Adaptive Filtering Methods for ECG Waveform Restoration during Cardiopulmonary Resuscitation

An artifact-free ECG is essential during cardiopulmonary resuscitation $(CPR)$ to decide adequate therapy. Chest compressions $(CCs)$ cause artifacts that alter the ECG waveform. This study analyzes the effectiveness of CPR artifact suppression filters to restore the ECG. For that purpose, artificial mixtures of artifact-free ECGs and CPR artifacts mixed at different signal-to-noise ratios $(SNR_{in})$ were used. Several configurations of three state-of-the-art filters were compared: Least Mean Squares $(LMS)$, Recursive Least Squares $(RLS)$ and Kalman. Performance was evaluated by comparing the artifact-free ECGs and the filtered ECGs in terms of the restored SNR $(SNR_{res})$ and the Pearson's Correlation coefficient $(PCC)$. RLS was the best option with a mean $SNR_{res}$ of 3.3 dB at the optimal working point, slightly above the LMS algorithm in terms of $SNR_{res}$. The similarity of the restored and the clean ECG increased with $SNR_{in}$, obtaining PCC values above 0.6 for $SNR_{in} > -5$ dB. In conclusion, suitable filtering methods to restore ECG waveforms during CPR were proposed, that would enhance the reliability of the ECG analysis during CCs.