A sleep apnea detection method based on discrete wavelet transform and convolutional neural network

Currently, the most authoritative clinical method for diagnosing SA is polysomnography (PSG), which involves monitoring respiratory patterns, oxygen saturation, and electrocardiogram (ECG) during sleep to determine the presence of apnea events. However, this method is complex, inefficient, resource-intensive, and requires interpretation by trained professionals, making it unsuitable for widespread use in large populations. In this study, we propose a new method for detecting SA from ECG signals, this method first employs discrete wavelet transform technology to denoise the collected ECG signals, thereby improving signal quality. Subsequently, a trained neural network is used to automatically extract signal features and classify them. We use the dataset of single-channel overnight sleep ECG data from the Apnea-ECG database, which contains 70 cases, to validate the proposed method. Experimental results indicate that the neural network achieved accuracy, sensitivity, and specificity of 91.5%, 86.2%, and 89.2% in detecting SA, respectively. Compared to existing related research, the model proposed in this study exhibits superior performance.