Data Augmentation for Automatic Identification of Spatiotemporal Dispersion Electrograms in Persistent Atrial Fibrillation Ablation Using Machine Learning

Catheter ablation is increasingly used to treat atrial fibrillation (AF), the most common sustained cardiac arrhythmia encountered in clinical practice. A recent breakthrough finding in AF ablation consists in identifying ablation sites based on their spatiotemporal dispersion (STD). STD stands for a delay of the cardiac activation observed in intracardiac electrograms (EGMs) across contiguous leads. In practice, interventional cardiologists localize STD sites visually using the PentaRay multipolar mapping catheter. This work aims at automatically characterizing STD by classifying EGM data into STD vs. non STD groups using machine learning (ML) techniques. A dataset of 23082 multichannel EGM recordings acquired by the PentaRay coming from 16 persistent AF patients is included in this study. A major problem hampering the classification performance lies in the highly imbalanced dataset ratio. We suggest to tackle data imbalance using adapted data augmentation techniques including 1) undersampling 2) oversampling 3) lead shift 4) time reversing and 5) time shift. These tools are designed to preserve the integrity of the cardiac data and are validated by a partner cardiologist. They provide enhancement in classification performance in terms of sensitivity, which increases from 50% to 80% while maintaining accuracy and AUC around 90% with oversampling. Bootstrapping is applied to check the variability of the trained classifiers. Clinical relevance The machine learning techniques developed in this contribution are expected to aid cardiologists in performing patient-tailored catheter ablation procedures for treating persistent AF.