Comparison of Two Algorithms for ECG Signal Denoising: A Recurrent Neural Network and A Support Vector Regression

Sahar Keshavarzi (Isfahan University of Technology, Iran)

The Electrocardiogram (ECG) signal is usually degraded by noise. The use of a denoising technique, therefore, is required before applying any analyses (for example, for diagnoses purpose) to this signal. This paper investigated two algorithms for enhancing the ECG signal in the presence of noise: a Support Vector Regression (SVR) and a deep Recurrent Neural Network (RNN). We compared the performance of the two algorithms using three case studies that differed in terms of the size of data used to train the algorithms. The performance of both algorithms was evaluated at different SNRs (-5 dB to 15 dB, with the step size of 2 dB) using two objective metrics including Pearson correlation and mean squared error. The simulation results demonstrated that the RNN outperformed the SVR algorithm for ECG enhancement in noise for all the three case studies. In particular, the mean correlation scores at low SNRs (-5 dB to 1dB) averaged across the three studies were 0.7 and 0.85 for SVR and RNN, respectively. We also found that the performance of both algorithms was degraded by decreasing in SNR. In addition, improvement in the performance of both algorithms was observed when the size of training data increased.

Journal: International Journal of Simulation- Systems, Science and Technology- IJSSST V23

Published: Mar 31, 2022

DOI: 10.5013/IJSSST.a.23.01.02