MTCNet: A Multi-scaled Temporal Convolution Network for Sleep Apnea Detection from Single-lead ECG Signal

Sleep apnea is a common sleep disorder caused by decreased blood oxygen saturation during sleep. Polysomnography is the gold standard for early detection. Because of the complicated operation and high price, it is gradually out of the public view. In recent years, single-lead electrocardiogram (ECG) signal detection sleep apnea methods based on deep learning have significantly progressed. This paper proposes a multi-scale temporal convolutional network (MTCNet) to detect sleep apnea. MTCNet uses multi-scale convolutional networks to capture fine-grained and coarse-grained features. This design enhances the analysis of characteristics associated with sleep apnea. The proposed MTCNet model significantly improved sleep apnea detection compared to previous methods. In our proposed model, the accuracy reached 91.5%, and the average sensitivity and specificity reached 85.8% and 95.0%, respectively.