AI Insights: Calculating Risk
By Keith Loria
Radiology Today
Vol. 26 No. 3 P. 8

Heart disease in women remains significantly under-recognized and is still the primary cause of death. While breast arterial calcification (BAC) on mammograms has emerged as a biomarker of women’s cardiovascular disease (CVD) risk, there has been a lack of quantification tools and clinical outcomes studies analyzing this.

That was the catalyst for a group of researchers to come up with the idea for opportunistic screening. The study, “Automated Breast Arterial Calcification Score Is Associated With Cardiovascular Outcomes and Mortality,” published in the Journal of the American College of Cardiology, assessed the association of BAC (both presence and quantity) with CVD outcomes.

“One of the challenges is how we can better identify women early who may be at risk of heart disease and possibly implement preventive strategies,” says Nitesh Nerlekar, MBBS, FRACP, MPH, PhD, one of the authors of the study. “An ideal way to do this is to try and leverage an established workflow that women already engage with, such as screening mammography.”

Lori B. Daniels, MD, FACC, coauthor of the study and a board-certified cardiologist at the University of California, San Diego Health, notes this concept is nothing new and BAC has been studied over the decades but, historically, it’s been a very laborintensive process; the radiologist has to figure out which calcium is in the arteries, which is in the fat, and make an assessment on how much is there.

“More recently, there was a study where there was a semiquantitative methodology, where it was still a radiologist sitting there, dividing the breast into quadrants on both sides and then scoring each quadrant on a zero, mild, moderate, severe scale, and then adding up all the quadrants for a score,” she says. “But you can imagine that takes a very long time, and radiologists are reading hundreds and hundreds of mammograms.”

Artificial Assistance
When CureMetrix, a San Diegobased company, developed a way to use automated, AI-based technology to read these mammograms, distinguish what’s in the arteries from what’s not, and come up with a quantitative measurement, it allowed the study authors to look at mammograms going back years and see what happened to these women.

The retrospective, single-center study focused on women who underwent screening mammograms between 2007 and 2016, utilizing nine years’ worth of data from approximately 18,000 participants. BAC was quantified through an AI-generated score, which was analyzed as both a binary and continuous variable. Regression analyses investigated the relationship between BAC and outcomes, including mortality, as well as a composite of acute myocardial infarction, heart failure, stroke, and mortality. The analyses were adjusted for factors such as age, race, diabetes, smoking status, blood pressure, cholesterol levels, and a history of CVD and chronic kidney disease.

“The concept of BAC detection from a mammogram is so appealing because it is a representation of vascular calcification which you could extrapolate. It represents a ‘memory’ of exposure to vascular risk factors; BAC can identify individuals with CVD risk factors,” Nerlekar says. “The appeal is that it is personalized to that individual, and it can be readily incorporated into an already established screening program without the need for any additional testing or radiation exposure.”

According to the study, “BAC was quantified using a validated, proprietary investigational software (cmAngio, CureMetrix) based on a deep neural, AI network, and previously trained with an 80:20 split using over 34,000 2D full-field digital mammograms and digital breast tomosynthesis mammograms obtained from multiple sites across 13 health care facilities in Australia, Brazil, and the United States.”

During development, two of 11 Mammography Quality Standards Act-certified radiologists reviewed each case. The performance of the software for detecting BAC, as assessed by area under the receiver operating characteristic curve, was 0.98, with a sensitivity of 94% and specificity of 96%.

Nerlekar explains that, as with any new biomarker, BAC needs validation to be generalizable across different sites, ethnicities, and clinical groups. Tantamount to that is the need for education about BAC and its implications to all the key stakeholders, including patients and general and specialist medical practitioners.

“With adequate awareness, it can be used as a tool to augment clinical practice,” Nerlekar says. “The CSBI have actually made BAC reporting part of their latest guidelines, and I am sure further bodies will start to follow suit as further research emerges. We are in the process of developing additional workflows to assist radiologists and avoid detracting them from the core business of mammography—cancer detection.”

Small Steps
While BAC has been discussed for more than three decades, understanding and awareness about it are still nascent, and it remains significantly under-reported.

“There is excellent qualitative research that indicates women want to know if they have BAC present, and a large majority would be willing to undergo additional cardiovascular testing,” Daniels says. “However, the determination of exactly how it should be presented to ensure the most favorable outcome is still to be determined and will hopefully become apparent as we progress with our prospective trials.”

Based on the study’s findings, there are two levels to consider when recommending follow-up or management strategies for patients identified with significant levels of BAC.

“Firstly, the presence of BAC does not immediately suggest there is prevalent CVD; rather, it is a highlight that there is potential chronic disease because BAC is also associated with other conditions, such as chronic kidney disease, and it is also associated with the major CVD risk factors of diabetes, hypertension and hyperlipidemia,” Nerlekar says. “You can have these risk factors before you have developed coronary atherosclerosis. Therefore, in my mind, the first step is to screen them for CVD risk factors. Thereafter, you may unmask the need for additional testing.”

The danger in creating an absolute standardized approach, he notes, is that there’s still a need for further prospective data. “We are already in discussion with several large centers with diverse populations who are keen to collaborate and explore the breadth of BAC across all demographic strata,” Nerlekar says.

“We are also in preparation to roll out a prospective study to assess the effects of implementing BAC in the clinical arena and the downstream effects of such a strategy. This will be performed in a multicenter international collaboration. We hope any interested centers will be willing to join in on this journey.”

The study’s implications extend beyond individual risk assessment. Nerlekar outlined a vision for integrating BAC assessment into standard practice, allowing health care providers to better communicate risks and develop tailored management strategies for patients. Recommendations include increased monitoring or intervention for those with significant BAC findings.

“I believe the first thing needed is to understand if women will be willing to engage in seeking CVD assessment if they have BAC present,” Nerlekar says. “Initial qualitative surveys indicate yes, but we need to see this in practice; otherwise, there needs to be more efforts placed on educational awareness of this marker, thereafter understanding the modifiability of risk based on presence of BAC. This is essential because, again, if there is no ability to alter risk, then it is not as useful as a risk arbiter.”

— Keith Loria is a freelance writer based in Oakton, Virginia. He is a frequent contributor to Radiology Today.