Automatic screening of Obstructive Sleep Apnea from the ECG based on Empirical Mode Decomposition and wavelet analysis

Abstract

This study proposes three different methods to evaluate Obstructive Sleep Apnea (OSA) during sleep time solely based on the ECG signal. OSA is a common sleep disorder produced by repetitive occlusions of the upper airways, which produces a characteristic pattern on the ECG. Extraction of ECG characteristics as the heart rate variability and the QRS peak area offer alternative measures for cheap, non-invasive and reliable pre-diagnosis of sleep apnea. 50 of the 70 recordings from the database of the Computers in Cardiology Challenge 2000, freely available on Physionet, have been used in this analysis, subdivided in a training and a testing set. We investigated the possibilities concerning the use of the recently proposed method Empirical Mode Decomposition in this application and compared it with the established Wavelet Analysis. From the results of these decompositions the eventual features were extracted, complemented with a series of standard HRV time domain measures and three extra non-linear measures. Of all features smoothed versions were calculated. From the obtained feature set, the best performing feature subset was used as the input of a Linear Discriminant Classifier. In this way we were able to classify the signal on a minute-by-minute basis as apneic or non-apneic with an accuracy of around 90% and to perfectly separate between apneic and normal patients, using around 20 to 40 features and with the possibility to do this in three alternative ways.