Catheter Ablation -> Atrial Fibrillation & Atrial Flutter: -> Ablation Techniques D-PO02 - Poster Session II (ID 47) Poster

D-PO02-185 - Extensible Artificial Intelligence Model Predicts Post-ablation Af Recurrence Using Coronary Sinus Electrogram (ID 244)

Disclosure
 D. Huang: Nothing relevant to disclose.

Abstract

Background: Atrial fibrillation (AF) is a major public health problem with significant adverse outcomes and catheter ablation is a widely adopted treatment. The CABANA trial showed that catheter ablation reduced AF recurrence to a greater extent than medications. However, some of patients who underwent this procedure still experience relapse. Here, we present an innovative way to identify this subgroup using an artificial intelligence (AI) -assisted coronary sinus electrogram.
Objective: Our hypothesis is that credible features in the electrogram can be extracted by AI for prediction, therefore rigorous drug administration, close follow-up or potential second procedure can be applied to these patients.
Methods: 67 patients from two independent hospitals (SPH & ZSH) with non-valvular persistent AF undergoing circumferential pulmonary vein isolation were enrolled in this study, 23 of which experienced recurrence6 months after the procedure. We collected standard 2.5-second fragments of coronary sinus electrogram from ENSITE NAVX (SPH) and Carto (ZSH)system before the ablation started. A total of 1429 fragments were obtained and a transfer learning-based ResNet model was employed in our study. Fragments from ZSH were used for training and SPH for validation of deep convolutional neural networks (DCNN) . The AI model performance was evaluated by accuracy, recall, precision, F-Measure and AUC .
Results: The prediction accuracy of the DCNN in single center reached 96%, while that in different ablation systems reached 74.3%. Also, the algorithm yielded values for the AUC, recall, precision and F-Measure of 0.76, 86.1%, 95.9% and 0.78, respectively, which shows satisfactory classification results and extensibility in different cardiology centers and brands of electroanatomic mapping instruments.
Conclusion: Our work have revealed the potential intrinsic correlation between coronary sinus electrical activity and AF recurrence using DCNN-based model. Moreover, the DCNN model we developed shows great prospects in the relapse prediction for personalized post-procedural management.
Collapse