Removing CPR artifacts from the ventricular fibrillation ECG by enhanced adaptive regression on lagged reference signals

Removing cardiopulmonary resuscitation (CPR) related artifacts from human ventricular fibrillation (VF) ECG signals would provide the possibility to continuously detect rhythm changes and estimate the probability of defibrillation success. This would avoid "hands-off" analysis times which diminish the cardiac perfusion and thus deteriorate the chance for a successful defibrillation attempt. Our approach consists in estimating the CPR-part of a corrupted signal by an adaptive regression on lagged copies of a reference signal which correlate with the CPR artifact signal. The algorithm is based on a state-space model and the corresponding Kalman recursions. The preliminary evaluation based on a small pool of artifact-free VF and asystole CPR data outperform comparable previous studies. In comparison with ordinary least-squares (OLS) regression the proposed algorithm shows improvements for low SNR corrupted signals and yields better estimates of the mean frequency of the true VF ECG signal.