Po-03-092 machine learning analysis of the predictors of af recurrence post-ablation in patients undergoing pre-ablation mri

Catheter ablation, a widely performed rhythm control strategy for atrial fibrillation (AF), has variable results in the complete suppression of arrhythmia recurrence, especially in obese patients. Structural remodeling of the left atrium (LA), evidenced by LA enlargement and fibrosis, is a well-known substrate for AF. Epicardial adipose tissue (EAT) is thought to be a crucial factor mediating cardiovascular disease in obesity, it can result in electrical heterogeneity and local conduction block forming a substrate for AF. We aimed to apply a machine learning approach to assess the predictors of AF recurrence following catheter ablation in AF patients undergoing pre-ablation cardiac magnetic resonance (CMR). 101 AF patients underwent CMR prior to catheter ablation. Fibrotic remodeling was assessed using the late gadolinium enhancement sequence. EAT was assessed using the novel fat-water separation 3D Dixon sequence. Panel A shows posterior and anterior views of a representative LA model showing fibrotic remodeling and EAT. Patients were followed for AF recurrence using ambulatory monitoring and 12-lead ECGs. A random forest classifier with 5x stratified cross-validation was trained to predict AF recurrence post-ablation and SHapley Additive exPlanations (SHAP) analysis was used to assess feature importance in the prediction of AF recurrence. During an average follow-up period of 1 year, post-ablation AF recurrence occurred in 31 (30.7%) patients. Patients with AF recurrence post-ablation had higher LA EAT index (20.7 [16.9, 30.4] vs 13.7 [10.5, 20.1] mL/m2, p<0.001), and higher LA volume index (66 [52.6, 77.5] vs 49.9 [37.7, 61.8] mL/m2, p=0.001). There was a trend toward a higher fibrosis burden in patients who had an AF recurrence (18.6% vs 16.1%, p=0.076). The classifier predicted AF recurrence post-ablation with an AUC of 0.67 ± 0.09 (Panel B). SHAP analysis revealed that for the overall population, the three most important features in the prediction of AF recurrence post-ablation were LA volume index, LA EAT index, and LA fibrosis (Panel C). LA enlargement, fibrosis and EAT, markers with clear ties to AF pathophysiology, are predictors of AF recurrence after ablation. Future studies testing the value of an imaging-based ablation failure prediction calculator, as well as studies to elucidate the mechanism of ablation failure, are required.