Automatic Detection of High-Quality Fibrillatory Waves Segments from Atrial Fibrillation Electrocardiographic Recordings

The conception of techniques for extracting fibrillatory waves (f-waves) from surface electrocardiograms (ECGs) has enabled their analysis to detect atrial fibrillation (AF) and enhance arrhythmia treatment. Nevertheless, the performance of these methods is weakened by the presence of artifacts and variable morphology of the QRST complexes. Hence, this study introduces a two-step approach aimed at refining the identification of f-waves segments free from artifacts and ventricular residues. The first step detects artifacts by setting a threshold on the amplitude of ECG-derived f-waves signals. The second step performs a binary classification of signals (i.e., with or without ventricular residues) by quantifying the residual ventricular activity in QRST-canceled signals through different metrics. The method was optimized on a database of ECG signals recorded in 148 patients with persistent AF and its potential was showcased in relation to the prediction of catheter ablation (CA) outcome. The RuVR metric, calculated as the product of the root mean square value of the f-waves in the QRST interval by their maximum value, emerged as the best single parameter for detecting segments with ventricular residues (with 71.66% accuracy). The combination of RuVR with other metrics through an ensemble model of decision trees, improved the accuracy of the classifier to nearly 82%. The application of the selection methodology to the prediction of CA outcome by dominant frequency in persistent AF enhanced discriminative capability. These results highlight the importance of ensuring the quality of the extracted f-waves segments from surface ECGs before applying further analysis steps.