PROJECT SUMMARY Although ablation to isolate pulmonary vein (PV) triggers has revolutionized atrial fibrillation (AF) management, performing effective AF ablation remains challenging. The procedure remains limited by targeting of ill-defined substrates, a 2-6% risk of major complications and limited success (single procedure 5-year success as low as 17-56%; 63-81% after the last ablation). A major recommendation of a recent NHLBI-sponsored report on the research needs and priorities for AF catheter ablation was to study how cardiac structure affects AF ablation success. There is a clear unmet need for non-invasive imaging tools to aid in improved patient selection, anatomic targeting and personalization of ablation or medical therapies. Our team has developed novel computational imaging (radiomics) methods to analyze cardiac computed tomography (CT) scans that were shown to predict the risk of recurrent AF post-ablation (AUC=0.84, N=167). These approaches included novel morphologic, fractal and atlas based features that teased out differences between PVs and the left atrial appendage (LAA), solely from analyses of CT scans. We propose to build upon our preliminary data using radiomic (computer extracted) features from radiographic images to use supervised and unsupervised machine learning methods that can analyze digitized radiographic and electro-anatomic images from the left atrium (LA) and PVs in over 2000 patients from two large AF ablation centers (Cleveland Clinic, Vanderbilt). Our project will focus on tackling the following main objectives: 1) Identify, evaluate and validate radiomic features and imaging-clinical nomograms predictive of recurrent AF after ablation; 2) Identify and validate regional radiomic sites predictive of post-ablation AF recurrence with the goal of identifying personalized targets for patients undergoing AF ablation; and 3) Identify biological correlates of radiomic features to understand the arrhythmogenic mechanisms underlying anatomic susceptibility to recurrent AF, using genomic analyses. Our 3 aims will test the following hypotheses: 1) Radiographic imaging can detect anatomic features that predict AF recurrence after ablation; 2) Regional radiomic features can predict sites that can be considered for additional ablation; and 3) Radiomic morphologic features are correlated with electroanatomic features and genomic variants associated with AF susceptibility. Tools developed will enable integration of radiographic and clinical data that may lead to improved patient selection, anatomic targeting and personalization of ablation or medical therapies. Successful project completion will yield a novel artificial intelligence-based imaging platform that can be tested for personalized targeting of AF ablation, as well as insights into the biologic basis of AF.