Automated Recognition of Obstructive Sleep Apnea Using Ensemble Support Vector Machine Classifier

Abstract

ECG is mainly used to diagnosis the obstructive sleep apnea (OSA) with a high degree of accuracy in clinical care applications. We have developed a real-time algorithm for the detection of sleep apnea disease based on electrocardiograph (ECG). In this study, features from ECG signals were extracted from 12 normal and 58 OSA patients from physionet apnea ECG database. The baseline noise, motion drift and muscle noise in raw ECG signals are removed using median filter and Daubechies wavelet filter. QRS detection algorithm extracts R-wave amplitude and R-wave time duration from de-noised signal. The proposed QRS detection algorithm contains four stages. The stages are calculation of QRS-complex slope, squaring function, moving-window integration and calculation of R-peak and QRS detection. Time domain features are calculated from the heart rate variability and ECG-derived respiration (EDR). Support vector machine (SVM) and ensemble support vector machine techniques are used for the detection of OSA.