AI Tool Demonstrates High Accuracy for Coronary Calcium Scoring on Routine Chest CTs

Researchers at Mass General Brigham have developed and validated an artificial intelligence (AI) model capable of accurately estimating coronary artery calcium (CAC) scores from non-gated chest CT scans, potentially transforming cardiovascular risk screening. The findings, recently published in Scientific Reports, highlight a scalable approach to opportunistic atherosclerosis detection using imaging data already collected for other clinical purposes.

The AI tool focuses on calculating the Agatston score—a key indicator of coronary artery disease—without the need for dedicated cardiac-gated CT scans. This innovation could increase access to early cardiovascular risk assessment, reduce the need for additional imaging, and avoid extra radiation exposure.

The model was tested on 491 non-contrast chest CTs acquired across five hospitals between January 2022 and December 2023. For two-thirds of the cases, matching cardiac-gated CTs were available for comparison. When benchmarked against radiologist consensus on the non-gated scans, the AI achieved a quadratic weighted kappa of 0.959 and an accuracy of 92.3% in categorizing Agatston scores (0, 1–99, 100–399, ≥400). Compared to gated cardiac CT, the kappa was 0.906, with a strong Spearman correlation of 0.942.

The model also performed consistently across different patient demographics and technical conditions, including age, sex, race, ethnicity, scanner type, and radiation dose settings. For continuous Agatston scores, the model achieved a Spearman correlation of 0.975 with radiologist-generated reference values, further validating its performance.

According to the study team, these results support the use of AI to extend coronary artery disease screening to a broader patient population. Non-gated chest CTs are commonly acquired for evaluating lung or other thoracic conditions, presenting an untapped opportunity to assess cardiovascular risk without additional testing.

The researchers acknowledged limitations, including the retrospective design and the need for prospective studies to measure clinical impact. They noted that future work will assess how integration of AI scoring into routine workflows affects patient outcomes and resource utilization.

This AI-driven approach could be a pivotal tool in addressing the growing burden of cardiovascular disease by enabling earlier diagnosis and preventive care—especially in settings where access to cardiac-specific imaging is limited.