Every year, millions of chest CT scans are performed for lung screenings or other reasons, producing images rich with untapped clues about a patient’s health. Now, researchers say a new artificial intelligence tool could unlock those clues, potentially reshaping how doctors identify cardiovascular risk long before symptoms appear.

Massachusetts General Brigham researchers, in collaboration with the U.S. Department of Veterans Affairs, have developed AI-CAC, a deep-learning algorithm that analyzes routine chest CT scans to detect and quantify coronary artery calcium (CAC), a key predictor of future heart disease. Their findings, published on the hospital’s website, and The New England Journal of Medicine, show that the tool can accurately flag patients at higher risk for heart attacks and long-term mortality. Their research, and similar studies conducted, suggest that AI could become a powerful ally in early cardiovascular prevention.

Knowing what to do with mountains of data is a problem across many sectors, and trying to figure out what all that information holds can be a years-long, if not career-long pursuit, but  AI’s ability to do such is transforming many industries, and perhaps the most profound application is in healthcare.  

“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,” said Dr. Hugo Aerts, director of the Artificial Intelligence in Medicine Program at Mass General Brigham. “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.”

The core challenge lies in the type of scans. The standard for measuring CAC is a gated CT scan, which syncs with a patient’s heartbeat to reduce motion. But most CTs done for general clinical purposes are non-gated, meaning they were never designed to evaluate heart risk.  Dr. Aerts and his team believed those images still held valuable information — they just needed a tool powerful enough to find it.

To test their theory, the team built AI-CAC using chest CT scans collected from 98 VA medical centers, training the model on “expert-labeled data” and testing it on 8,052 additional scans to simulate how it might work in everyday practice.

The AI-CAC correctly determined whether a scan contained calcium deposits 89.4% of the time, and accurately distinguished moderate from high-risk cases 87.3% of the time. Patients with CAC scores above 400 faced a 3.49-times greater risk of death over 10 years than those with a score of zero. In a sample of patients with very high scores, four cardiologists verified that 99.2% of them would benefit from lipid-lowering therapy.

“At present, VA imaging systems contain millions of non-gated 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 nongated scans for purposes of cardiovascular risk evaluation and to enhance care,” said Dr. Raffi Hagopian, a cardiologist and researcher in the VA’s Applied Innovations and Medical Informatics group in Long Beach. “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.”

For decades, cardiologists have known that CAC is one of the strongest predictors of cardiovascular events, but the need for specialized scans has kept widespread screening out of reach, researchers say. By turning ordinary chest CTs into dual-purpose, AI-CAC may make risk detection routine.

The researchers acknowledge limitations. Because the model was trained solely on data from veterans, more research is needed to ensure its accuracy across broader populations. The team hopes to test whether AI-CAC can also monitor the effectiveness of cholesterol-lowering therapies over time.

A parallel study led by VA Long Beach Healthcare System and Harvard Medical School reached similar conclusions. “Using AI has the potential to provide clinicians with a powerful new tool to identify heart disease risk for patients earlier on, enabling them to proactively manage their health,” said Dr. Hagopian. “By helping shift medicine from a reactive approach to the proactive prevention of disease, we can save lives, improve outcomes, and reduce healthcare costs.”

Dr. Jeannie Yu, staff cardiologist with the VA Long Beach Healthcare Group, said VA’s vast imaging archive makes it uniquely suited for this. “With access to millions of chest CT scans and outcomes data, VA is able to conduct groundbreaking cardiovascular research at unprecedented scale, powered by artificial intelligence, to provide clinical insights and precision medicine that enables the potential to identify and treat heart disease earlier, not only for our Veterans, but also the public at-large.”

As medicine continues to embrace AI, tools like AI-CAC may play a larger role, delving into those mountains of data, scans, surveys, and other information, to reveal what could be hiding in plain sight.