Bioelectric model of atrial fibrillation: applicability of blind source separation techniques for atrial activity estimation in atrial fibrillation episodes

In this contribution, we present the theoretical justification that give support to the suitability of blind signal separation (BSS) techniques for the estimation of the atrial activity (AA) present in ECGs of persistent atrial fibrillation (AF). The application of BSS methods to this problem needs the fulfillment of several conditions regarding AA, ventricular activity (VA) and the fashion in which both activities arise on the body surface, that will be justified along the paper. To empirically validate the model, an ICA method is applied to 10 real 12-lead recordings of AF. The identification of AA is put forward based on kurtosis and spectral analysis. The kurtosis value of the estimated AA was always below zero (k_AA = -0.36 ± 0.14), while for the VA was above 20 (k_VA = 30.44 ± 7.83). As conclusion, the verification of the AF bioelectric model makes feasible the application of BSS, and this contribution has indeed justified, for the first time, the theoretical background that supports the applicability of these techniques and empirically its usefulness to solve the AA extraction problem.