Denoising of Electriocardiogram (ECG) Signal

Abstract

Plenty of information related to physiology of the heart can be collected from the electrocardiogram (ECG) signal. In reality, the ECG signal is interfered by the various noise sources. To extract the correct information related to physiology of the heart, the cancellation of noise present in the ECG signal is needed. In this thesis, the investigation on effectiveness of the empirical mode decomposition (EMD) with non - local mean (NLM) technique by using the value of differential standard deviation for denoising of ECG signal is performed. Differential standard deviation is calculated for collecting information related to the input noise so that appropriate formation in EMD and NLM framework can be performed. EMD framework in the proposed methodology is used for reduction of the noise from the ECG signal. The output of the EMD passes through NLM framework to preserve the edges of the ECG signal and cancel the noise present in the ECG signal after the EMD process. The performance of the proposed methodology based on EMD with NLM framework has been validated by using added white and color Gaussian noise to the clean ECG signal from MIT-BIH arrhythmia database at different signal to noise ratio (SNR). The proposed denoising technique shows lesser percent root mean square difference (PRD), mean square error (MSE), and better SNR improvement compared to other well-known methods.
In this thesis comparison of the different technique for removal of muscle artifacts and baseline wander noise. Different methodology for removal of muscle artifacts are conventional filtering, wavelet denoising, and non-local mean (NLM) technique, in these wavelet denoising gives better SNR improvement and lesser MSE and PRD. Similarly for baseline wander removal, the performance of different techniques like two-stage median filter, single-stage median filter, two-stage moving average filter, single-stage moving average filter, low-pass filter, and band-pass filter have been evaluated using added baseline wander noise to synthetic ECG signal at different sampling frequency, among these two-stage median filter gives better SNR improvement and lesser MSE and PRD.