Automated Coronary Artery Calcium Scoring Using Deep Learning: Validation Across Diverse Chest CT Protocols

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Healthcare
• Cardiovascular Disease Risk Assessment

Background:

• Coronary artery calcium (CAC) scoring is crucial for atherosclerotic cardiovascular disease (ASCVD) risk stratification.
• Routine non-gated chest CT (NCCT) use has increased, presenting an opportunity for opportunistic CAC assessment.
• Current methods for CAC quantification on NCCT are not widely integrated into clinical workflows.

Purpose of the Study:

• To develop and validate a deep learning (DL) model for automated, protocol-agnostic CAC quantification.
• To enable workflow-ready CAC scoring from routine chest CT scans.
• To support opportunistic cardiovascular risk stratification in clinical practice.

Main Methods:

• A retrospective study involving 2132 chest CT scans (routine, CT-CAC, CT-COVID) from patients without established ASCVD.
• Training and validation of a DL-based CAC segmentation model against manual annotations.
• Evaluation of agreement using intra-class correlation coefficients (ICC) and Cohen's kappa.
• Calculation of diagnostic performance metrics including sensitivity, specificity, and F1 scores.

Main Results:

• The DL model showed high reliability for Agatston scores (ICC=0.987) and strong agreement in CAC categories (Cohen's κ=0.86-0.95).
• Excellent diagnostic performance was observed for CAC >100 (F1=0.956) and CAC >300 (F1=0.967).
• Good agreement was confirmed through external validation in the Mashhad COVID Study (κ=0.8) and SBU COVID study (F1=0.928 for moderate-to-severe CAC).

Conclusions:

• The developed DL model provides accurate and workflow-ready CAC quantification.
• The model is effective across various chest CT scan types, including routine and pandemic-era scans.
• This technology supports cost-effective, opportunistic cardiovascular risk stratification in clinical settings.