A stacked contractive denoising auto-encoder for ECG signal denoising

Authors:

Abstract
As a primary diagnostic tool for cardiac diseases, electrocardiogram (ECG) signals are often contaminated by various kinds of noise, such as baseline wander, electrode contact noise and motion artifacts. In this paper, we propose a contractive denoising technique to improve the performance of current denoising auto-encoders (DAEs) for ECG signal denoising. Based on the Frobenius norm of the Jacobean matrix for the learned features with respect to the input, we develop a stacked contractive denoising auto-encoder (CDAE) to build a deep neural network (DNN) for noise reduction, which can significantly improve the expression of ECG signals through multi-level feature extraction. The proposed method is evaluated on ECG signals from the bench-marker MIT-BIH Arrhythmia Database, and the noises come from the MIT-BIH noise stress test database. The experimental results show that the new CDAE algorithm performs better than the conventional ECG denoising method, specifically with more than 2.40 dB improvement in the signal-to-noise ratio (SNR) and nearly 0.075 to 0.350 improvements in the root mean square error (RMSE).
Funding Information
  • Natural Science Foundation of Hebei Province (F2015201112)
  • Colleges and Universities in Hebei Province Science and Technology Research Point Project (ZD2015067)
  • Funds for Distinguished Young Scientists of Hebei Province (F2016201186)
  • National Natural Science Foundation of China (61473112)

This publication has 24 references indexed in Scilit: