AI spots heart disease warning signs in routine chest scans

Heart disease still tops the list of causes of death in the United States, taking more than 700,000 lives in 2022 alone.

Millions of people undergo computed tomography (CT) of the chest each year for reasons unrelated to the heart, but these scans also serve as silent heart scans. They capture calcium deposits that warn of future coronary artery trouble.

Dr. Hugo Aerts and his team at Mass General Brigham, working with the U.S. Department of Veterans Affairs, built an algorithm named AI‑CAC to turn overlooked data from CT scans into an early warning system.

Hidden calcium on scans predicts heart risk

The white flecks of coronary artery calcium are mineral scars of past plaque damage. Decades of studies show that any calcium is a red flag, and a score above 400 can triple the 10‑year risk of death or heart attack .

Despite this, routine nongated chest CTs are generally read for lung or spine findings, leaving the heart unattended. That gap means people often discover their risk only after symptoms appear, when lifestyle changes or statins have less time to work.

AI finds calcium that doctors often miss

AI‑CAC is built on deep learning, a method that lets software teach itself patterns by reviewing thousands of labeled images. The developers trained the network on scans from 98 VA hospitals, capturing a wide mix of scanners and patient body types.

On an 8,052‑scan test set the model spotted any calcium with 89 percent accuracy and correctly split scores above or below 100 in 87 percent of cases. Those numbers rival expert readers yet require a fraction of the time, seconds rather than minutes.

“Millions of chest CT scans are taken each year, often in healthy people.” said Aerts. He believes that teaching the algorithm to comb existing archives can “enable physicians to engage with patients earlier, before their heart disease advances to a cardiac event.”

Heart scan scores predict death risk

Patients whose AI‑CAC score topped 400 faced a 3.49‑fold jump in 10‑year all‑cause mortality compared with those who scored zero.

Four independent cardiologists reviewed a random sample of these high‑score scans. The experts agreed that 99 percent of the patients would benefit from cholesterol‑lowering medication, underscoring the model’s clinical relevance.

The strength of the project lies in its size and diversity; many prior calcium‑AI efforts used gated scans from a single vendor. A 2021 Stanford‑led paper showed similar accuracy, but with just a few hundred test cases.

From veterans to everyday clinics

The dataset came exclusively from veterans, a group that is older and more often male than the general public.

Study lead author Dr. Raffi Hagopian acknowledged the limitation, yet called it an ideal first step. He noted that VA imaging archives already hold “millions of nongated chest CT scans” suited for opportunistic screening.

Rolling the tool into other health systems will require validation across broader demographics, different scanner models, and varied radiation doses. That work is starting, with several academic centers planning head‑to‑head tests later this year.

“Using AI for tasks like CAC detection can help shift medicine from a reactive approach to the proactive prevention of disease,” said Dr. Hagopian. He pointed to the potential for fewer heart attacks, lower costs, and better shared decision‑making between patients and clinicians.

Heart scan tool fits existing guidelines

The 2019 American College of Cardiology/American Heart Association guideline already endorses starting statin therapy when a calcium score reaches 100 in select patients.

Today, getting that number usually means scheduling a separate gated scan that insurers may not cover, limiting uptake.

AI‑CAC circumvents the hurdle by reading the scans people have already undergone for other reasons. If future trials confirm its performance, a simple software update in the radiology workstation could make calcium scoring routine.

Opportunities and concerns

Automating calcium detection promises to relieve radiologists of a tedious task and flag high‑risk patients before they leave the imaging suite. However, false positives could provoke anxiety or unnecessary follow‑up, so clinicians will need clear protocols.

Data security is another concern; VA hospitals operate behind stringent firewalls, and private centers will expect equally robust safeguards.

Finally, health systems must decide who owns the liability when an algorithm misfires and how to reimburse the extra counseling that follows each alert.

AI and heart scans in more hospitals

The next milestones include testing AI‑CAC in community hospitals, monitoring whether early statin prescriptions actually change outcomes, and integrating the calcium score into electronic health record dashboards alongside blood pressure and cholesterol.

Developers are also exploring whether the neural network can measure aortic and valvular calcium in the same pass, extending its reach without extra radiation.

Clinicians have long known that calcium foretells trouble, but counting those specks by hand never scaled. AI‑CAC offers a path to put that knowledge to work for every patient whose chest has ever met a CT scanner.

The study is published in the journal NEJM AI.

—–

Like what you read? Subscribe to our newsletter for engaging articles, exclusive content, and the latest updates. 

Check us out on EarthSnap, a free app brought to you by Eric Ralls and Earth.com.

—–