Surface-ECG atrial activity extraction via blind source separation: spectral validation

The isolation of atrial activity (AA) from the surface electrocardiogram (ECG) remains a challenging unsolved problem. Different methods based on the ventricular activity (VA) cancellation have been proposed, validated from the performance obtained over synthesized AF recordings where the AA is known. This performance is measured by comparison (e.g. correlation and quadratic error) of real and estimated AA lead by lead. Other methods estimate one AA as a global contribution of all leads, and so, performance can not be computed in the same way. In this case, the present work explores this problem from a spectral-domain perspective. Several AA extracted from real atrial fibrillation (AF) and atrial flutter (AFL) patients have been analyzed. A spectral analysis reveals the existence of characteristic patterns.