Sleep Staging Based on Cardiopulmonary Coupling Features Derived from Single-Lead ECG Signal

Sleep staging is an important tool for diagnosing various sleep disorders. However, polysomnography (PSG), which serves as the gold standard for sleep detection, is challenging to acquire in non-clinical settings. Electrocardiogram (ECG) signals are readily obtainable and have been utilized in sleep staging research. The majority of existing studies have not taken into account neighborhood information and cardiopulmonary coupling technology. Actually, medical experts usually consider about 30 minutes or 1 hour of contextual information when staging a segment of ECG, so neighborhood segments can provide auxiliary information. In addition, clinical studies have proved that cardiopulmonary coupling technology plays a key role in sleep staging research. In view of these, we propose a sleep staging method based on ECG, cardiopulmonary coupling technology and deep neural network. More specifically, we first utilize the CNN model for automatic feature extraction and fusion of cardiopulmonary coupling features and the original ECG. Then, noises effects caused by neighborhood information are removed through an efficient channel attention (ECA) mechanism. Finally, sleep stages are determined for signal segments. To verify the robustness of the model, we evaluate it on four publicly available datasets and a private dataset. Experiments show that our method outperforms other benchmarks.