ECG Signal-Based Characterization and Classification of Sleep Apnea

This study proposes a cost-effective alternative using electrocardiogram (ECG) signals to characterize and classify obstructive sleep apnea (OSA). Sleep apnea is a prevalent disorder traditionally diagnosed using the EEG signals of polysomnography. ECG signals can be obtained easily without any complex procedures, unlike expensive and cumbersome polysomnography (PSG). The ECG signals were preprocessed using Undecimated Wavelet Transform (UDWT), with Coiflet5 proving particularly effective among the rest of the wavelets for noise reduction. R-peaks were detected using Undecimated Wavelet Transform (UDWT) combined with time window segmentation, and heart rate variability (HRV) features were derived. Time domain, frequency domain, and non-linear measures were extracted from the denoised ECG signals. The classification was attempted using several machine learning models, such as logistic regression, support vector machines, k-nearest neighbors, decision trees, and random forest to detect sleep apnea. Out of all the classifiers, the XGBoost classifier demonstrated superior performance, achieving an accuracy of 85.6%, highlighting the proposed approach's efficacy in accurately classifying sleep apnea using ECG signals. All the work was carried out using the PhysioNet Apnea-ECG database. This research provided a viable and efficient alternative to traditional PSG-based diagnosis.