Comparação dos filtros ICA, passa-alta e Wavelet, aplicados em sinal EMG contaminado por sinal ECG

Abstract

Introduction and bibliographical review: Surface electromyography represents the summation of action potentials that travels along muscles membranes, detected by a surface electrode. Due to it being a method of biomedical signals acquisition, it’s subjected to several interferences, such as cable movement, power line noise, motor activity of muscles that differ from the detection point, problems on the skin-electrode interface and noise from the electrocardiographic signal – in this case, on acquisitions near the thoracic region, closer to the heart. The last one presents the biggest removal difficulties because it emits a signal of a complex shape and which frequency spectrum is in the same range as the electromyogram. In order to remove this complex noise, several approaches were suggested in literature, such as adaptative sampling, subtraction by a template, adaptative filtering, independent component analysis (ICA), wavelets or high-pass filter; the focus of this paper is the last 3 methodologies, using the R language on RStudio software. Methodology: Synthetic electromyographic and electrocardiographic signals were created using libraries present on this software in order to use the filtering methods. The output signals of each procedure were then compared with the produced synthetic electromyographic signal to check how good the filtration was. The comparison parameters used were the root medium square error, spacial correlation, spectral similarity, median frequency and mean frequency. The EMG signal was then contaminated with the electrocardiographic signal to simulate a surface electromyography with ECG noise. The ICA was used in two ways: one using both the contaminated electromyographic signal and the electrocardiographic noise as input, the other was using two distinct synthetic electromyographic signals as input. Results: Among the presented methods, stood out the ICA with the electrocardiographic noise as reference, because it removed the noise almost completely. Among the methods that didn’t use the noise as reference, stood out the method using wavelets, specially Daubechies 10 with the “op1” approach, presenting values closest to the ones of the synthetic noise-free electromyogram. In contrast, the methodology that showed the worst results was using two distinct synthetic electromyographic signals, both with the ECG noise, as input for the ICA, because it didn’t make a good separation of the noise and the signal, presenting parameters that differ the most from the expected compared to the other methods.