AI-powered mammograms: A new window into heart health

The U.S. Centers for Disease Control and Prevention recommends that middle-aged and older women get a mammogram — an X-ray of the breast — to screen for breast cancer every one or two years. About 40 million mammograms are performed in the United States each year. While breast artery calcifications can be seen on the resulting images, radiologists do not typically quantify or report this information to women or their clinicians. The new study, which used an AI image analysis technique not previously used on mammograms, demonstrates how AI can help fill this gap by automatically analyzing breast arterial calcification and translating the results into a cardiovascular risk score.

“We see an opportunity for women to get screened for cancer and also additionally get a cardiovascular screen from their mammograms,” said Theo Dapamede, MD, PhD, a postdoctoral fellow at Emory University in Atlanta and the study’s lead author. “Our study showed that breast arterial calcification is a good predictor for cardiovascular disease, especially in patients younger than age 60. If we are able to screen and identify these patients early, we can refer them to a cardiologist for further risk assessment.”

Heart disease is the leading cause of death in the United States but remains underdiagnosed in women and there is also lagging awareness. Researchers said the use of AI-enabled mammogram screening tools could help identify more women with early signs of cardiovascular disease by taking better advantage of screening tests that many women routinely receive.

A buildup of calcium in blood vessels is a sign of cardiovascular damage associated with early-stage heart disease or aging. Previous studies have shown that women with calcium buildup in the arteries face a 51% higher risk of heart disease and stroke.

To develop the screening tool used for this study, researchers trained a deep-learning AI model to segment calcified vessels in mammogram images — which appear as bright pixels on X-rays — and calculate the future risk of cardiovascular events based on data obtained from the electronic health record data. The segmentation approach is what separates this model from previous AI models developed for analyzing breast artery calcifications. Researchers said the model is also strengthened by its use of a large dataset for training and testing, which included images and health records from over 56,000 patients who had a mammogram at Emory Healthcare between 2013 and 2020 and had at least five years of follow-up electronic health records data.

“Advances in deep learning and AI have made it much more feasible to extract and use more information from images to inform opportunistic screening,” Dapamede said.

Overall findings showed the new model performed well at characterizing patients’ cardiovascular risk as low, moderate or severe based on mammogram images. After calculating the risk of dying from any cause or suffering an acute heart attack, stroke or heart failure at two years and five years, the model showed that the rate of these serious cardiovascular events increased with breast arterial calcification level in two of the three age categories assessed — women younger than age 60 and age 60-80, but not in those over age 80. This makes the tool particularly well suited for providing early warning of heart disease risk in younger women, who can benefit more from early interventions, researchers said.

The results also showed that women with the highest level of breast arterial calcification (above 40 mm²) had a significantly lower five-year rate of event-free survival than those with the lowest level (below 10 mm²). For example, 86.4% of those with the highest breast arterial calcification survived for five years compared with 95.3% of those with the lowest level of calcification. This translates to approximately 2.8 times the risk of death within five years in patients with severe breast arterial calcification compared to those with little to no breast arterial calcification.

The AI model was developed as a collaboration between Emory Healthcare and Mayo Clinic and is not currently available for use. If it passes external validation and gains approval from the U.S. Food and Drug Administration, researchers said the tool could be made commercially available for other health care systems to incorporate into routine mammogram processing and follow-up care. The researchers also plan to explore how similar AI models could be used for assessing biomarkers for other conditions, such as peripheral artery disease and kidney disease, that might be extracted from mammograms.