From: Analysis and applications of respiratory surface EMG: report of a round table meeting
 | Specifics | Common pitfalls | Best practice |
---|---|---|---|
ECG removal | |||
High-pass filtering | Specific for first ‘raw’ data checks and preliminary step for some preprocessing methods (gating) The higher the ratio between respiratory and cardiac signal power, the lower the required cutoff frequency to reduce impact of cardiac activity on the filtered signal (such as in small-distance electrode setups and parasternal EMG when muscle activation is strong | Distortion of the spectrum Reduction in EMG amplitude Absence of respiratory EMG amplitude in the filtered signal when there is cardiac interference and/or very weak respiratory muscle activation | If used as the sole method, a high cutoff frequency should be employed and adjusted to minimize impact of cardiac activity Mean Absolute Value (MAV) is recommended to obtain the respiratory waveform in cases where QRS peaks are still present in the resulting signal Lower cutoff frequencies (< 50Hz) should be combined with other preprocessing techniques to fully remove the QRS complex |
Gating | Envelope calculation when EMG amplitude is to be maintained Requires robust detection of R-peaks | Cannot be used with tachycardia Substantial loss of temporal information Not suitable for detecting respiratory onset/offset with high precision | Pan-Tompkins algorithm should be used to detect R-peaks Combination with 20 Hz high-pass filter to remove P and T waves Window length should be adjusted to the duration of the QRS complex Appropriate gate-filling techniques must be used (interpolation or median) |
Wavelet | Go-to method for ECG removal in far-distance electrode setups (with strong ECG interference) when resp muscle activation is small Best method when R-peaks cannot be robustly detected (e.g., many ectopic beats, patients with arrhythmias) | Inadequate setting of Fs, level of decomposition, thresholds Thresholding might cutoff large EMG activity bursts | Pre-filtering is not required Number of decomposition levels depends on sampling frequency and should be adjusted to the P-/T- waves and motion artifacts (10–20 Hz): 5 levels for fs of 1000 Hz, increase/decrease level when fs doubles/halves Resulting wavelet-bands and thresholds should be checked visually Daubechies 2 and 4 wavelets have demonstrated good performance in denoising respiratory EMG [29, 35, 36] Fixed threshold; start with a threshold set at 4.5 times the standard deviation of the decomposition level (σk) |
Envelope | |||
General recommendation: Use centered window with length 250 ms, deviate when application demands | |||
Root Mean Square (RMS) | Most generally used Power of the signal can be used based on RMS (and compared with that obtained by spectral methods) | N/A | Step size of the moving window should be considered (1 sample step is feasible) |
Average Rectified Value (ARV) | Less affected by high amplitude peaks (like remaining QRS artifacts) than RMS | ||
Mean Absolute Value (MAV) | Â | Â | Combination with HPF |
Fixed sample entropy (fSampEn) | More robust than RMS and ARV, i.e., less affected by high amplitude peaks caused by remaining artifacts | Step size of the moving window: 1 sample step can be computationally expensive (for fSampEn) | Application directly to raw data, no other filtering needed Embedded dimension (m = 1) Tolerance value (r) set to 0.2–0.3 times the standard deviation of the sEMG signal |