Artificial Intelligence-Enabled ECG Algorithm for the Prediction of Coronary Artery Calcification.

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging Analysis

Background:

• Coronary artery calcium (CAC) is a key indicator of atherosclerosis and cardiovascular risk.
• Current CAC scoring methods (CT scans) are limited by cost, radiation, and availability.
• Electrocardiograms (ECGs) are cost-effective and widely accessible diagnostic tools.

Purpose of the Study:

• To develop and validate artificial intelligence (AI) models capable of predicting CAC scores using only ECG data.
• To assess the generalizability of AI-predicted CAC models across different institutional datasets.
• To explore the potential of ECG-based AI for cardiovascular risk stratification.

Main Methods:

• Deep convolutional neural networks (CNNs) based on residual networks were constructed using raw ECG waveforms.
• AI models were trained to predict binary CAC levels: ≥100, ≥400, and ≥1,000.
• Model performance was evaluated using area under the receiver operating characteristic curve (AUROC) on internal and external validation datasets.

Main Results:

• AI models demonstrated strong performance in predicting CAC on internal datasets (AUROC 0.753-0.835).
• The models exhibited good generalizability, performing well on an external validation dataset (AUROC 0.718-0.803).
• Predictive performance improved with higher CAC score thresholds (≥100 to ≥1,000).

Conclusions:

• ECG-based AI models can effectively predict the presence and severity of coronary artery calcium.
• This AI approach offers a non-invasive, low-cost alternative for cardiovascular risk assessment.
• Widespread implementation could enable earlier detection and prophylactic treatment for ischemic heart disease.