%, 50-50 hold-out validation 10-fold cross-validation, respectively. Besides, both these scenarios, 96% area under ROC. Importantly, provided better performance results than most existing methodologies.