Algorithms for Atrial Signal Extraction in Atrial Fibrillation ECGs: A Comparison Based on the Correlation Between Endocardial and Surface Dominant Frequency

The non-invasive analysis of atrial fibrillation (AF) relies on the extraction of atrial activity from surface ECG recordings. The present study compares three different methods for AA extraction from multi-lead ECG recordings: The adaptive singular value QRST cancellation, the spatio-temporal QRST cancellation and Independent Component Analysis (ICA). A criterion for assessing the performance of the extracting techniques on real data is proposed, based on the correlation r between surface and endocardial atrial fibrillation dominant frequency. Performance results obtained with the proposed criterion are compared with those obtained considering the spectral concentration index (SC) of the estimated atrial signal as an estimator of extraction quality. On a database of 20 surface 12-lead ECG and endocardial recordings of persistent AF, results show that higher SC corresponds to better dominant frequency correlation. In addition, the ICA-based method was found to perform better in terms of this two criteria (SC= 68.2% ± 10.4% and r = 0.57 , p < 0.01).