Abstract Body (Do not enter title and authors here): Background:
Pulmonary vein isolation (PVI) has been shown to reduce morbidity and mortality in patients with atrial fibrillation (AF). However, a major challenge is identify which patients will or will not respond.

Hypothesis:
We hypothesized that patients in whom AF activity predominantly exits the pulmonary veins (PV), identified using a novel artificial-intelligence (AI) wave tracking approach, will respond to ablation compared to patients in whom activity predominantly enters the PVs.

Methods:
We examined 26,640 electrograms in N=37 patients at AF ablation (age 66.3 ± 8.6 years, 13.5% females, 59.5% non-paroxysmal AF). Patients had AF recordings with multipolar catheters. We analyzed AF in 3X3 electrode grids, representing 1-3 cm² areas that approximate minimum tissue wavelength, near and remote from PVs. For each grid, we determined if AF waves exited PVs (i.e. active PVs) or entered PVs using AI-algorithms to assign electrode activations and grid vectors. We compared the numbers of waves that exited versus entered PVs relative to ablation success, defined using ambulatory ECGs over 1 year.

Results:
Spatial grids in Fig A show AF waves exiting the left superior PV in a 60 year old man with persistent AF whose PVI was successful. Fig B shows AF waves entering the PVs in a 70 year old woman whose subsequent PVI was unsuccessful. Examining 720 grids per patient (48 four second grids over 1 minute), AF waves varied considerably. Fig. C shows that more AF waves exited the PVs in patients with PVI success than in patients with PVI failure (p=0.03). Conversely, more AF waves headed towards the PVs in patients with PVI failure than with PVI success (p=0.04).

Conclusions:
In this study, AI-based wave identification enabled the identification of AF patients in whom PVI was more or less successful. The approach to determine active versus passive PVs could be extended to other metrics, including non-invasive tools.