Toward Online Removal of Cardiac Interference From Trunk Electromyography by Morphological Modeling of the Electrocardiography

Abstract

Trunk electromyography (EMG) has been widely used in many biomedical applications, which is usually contaminated by electrocardiography (ECG) interference. Several methods have been proposed for ECG removal from trunk EMG. However, most of them are either inaccurate or unsuitable for online applications, e.g., prosthesis control. The aim of the present study is therefore to develop an accurate ECG removal algorithm suitable for online applications. Each ECG wave was modeled by Gaussian kernel functions and subtracted from the trunk measurement to obtain a clean EMG. Two synthetic datasets were generated by mixing a real EMG with a healthy ECG and a dysrhythmia ECG, respectively. Average rectified value (ARV) and mean frequency (MF) were calculated from the reconstructed EMG and the clean EMG for performance evaluation. Moreover, real trunk EMG was recorded under isometric contractions with different forces. Correlation coefficient (CC) between the amplitude of the reconstruct EMG and the contraction force was calculated as performance metric. Small root mean square errors were observed in ARV and MF between the clean EMG and reconstructed EMG, i.e., $2.5\pm 0.7 ~\mu \text{v}$ and 2.0± 0.4 Hz for the synthetic dataset containing healthy ECG and $3.1\pm 1.7 ~\mu \text{v}$ and 3.0± 1.2 Hz for that containing dysrhythmia ECG. High CC (0.91± 0.12) between EMG amplitude and contraction force was observed for real trunk EMG. Our algorithm outperforms many of the state-of-the-art algorithms and is implemented in each cardiac cycle, enabling possible online applications such as prosthesis control.

Keywords

• Trunk electromyography
• cardiac interference removal
• electrocardiography modeling
• Gaussian kernel functions