Data Augmentation for Automatic Identification of Spatiotemporal Dispersion Electrograms in Persistent Atrial...

Area of Science:

• Cardiology
• Medical technology
• Machine learning

Background:

• Catheter ablation is a common treatment for atrial fibrillation (AF).
• Identifying specific ablation sites based on spatiotemporal dispersion (STD) is a recent advancement.
• Current methods for localizing STD sites rely on visual interpretation by cardiologists using specialized catheters.

Purpose of the Study:

• To develop and validate machine learning models for the automatic characterization and classification of STD sites in electrograms (EGMs).
• To address the challenge of imbalanced datasets in machine learning models for AF ablation.
• To enhance the precision and efficiency of identifying ablation targets for persistent AF.

Main Methods:

• Utilized a dataset of 23,082 multichannel EGM recordings from 16 persistent AF patients.
• Implemented data augmentation techniques including undersampling, oversampling, lead shift, time reversing, and time shift to handle data imbalance.
• Employed machine learning for classifying EGM data into STD and non-STD groups.
• Applied bootstrapping to assess classifier variability.

Main Results:

• Data augmentation, particularly oversampling, significantly improved classification sensitivity from 50% to 80%.
• Accuracy and Area Under the Curve (AUC) were maintained around 90% with oversampling.
• The developed ML techniques demonstrated effectiveness in classifying STD versus non-STD EGM data.

Conclusions:

• Machine learning models, enhanced by data augmentation, can accurately identify STD sites for AF catheter ablation.
• These automated tools are expected to assist cardiologists in tailoring ablation procedures for persistent AF patients.
• The study highlights the potential of ML to improve the efficacy of AF treatments.