Background: Superimposition of farfield (FF) and nearfield (NF) bipolar voltage electrograms (BVE) complicates the confirmation of pulmonary vein (PV) isolation after catheter ablation of atrial fibrillation. Our aim was to develop an automatic algorithm based on a single-beat analysis to discriminate PV NF from atrial FF BVE from a circular mapping catheter during the cryoballoon PV isolation.
Methods: During freezing cycles in cryoablation PVI, local NF and distant FF signals were recorded, identified and labelled. BVEs were classified using four different machine learning algorithms based on four frequency domain (high-frequency power (PHF), low-frequency power (PLF), relative high power band, PHF ratio of neighbouring electrodes) and two time domain features (amplitude (Vmax), slew rate). The algorithm-based classification was compared to the true identification gained during the PVI and to a classification by cardiac electrophysiologists.
Results: We included 335 BVEs from 57 consecutive patients. Using a single feature, PHF with a cut-off at 150 Hz showed the best overall accuracy for classification (79.4%). By combining PHF with Vmax, overall accuracy was improved to 82.7% with a specificity of 89% and a sensitivity of 77%. The overall accuracy was highest for the right inferior PV (96.6%) and lowest for the left superior PV (76.9%). The algorithm showed comparable accuracy to the classification by the EP specialists.
Conclusions: An automated farfield-nearfield discrimination based on two simple features from a single-beat BVE is feasible with a high specificity and comparable accuracy to the assessment by experienced cardiac electrophysiologists.
Keywords: Bipolar voltage electrogram; Farfield; Machine learning; Nearfield; Pulmonary vein isolation.
© 2023. The Author(s).