Heart Failure -> Heart Failure Management: -> Monitoring D-MP02 - Newer technologies in managing and predicting outcomes in HF and AF patients (ID 50) Moderated ePoster

D-MP02-06 - Ct Image Is Better Than Clinical Parameters In Prediction Of Recurrence Using Artifical Intelligence In Atrial Fibrillation Ablation (ID 6)

Abstract

Background: The artificial intelligence (AI) has been applied in image recognition to facilitate clinical practice.
Objective: We aimed to compared the accuracy in prediction of atrial fibrillation (AF) recurrence between AI model usingpre-ablation pulmonary vein computed tomography (PVCT) geometric slicesand AI model of clinical parameters in patients with paroxysmal AF.
Methods: We retrospectively analyzed 521 PAF patientswho have undergone catheter ablation of PAF (Table 1). PVCT geometric slices (2-3mm for each slice, 20-200 slices for each patient, total 36943 images of slices in 521 AF patients) were used in the deep learning process (ResNet34) for prediction of AF recurrence.The clinical data was collected for analysis.
Results: Over 1-year follow-up, there was no AF recurrence in 358 (68.7%) patientsand AF recurrence in 163 (31.3%) patients.For each PVCT image, the AUC, accuracy, sensitivity, and specificity of prediction for post-ablation AF recurrence can achieve 0.86, 80.9%, 100.0%, and 62.7% in each patient, respectively.For each patient, the AUC, accuracy, sensitivity, and specificity of prediction for post-ablation AF recurrence can achieve 0.85, 78.5%, 100.0%, and 57.1% in each patient, respectively. For clinical datasets, the AUC, accuracy, sensitivity, and specificity were0.58, 73.6%, 57.1%, and 75.2%, respectively (Figure).
Conclusion: Deep-learning AI using pre-ablation cardiac CT can be applied in the prediction of recurrence in PAF patients receiving catheter ablation. The application of this model can classify patients with high risk for AF recurrence and provide more information for shared decision making in terms of AF treatment.
Collapse