An ECG denoising method based on adversarial denoising convolutional neural network

The Electrocardiogram (ECG) is widely used to diagnose heart disease. However, ECG recordings are often polluted by different noises in real situations. It is of great significance to propose an effective method for noise rejection of ECG signal. With the development of deep learning, neural networks have been widely utilized in signal processing. Most of the existing models of deep neural networks have millions of parameters, which places higher requirements on medical computing power. Based on this, this paper proposes a new denoising framework named the Adversarial Denoising Convolutional Neural Network (ADnCNN). We adopt the ADnCNN model to learn the residual signals in the noisy signal to obtain clean signal. A discriminator network is used for two-class classification of the denoised and clean signals, which are fed back to the ADnCNN model for parameter adjustment. The proposed method is validated by dividing three different types of datasets on the MIT-BIH arrhythmia database, the MIT-BIH noise stress test database, and the QT database for experiments. The denoising performance of the ADnCNN model is represented by the Signal-to-Noise Ratio (SNR), Root Mean Square Error (RMSE) and Percentage Root Mean Square Difference (PRD). The generality and adaptability of the model are also demonstrated. Simulation results show that the proposed method has advantages compared to recently proposed models.