Deep Learning Prediction of Left Atrial Structure and Function from 12-lead Electrocardiograms

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Atrial cardiopathy, characterized by abnormal atrial structure and function, often precedes atrial fibrillation (AF) and other cardiovascular issues.
• Current detection methods rely on costly and less accessible cardiac imaging techniques.

Purpose of the Study:

• To develop and validate a deep learning model (ECG-AI) that predicts left atrial structure and function using 12-lead electrocardiograms (ECGs).
• To assess the efficacy of ECG-AI in identifying individuals at high risk for AF and associated cardiovascular complications.

Main Methods:

• A deep learning model (ECG-AI) was trained using 12-lead ECGs from 21,749 individuals, paired with cardiac magnetic resonance imaging (CMR) scans.
• The model's predictions of atrial cardiopathy were validated in two external cohorts, comparing its performance against imaging measures and clinical risk factors.
• ECG-AI's ability to predict subclinical AF was evaluated in a screening population using 14-day cardiac monitoring.

Main Results:

• ECG-AI measures of atrial cardiopathy showed strong associations with new-onset AF, heart failure, and ischemic stroke, even after adjusting for clinical factors.
• The model outperformed traditional imaging measures and clinical risk factors in predicting these cardiovascular events.
• Increased left atrial volume, as predicted by ECG-AI, was linked to a 66% greater risk of cardioembolic stroke per standard deviation.
• ECG-AI demonstrated superior prediction of subclinical AF compared to a clinical risk prediction tool in a screening setting.

Conclusions:

• ECG-AI is an inexpensive and accessible tool for identifying individuals with atrial cardiopathy and high risk for AF.
• This AI-driven approach offers a promising alternative to traditional imaging for cardiovascular risk assessment and early detection of AF.