CPR Artifact Reduction in the Human ECG Using Independent Component Analysis

Abstract

Currently, automated external defibrillators (AEDs) require interruption of cardiopulmonary resuscitation (CPR) to analyze ECG. To avoid interruptions of chest compressions, which decrease the defibrillation success rate, there are several algorithms available for reducing CPR artifacts from the corrupted ECG, but, based on recent research reports, none of the algorithms checked showed superior sensitivity and specificity.

Independent component analysis (ICA) is an established tool for signal extraction. However, as far as the authors know, artifact removal from the CPR corrupted ECG has not been tried with ICA. As a new approach to solve the problem of removing CPR induced noise, the use of ICA is evaluated in this paper. By measuring four ECG channels during CPR on a porcine model, data were obtained for testing ICA algorithms. After applying ICA to corrupted signals with small ECG amplitude (low SNR), the sensitivity increased from 75% (corrupted signal) to 100% using the selected independent component and specificity from 80% to 89%, taking the AEDs decision whether the rhythm is shockable or not. When checking the similarity between the original, the corrupted and the reconstructed signal, the computed correlation values indicated an improvement compared to the corrupted signal. We conclude that ICA was successful in separating the artifacts from the corrupted ECG in our experimental setup. It must be noted that the correct independent component was selected by visual inspection only.