Explainable Machine Learning to Predict Anchored Reentry Substrate Created by Persistent Atrial Fibrillation Ablation...

Area of Science:

• Computational electrophysiology
• Medical imaging analysis
• Machine learning in cardiology

Background:

• Postablation arrhythmia recurrence affects ~40% of patients with persistent atrial fibrillation.
• Fibrotic remodeling and ablation-induced scar are implicated in recurrent arrhythmias.
• The interaction between ablation scar and residual fibrosis is poorly understood.

Purpose of the Study:

• To test the hypothesis that ablation creates substrate for anchored reentry-driven recurrent arrhythmia.
• To identify specific nonconductive tissue patterns conducive to arrhythmia.
• To understand the mechanisms of arrhythmia recurrence after ablation for persistent atrial fibrillation.

Main Methods:

• Computational simulations using patient-specific left atrial models from LGE-MRI.
• Development and training of a random forest machine learning classifier to identify arrhythmogenic substrate.
• Analysis of nonconductive tissue characteristics (size, proximity to fibrosis).

Main Results:

• A machine learning classifier accurately identified arrhythmogenic substrate (AUC: 0.91±0.03).
• A distinctive nonconductive tissue pattern, defined by size and proximity to residual fibrosis, was identified as arrhythmogenic.
• Ablation appears to transform substrate favoring functional reentry into one favoring anchored reentry.

Conclusions:

• Persistent atrial fibrillation ablation can create substrate conducive to anchored reentry.
• Explainable machine learning and computational simulations are valuable tools for studying arrhythmia mechanisms.
• Understanding substrate changes postablation is crucial for improving treatment outcomes.