A Straightforward Methodology to Distinguish Complex Fractionated Atrial Electrograms of Paroxysmal from Persistent Atrial Fibrillation

Many indices aimed at discriminating between paroxysmal and persistent atrial fibrillation (ParAF vs. PerAF) have been previously studied and assessed via statistical tests in order to suggest optimized approaches to catheter ablation (CA) of AF. However, clinicians demand the use of simple classification methods of straightforward comprehension. The present work exploits AF cycle length (AFCL), dominant frequency (DF), sample entropy (SE) and determinism (DET) of recurrent quantification analysis, applied to AF recordings of complex fractionated atrial electrograms (CFAEs), aimed at creating straightforward models to discriminate between ParAF and PerAF. AFCL and DF were calculated on the full AF recordings, whereas SE and DET were computed on segments of 1, 2 and 4s. First, correlation matrix filters removed redundant information and Random Forests made a ranking of variables by relevance. Next, coarse tree classificators were created, combining optimally high-ranked indexes which were tested with leave-oneout cross-validation. After analyzing all the possible combinations of highly ranked features, the best classification performance provided an Accuracy (Acc) of 88.2% to discriminate ParAF from PerAF, while DET provided the highest single Acc of 82.4%. As conclusion, the careful selection of limited sets of indices feeding straightforward classificators are able to discriminate accurately between CFAEs of ParAF and PerAF.