AI is increasingly being used to help determine breast cancer risk, with two studies showing its value for primary prevention and in women of diverse ancestry.
The first study, in Science Translational Medicine, revealed that AI could outperform standard clinical risk tools at identifying women of European ancestry who were at greatest risk of developing breast cancer within the next decade.
Because current AI-based models mostly predict breast cancer risk in the short term, Mikael Eriksson, PhD, from the Karolinska Institute, and fellow researchers created a model for invasive and in situ breast cancer that covered a 10-year period.
“Considering that a tumor can take five to 20 years to develop into a screen- or clinically detected cancer, a 10-year or lifetime risk projection time is reasonable,” they noted.
The researchers validated their image-derived AI-based 10-year risk model using digital mammograms from 8676 women in two cohorts from the U.S. and Sweden, of whom 1633 had breast cancer.
They compared their model with three other risk assessment tools: the Tyrer-Cuzick-v8, which uses personal and family history to determine the lifetime risk of breast cancer; the Breast Cancer Surveillance Consortium v3 tool, which estimates the risk of developing invasive risk cancer over five years; and the Mirai AI algorithm for predicting breast cancer risk.
The model calculated that the 10-year risk of breast cancer was 3.83% for the North-American group and 3.14% for the Scandinavian group, and it showed promise for predicting the risk of invasive tumors.
The AI model performed significantly better at predicting invasive breast cancer than the three comparator models both overall, in women aged 50 to 69 years and those with estrogen-receptor-positive breast cancers.
“An image-derived AI-based risk model developed for 10-year risk assessment for identifying individuals in mammographic screening who may benefit from primary prevention strategies identifies up to 40% of breast cancers to be at high risk at study entry in U.S. and Swedish validation case cohorts per clinical guidelines,” summarized Eriksson and co-workers.
The second study, in Science Advances, revealed how AI could be of value for women from more diverse backgrounds, and that it had value regardless of race and ethnicity.
The research was driven by the observation that conventional prediction models incorporating genetic and clinical factors including breast density underperform in women who do not have European ancestry.
Shu Jiang, PhD, from Washington University School of Medicine in St Louis, and co-workers set out to evaluate the generalizability of an AI-derived mammogram risk score (MRS), a summary of texture features which captures intrinsic breast-tissue characteristics that are the basis for cancer to develop.
To do this, they used data from two large North American cohorts that represented more than 226,000 racially diverse women undergoing routine breast screening with mammography.
This was then used to examine the generalizability of MRS with breast cancer risk across non-Hispanic white (NHW), non-Hispanic Black (NHB), East Asian, South Asian, and Indigenous women.
Jiang and team found that the MRS was a strong predictor of breast cancer risk independent of race or ethnicity, demonstrating its potential for broader clinical utility.
Across the two independent screening cohorts, the MRS increased with age in line with rising breast cancer risk. It was a strong and consistent predictor of breast cancer risk across race and ethnic groups as shown by the association study and comparison of distributions and demonstrated excellent calibration in all groups.
Its remained robust across full-field digital mammograms or synthetic two-dimensional digital breast tomosynthesis, although the limited numbers of images precluded an analysis of how performance varied across manufacturers.
“Overall, these results support that MRS is a powerful breast cancer risk predictor that does not depend on race or ethnicity, supporting its potential for broader clinical adoption and use in varied populations of women undergoing routine screening mammography to identify those at increased risk,” Jiang and coworkers concluded.
The post AI Shows Value for Breast Cancer Prediction appeared first on Inside Precision Medicine.