Deep learning model to generate patient-specific pulmonary vein isolation lines from successful atrial fibrillation ablation cases: a proof-of-concept study.

Pulmonary vein isolation (PVI) is an established standard ablation for atrial fibrillation (AF), however, AF recurrence remains a major clinical challenge. We developed a deep learning model to generate patient-specific PVI lines on pre-ablation 3D voltage maps, using lesion sets from successful AF ablation cases with documented freedom from recurrence for more than one year as ground truth. Using a U-Net-based architecture trained and evaluated on 513 maps from 171 such cases, the model reproduced the anatomical and electrophysiological features of these PVI lesion sets. On the held-out test set, the model achieved a mean Intersection over Union of 0.87 and a Dice score of 0.93. As a proof of concept, these findings suggest that the model can reproduce patient-specific PVI patterns associated with successful outcomes; whether this translates into reduced recurrence requires prospective clinical validation.