According to a study published in NEJM AI, Mass General Brigham researchers collaborated with the United States Department of Veterans Affairs (VA) to create a new AI tool that can search through previously collected CT scans and identify individuals with high coronary artery calcium (CAC) levels, which puts them at a higher risk for cardiovascular events.
Image Credit: SewCream/Shutterstock.com
Their findings revealed that the AI-CAC tool was highly precise and predictive of future heart attacks and 10-year mortality. Their findings indicate that widespread use of such a tool might assist doctors in assessing their patients' cardiovascular risk.
Millions of chest CT scans are taken each year, often in healthy people, for example, to screen for lung cancer. Our study shows that important information about cardiovascular risk is going unnoticed in these scans. Our study shows that AI has the potential to change how clinicians practice medicine and enable physicians to engage with patients earlier, before their heart disease advances to a cardiac event.
Hugo Aerts, PhD, Study Senior Author and Director, Artificial Intelligence in Medicine Program, Mass General Brigham
Chest CT scans can reveal calcium deposits in the heart and arteries, which raises the risk of a heart attack. The gold standard for assessing CAC is “gated” CT scans, which coordinate with the heartbeat to eliminate motion during the scan. However, most chest CT images used for regular clinical purposes are “nongated.”
The researchers discovered that CAC may still be identified on these nongated scans, so they created AI-CAC. This deep learning system probes through the nongated scans and quantifies CAC to assist in forecasting the risk of cardiovascular events.
They trained the model using chest CT scans obtained as part of normal treatment for veterans at 98 VA medical sites. Then they assessed AI-CAC's performance on 8,052 CT images to emulate CAC screening in routine imaging examinations.
The researchers discovered that the AI-CAC model was 89.4% accurate in detecting whether a scan included CAC. For those with CAC, the model was 87.3% accurate in predicting whether the score was greater or lower than 100, suggesting a moderate cardiovascular risk.
AI-CAC was also predictive of 10-year all-cause mortality—patients with a CAC score of 400 or above had a 3.49 times greater risk of dying over a decade than those with a score of zero. Four cardiologists confirmed that 99.2% of the patients identified by the model as having very high CAC scores (more than 400) would benefit from lipid-lowering treatment.
At present, VA imaging systems contain millions of nongated chest CT scans that may have been taken for another purpose, around 50,000 gated studies. This presents an opportunity for AI-CAC to leverage routinely collected non-gated scans for purposes of cardiovascular risk evaluation and to enhance care. Using AI for tasks like CAC detection can help shift medicine from a reactive approach to the proactive prevention of disease, reducing long-term morbidity, mortality, and healthcare costs.
Raffi Hagopian, MD, Study First Author, Cardiologist and Researcher, Applied Innovations and Medical Informatics Group, VA Long Beach Healthcare System
The study’s limitations include the fact that the algorithm was created only for veterans. The team aims to perform additional research in the general population to see if the technique can predict the effect of lipid-lowering drugs on CAC scores.
Veterans Affairs health care system funded the study.
Journal Reference:
Hagopian, R., et al. (2025) AI Opportunistic Coronary Calcium Screening at Veterans Affairs Hospitals. NEJM AI. doi.org/10.1056/aioa2400937.