Deep learning modeling in patients scheduled for premature ventricular contraction (PVC) ablation

Aim: To determine the localization of outflow tract premature ventricular contractions (PVCs) with high accuracy and sensitivity using artificial intelligence models trained with 12-lead surface ECG data, and to investigate their relationship with the prediction of recurrence after RF catheter ablation. Materials and Methods: Outflow tract PVCs catheter ablations performed at our center between July 2020 and July 2025 were analyzed retrospectively. A deeper and wider 1-D convolutional neural network architecture (1-D CNN, MultiQRSNet), using a 13-channel input for multi-channel classification from 12-lead ECGs, was developed. Three models were trained: (i) Right–Left Model, (ii) Right–Left–Intramural Model, (iii) Right–Left Recurrence Model. Inter-group results were compared; recurrence predictors were evaluated with multivariate analyses. Results: Patients with right ventricular outflow tract (RVOT) PVCs were younger (45.8±15.7 vs 58.4±11.3 and 55.1±13.9 years; p<0.001), had a lower BMI (27.1±3.9 vs 30.5±5.4 and 29.9±5.5 kg/m²; p=0.001), and fewer comorbidities (hypertension 22% vs 53% and 46%; p=0.004; diabetes 5% vs 22% and 27%; p=0.012) compared to the left ventricular outflow tract (LVOT) and intramural groups. The performance of all three models was statistically significant (p<0.01). The Right–Left model achieved 98.22% accuracy and 98.22% sensitivity in distinguishing between right and left outflow tracts. The Right–Left–Intramural model obtained 60.73% accuracy in the three-class classification. The Right–Left Recurrence model reached 69.83% accuracy in the "cure" and "recurrence" classification. In the multivariate analysis, the ABC-VT risk score was identified as an independent predictor of recurrence (OR: 1.21; p=0.044). Conclusion: The artificial intelligence model developed in our study allows for the identification of intramural foci, in addition to distinguishing between right and left outflow tracts with high precision. For recurrence prediction, the Right-Left Recurrence model can be used in conjunction with the ABC-VT risk score. Our results may aid in treatment selection for patients presenting with frequent PVCs, helping to reduce procedural site complications, procedure time, and costs. Keywords: artificial intelligence; intramural foci; outflow tract PVCs;