Early prediction of the efficacy of local epicardial radiofrequency ablation for the robotic treatment of persistent atrial fibrillation.

Background: Early prediction of the efficacy of local epicardial radiofrequency ablation (LERFA) is crucial for optimizing the robotic treatment of persistent atrial fibrillation. Objective: This study aimed to develop a machine learning model that accurately predicts LERFA efficacy within the first 5 seconds of the procedure, to stop ineffective procedures and reduce unnecessary cardiac tissue damage. Methods: Impedance data from 92 patients who underwent robotic LERFA were analyzed, with a total of 2486 LERFAs included in the final dataset. LERFA efficacy predictors, including zero-time impedance value, slope, and harmonic components, were extracted from the first 5 seconds of each time-impedance curve. Several supervised machine learning approaches were then tested to predict LERFA efficacy. Results: Random Forest demonstrated the highest performance, achieving 94.5% accuracy, 88.3% sensibility, and 97.2% specificity. This Random Forest model significantly outperformed the benchmark approach based on the zero-time impedance value alone, which achieved an accuracy of only 55.6% and a specificity of only 37.7%. Conclusion: The developed model enables fast and accurate prediction of LERFA efficacy, potentially reducing the number of completed LERFAs by 56.8%. This reduction results in minimal damage to cardiac tissue, a lower risk of complications, a reduction in operating time, and greater precision and safety in the ablation process.