An Artificial Intelligence-Enabled ECG Algorithm for the Prediction and Localization of Angiography-Proven Coronary...

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Diagnostics

Background:

• The diagnostic capability of artificial intelligence (AI) in conjunction with electrocardiograms (ECGs) for identifying significant coronary artery disease (CAD) remains largely unexplored.
• This study investigated the potential of AI to interpret ECGs for detecting significant CAD and localizing obstructed coronary arteries.

Purpose of the Study:

• To evaluate the efficacy of AI-driven analysis of standard 12-lead ECGs in diagnosing significant CAD.
• To determine if AI can accurately identify the specific coronary artery that is obstructed.

Main Methods:

• A multi-center retrospective cohort study was conducted using ECG data from patients with and without significant CAD, confirmed by invasive coronary angiography.
• Convolutional neural networks (CNN) models were trained on 12,954 ECGs from 2303 CAD patients and 2090 ECGs from 1053 control patients.

Main Results:

• The AI-enhanced CNN model achieved a Marco-average area under the ROC curve (AUC) of 0.869 for detecting significant CAD.
• The model demonstrated strong performance in localizing coronary artery obstructions: AUCs were 0.885 (left anterior descending), 0.776 (right coronary), and 0.816 (left circumflex).
• The AUC for CAD detection increased to 0.973 when ECGs exhibited features of myocardial ischemia.

Conclusions:

• This research demonstrates for the first time that an AI-enhanced CNN model can effectively utilize standard 12-lead ECGs to screen for significant CAD and identify the site of coronary obstruction.
• The AI-ECG approach offers a powerful, easily implementable tool for health check-ups, enabling early identification of high-risk individuals for future coronary events.