Accuracy of non-invasive frequency estimation during atrial fibrillation

Ablation procedures have become one of the most efficient treatments for termination of atrial fibrillation (AF). The aim of this study is to evaluate the accuracy of dominant frequency (DF) maps on the epicardium computed from non-invasive recordings as a clinical tool for the identification of AF sources. Spherical and realistic atrial models were used. Four fibrillation patterns with varying dominant frequency distributions were obtained. Surface potentials were computed by solving the forward problem and adding noise at signal-to-noise ratios (SNR) from 10 to 20 dB. For the spherical model, with 80% of the surface with a DF of 14 Hz and 20% with a DF of 21.5 Hz, RDM* between generated and inverse computed potentials was 0.56 without added noise and 1.05 with SNR=10 dB. However, for the same conditions, the RDM* of DF maps were 0.02 and 0.09, with DF errors ofO.0l+0.31 Hz and 0.26+1.39 Hz, respectively. For the realistic model, frequency reconstruction was consistent with generated electrograms, allowing an accurate estimation of the DF distribution with a maximum RDM* of 0.19 (SNR=10dB). Inverse computed DF maps reconstructed during insilico AF were more accurate than voltage maps. Noninvasive estimation of DF maps during AF is feasible and may help in procedure planning.