AI algorithm turns mammograms into a ‘two-for-one’ test for women’s heart health

An AI algorithm based only on routine mammogram images plus age can predict a woman’s risk of major cardiovascular disease as well as standard risk assessment methods, finds research published online in the journal Heart.

And because it uses existing health infrastructure, routine mammography may offer a cost-effective ‘two for one’ effective screening option for women, suggest the researchers.

Cardiovascular disease and its risk factors are underrecognized and undertreated in women, and risk prediction algorithms have underperformed in them, point out the researchers. And while newer risk scores perform better in women than in men, they are complex and their accuracy depends on extensive medical data, which isn’t always available, they add.

The extent of arterial calcium deposits (BAC) and tissue density in the breasts have been linked to cardiovascular disease risk, but BAC isn’t associated with obesity and is negatively associated with smoking, suggesting that it’s not effective by itself, they suggest.

They therefore set out to discover if an automated AI analysis of the full range of internal breast structure and characteristics from routine mammogram images might be more accurate at cardiovascular risk prediction.

They drew on 49,196 women with an average age of 59, enrolled between 2009 and 2020 in the Lifepool cohort registry, and living in Victoria, Australia.

At enrollment, the women provided initial background health information on their age, smoking status, alcohol intake, weight (BMI), any history of diabetes and use of high blood pressure and/or high cholesterol drugs and blood thinners.

Additional information included menopause status, reproductive history and use of hormone therapy, as well as factors potentially affecting the internal structure of the breast, such as radiation, surgery, and cancer.

Some 5% of the women were current smokers, 62% had a BMI of more than 25, 6% had type 2 diabetes, 33% were taking drugs for high cholesterol, 27% for high blood pressure, and 11% were taking a blood thinner.

During an average tracking period of nearly 9 years, 3392 of these women had a first cardiovascular ‘event’: coronary artery disease (2383); heart attack (656); stroke (434) or heart failure (731).

The researchers developed an AI algorithm based on the full complement of internal breast structures and features from the mammogram images plus the woman’s age to predict major cardiovascular disease risk over 10 years.

This AI algorithm was as good as modern risk scores based on age and various clinical factors, including the New Zealand ‘PREDICT’ tool and the American Heart Association ‘PREVENT’ calculator. And it was only slightly better when various clinical factors were added.

The researchers acknowledge several limitations to their findings, including that different scanners don’t produce exactly the same data; the cardiovascular risk factors used for comparison relied on self-report; and all deep learning models are entirely dependent on their training datasets.

But they say, “A key advantage of the mammography model we developed is that it did not require additional history taking or medical record data and leveraged an existing risk screening process widely used by women.

“Mammography has potential as a ‘two-for-one’ risk assessment tool, offering efficiencies for both community and the health care system.”

They concede that “the use of mammography images to predict cardiovascular risk is novel, but the use of machine learning models to do cardiovascular risk prediction is gaining traction.”

In a linked editorial, Professor Gemma Figtree and Dr. Stuart Grieve of the University of Sydney, point out that the poor performance of traditional risk factor algorithms in women is compounded by lack of awareness of the threat posed by heart disease to women by both women themselves and the health system.

“In contrast with what is commonly thought, breast cancer causes only about 10% of the total deaths globally compared with those resulting from cardiovascular disease,” they write.

“Mammography may therefore represent a ‘touch point’ for raising awareness about cardiovascular risk and disease in women,” they suggest.

But they add, “One of the challenges with new tools that show promise for improved cardiovascular risk assessment remains implementation.”