SLNet: A Hybrid Machine and Deep Learning Model for Sleep Apnea Episodes Detection From Single-Lead ECG Data

In this study, a comprehensive feature extraction and classification scheme for sleep apnea episodes detection using single-lead electrocardiogram (ECG) data, namely Single-Lead Network (SLNet), is proposed. SLNet involves ECG R-peak detection and heart rate variability estimation to extract time/frequency domain, and Poincaré plot features. Additionally, wavelet-based multi-resolution analysis (MRA) is employed to decompose the ECG signal and extract statistical and higher-order-crossings features from each MRA detail scale. Ultimately, to make predictions, SLNet utilizes hybrid machine/deep learning models. The performance of SLNet was evaluated on 70 single-lead ECG recordings from 32 SA patients drawn from the open access "MIT Physionet Apnea-ECG Database''.