Use of independent component analysis for reducing CPR artefacts in human emergency ECGs

We applied independent component analysis (ICA) to cardiopulmonary resuscitation (CPR)-corrupted human multichannel emergency ECGs with the aim of reconstructing the original ECGs.

Two ICA algorithms (EFICA and JADE) were selected. Data for ICA were acquired by simultaneously recording eight ECG channels during CPR on a porcine model. The algorithms' reconstruction performance was assessed by the Spearman correlation coefficient (SCC) and the shock advice algorithm of an AED. We then compared the performance of EFICA with the established second-channel adaptive matching pursuit method (AF).

ICA was applied to 918 corrupted ECG multichannel signals. The sensitivity of the AED's shock/no-shock decision increased from 93.5% (corrupted signal) to 99.5/99.8% (JADE/EFICA) in the selected independent component; specificity increased from 50.5% to 78.9/83.2% (JADE/EFICA). The SCCs comparing the reconstructed with the original signal (JADE: 0.75±0.15; EFICA: 0.76±0.15, n=918) were significantly higher than for the corrupted signal vs. the original (0.52±0.22). The SCC is significantly higher (p<0.01) using EFICA than AF (EFICA: 0.75±0.16; AF: 0.72±0.19, n=718). For all signals at all SNR levels, specificity did not differ significantly between EFICA (83.6%) and AF (80.2%). EFICA proved to be superior especially at low corruption levels (SNR<-5dB). Sensitivity was above 99.5% for both algorithms.

We have demonstrated that CPR artefacts in the emergency ECG can be reduced using ICA. EFICA and JADE are at least as successful in this regard as are other published algorithms. In particular, non-shockable signals with low SNRs (<-5dB) are reconstructed significantly better (p=0.01) with EFICA than with AF.