Can machine learning models using ultra-high-density mapping and clinical data predict ablation gaps and recurrence of atrial tachycardia in patients undergoing pulsed field ablation for atrial fibrillation?
Ultra-high-density mapping combined with machine learning can identify ablation gaps missed by standard PFA catheters and predict the risk of post-ablation roof-dependent atrial tachycardia.
BACKGROUND AND AIMS: Pulsed field ablation (PFA) has emerged to an innovative approach to achieve pulmonary vein isolation (PVI) in atrial fibrillation (AF) treatment. Despite its fast adoption and promising safety profile, insights into immediate ablation effects and lesion characteristics and their influence on follow-up arrhythmia recurrence using ultra-high-density mapping (UHDM) are sparse. This study aims to evaluate acute lesion dynamics and formation and their clinical impact using UHDM. METHODS: This study enrolled 204 patients undergoing PVI with a pentaspline PFA system. UHDM was used for pre- and post-ablation assessment of the pulmonary veins (PV) and left atrium (LA). Clinical and mapping data were analyzed to define immediate lesion formation. Machine learning (ML) techniques, including SMOTE for data augmentation, were utilized to predict the recurrence of atrial tachycardia during follow-up and understand their underlying mechanisms. RESULTS: UHDM of immediate outcomes showed typical isolation patterns around the PV ostia. UHDM revealed a significantly narrowed electrically intact bridge on the LA roof. Furthermore, UHDM detected 14 non-isolated PV gaps in 13 patients, matching the typical lesion distribution. Gaps were undetected by the pentaspline PFA catheter. During follow-up, LA roof-dependent tachycardia was the most common recurrent arrhythmia (n = 11, 5.4%). ML model demonstrated size of the electrically intact tissue-bridge within the LA roof after PFA combined with LA size as predictors of occurrence of this specific tachycardia (AUC, 0.86; Sens., 0.86; Spec., 0.82). Additionally, ML models identified LA size and persistent AF as key predictors of gap presence (AUC, 0.88; Sens., 0.90; Spec., 0.83). CONCLUSION: ML models build with UHDM and clinical data can identify patients at risk for LA roof-dependent atrial tachycardia during follow-up and ablation gaps. Ablation strategies adapted to this information may potentially improve long-term outcomes in AF management.
Feickert et al. (Thu,) studied this question.