Using ECGWiz: A Multimodal Deep Learning Framework
Atrial fibrillation (AF) is the most common cardiac arrhythmia, affecting millions worldwide. Current treatment strategies often require invasive electrophysiological mapping to understand the underlying atrial substrate and guide therapy selection. This project introduces ECGWiz, a novel multimodal deep learning framework that uses readily available 12-lead surface ECGs as a surrogate for invasive mapping.
By extracting substrate-level information non-invasively, ECGWiz enables clinicians to predict atrial electrical remodeling patterns and make informed therapy selection decisions without subjecting patients to invasive procedures.
12-lead surface ECG recordings from AF patients, paired with invasive electrophysiological mapping data for ground truth.
Multimodal deep learning model combining CNN-based signal processing with attention mechanisms for substrate prediction.
Cross-validated against invasive mapping outcomes. Demonstrated clinical utility for therapy selection in AF patients.
Established that standard 12-lead surface ECGs contain sufficient information to serve as surrogates for invasive atrial electrical mapping, enabling non-invasive substrate characterization.
Developed a novel multimodal deep learning architecture that integrates temporal and spatial ECG features for accurate prediction of atrial remodeling patterns.
Demonstrated potential for real-world clinical application in guiding AF therapy selection without requiring invasive procedures.
Incorporated attention visualization and feature importance analysis to provide clinically interpretable predictions.
Vatsaraj, I., Shade, J. K., Ali, R. L., Popescu, D. M., & Trayanova, N. A. (2025). "ECGWiz: Non-Invasive Prediction of Atrial Electrical Remodeling Using 12-Lead Surface ECGs." Journal of Electrocardiology.