Analysis and denoising of ECG signals

The Electrocardiogram (ECG) is a bioelectric signal generated from the contraction of the human heart muscle, which is recorded from the body surface using electrodes for the diagnosis purpose. During the acquisition or transmission, noises generated from the surrounded electrical equipment, the motion of the patient, the movement of the electrodes, or the contraction of the muscle around the heart usually interfere with the clinically recorded ECG signals. The interference of the noises with the recorded ECG signal in the spatial, temporal, and frequency domain mask the desired signal and reduce the amount of information which can be extracted from it, mislead the clinician, and obstruct the diagnosis process. Many filtering techniques have been proposed for reducing, extracting or separating the noise sources but finding a robust and reliable filtering technique, which completely remove the noise without distorting the morphological shape of the signal still form a challenge. Recently, Blind Source Separation (BSS) and model-based filtering techniques have shown good results in the area of biomedical signal processing. Here in this study, Robust Independent Component Analysis (RobustICA) has been used to separate the measured ECG signal mixture into independent components based on its statistical features then a clean signal has been estimated. On the other hand, model-based filtering has been done using the nonlinear dynamical model of ECG signal with the Extended Kalman Filter (EKF) to estimate a noise-free signal. The proposed algorithms have been used to remove muscle contraction artifact, baseline shift, and electrode motion artifact, which are the most common ECG noises. The efficiency of the proposed methods has been measured by using several noisy ECG signals generated by artificially adding baseline wander, muscle artifact, and electrode movement noises to normal ECG signal. This generates a database with a range of signal to noise ratio (SNR) from 20 to -20 Decibel to inspect the filtered ECG recordings by studying their SNR and morphology after the filtering process. Performance comparison of ICA and EKF are showed that ICA produces better results in reducing muscle artifact when compared with baseline wander and electrode movement artifacts reduction ratio, while EKF shows good results in the reduction of muscle artifact only when the SNR is low. Analysis results demonstrate that ICA is better than EKF in the reduction of baseline wander and electrode movement artifacts while both show good result in the reduction of muscle artifact.