Denoising Autoencoder for Eletrocardiogram Signal Enhancement

Eletrocardiogram (ECG) is a useful diagnostic method to detect electrical signals of both healthy or diseased hearts. However, ECG signals are often contaminated by various types of noise, such as baseline wander, electrode contact noise, and motion artifacts. A Denoising Autoencoder (DAE) is selected to enhance ECG signal in this paper. The ECG signals used for DAE training is initially corrupted by means of a stochastic mapping and then reconstructed to the initial uncorrupted signals. DAEs are stacked to build deep architecture, which will improve the expression with multi-level feature extraction. The method is evaluated on ECG signals from the MIT-BIH Arrthythmia Database. MIT-BIH Noise Stress Test Database was used for generating noise. Computer simulation results show that the proposed system performed better than typically used ECG denoising methods with significant improvement on signal to noise ratio (SNR) and root mean square error (RMSE).