Semi-Supervised Learning for Enhancing Ablation Outcomes in Persistent Atrial Fibrillation

Introduction: The outcome of ablations of AF targets in persistent atrial fibrillation (persAF) using dominant frequency (DF), rotors, and complex fractionated atrial electrograms (CFAEs) has been disappointing.Machine learning tools with novel electrogram (EGM) features from the three EGM domains (spectral, temporal, and statistical) by utilizing the labeled and unlabeled EGMs might enhance ablation outcomes.Methods: 20480 noncontact EGMs were collected from the left atrium of 10 persAF patients using the Ensite Array.51 high dominant frequency locations (3206 nodes (EGMs)) were ablated.1490 EGMs were labeled as positive responses (AF termination or AF cycle length (AFCL) increased (10msec), whereas 1716 EGMs had negative responses (AFCL increase (<10msec))) to catheter ablation.390 features were extracted, followed by feature selection and importance stages.Supervised learning (SL) was used to train and test the decision tree (DT) model.Subsequently, semi-supervised self-learning (SSSL) was used to take advantage of unlabelled data (17274 EGMs).Results: The accuracy and AUROC for SL were 72%, and 0.71; while for SSSL were 75.22%, and 0.73, respectively.Conclusions: SSSL played a significant role in increasing the model performance by 3.22% and 0.02 for accuracy and AUROC, respectively.