Researchers in Australia at the independent global medical research institute, The George Institute for Global Health, in collaboration with the University of New South Wales and the University of Sydney, have developed a machine learning model that can analyze mammography images to successfully predict the risk of cardiovascular events in women.
Cardiovascular diseases, such as coronary artery disease, arrhythmia, and heart valve disease, and including cardiovascular events such as heart attacks and strokes, are the leading cause of death worldwide, resulting in approximately 20 million deaths each year.
Every year, around nine million women die of cardiovascular disease, but despite these high numbers, several international studies have shown that the symptoms of cardiovascular disease and cardiovascular events, as well as their risk factors, are overlooked far more often in women than in men.
For example, a 2024 study showed that women who were hospitalized with a heart attack were less likely to receive the necessary treatment and more likely to die than men.
For this reason the researchers at The George Institute for Global Health wanted to find a way to use existing data to predict the risk for cardiovascular events in women.
“It’s a common misconception that cardiovascular disease predominantly affects men, resulting in underdiagnosis and undertreatment of the condition in women,” explained co-author Clare Arnott, PhD, associate professor and global director of the cardiovascular program at The George Institute.
“By integrating cardiovascular risk screening with breast screening through the use of mammograms—something many women already engage with at a stage in life when their cardiovascular risk increases—we can identify and potentially prevent two major causes of illness and death at the same time.”
In their study, published in the journal Heart and titled “Predicting cardiovascular events from routine mammograms using machine learning,” they developed a fully automated deep learning algorithm that can analyze whole breast architecture and characteristics to predict cardiovascular risk in women who undergo routine mammography screening for breast cancer.
While the idea to use mammogram images to understand this risk is not new, studies to date have focused merely on few features in mammographic images, such as breast arterial calcification. However, this has limitations, as the risk for cardiovascular events can come from many factors. Breast arterial calcification, for example, cannot be applied as accurately to older women.
In their study, the researchers looked at mammography data from 49,196 women aged 35-94 years, with a mean age of around 60 years, and a median follow-up of 8.8 years. Of these women, 3,392 reported during their follow-up that they had experienced a first major cardiovascular event, such as atherosclerosis, heart failure, heart attack, or stroke.
The researchers trained their automated algorithm to analyze the full range of internal breast structures and characteristics from these routine mammogram images, taking into account the women’s age to predict their major cardiovascular disease risk over ten years.
They then compared their AI model with other risk scores and calculators, which need multiple data points based on known cardiovascular risk factors, including blood pressure and cholesterol.
“We found that our model performed just as well without the need for extensive clinical and medical data,” said Arnott. “Our model is the first to use a range of features from mammographic images combined simply with age—a key advantage of this approach being that it doesn’t require additional history taking or medical record data, making it less resource intensive to implement, but still highly accurate.”
As a next step, the researchers aim to validate their algorithm in more diverse patient populations, utilizing different screening practices to assess the generalizability of their AI model and to refine it further.
