A study analyzing more than 30,000 digital mammograms found that breast tissue patterns—beyond density—can help predict a woman’s risk of breast cancer, including cases that go undetected during routine screening.
Researchers identified distinct patterns on mammograms, known as parenchymal phenotypes, using advanced radiomic techniques. These patterns reflect the internal structure and complexity of breast tissue. Unlike breast density, which provides a general measure, these phenotypes capture detailed features such as texture and spatial distribution, according to Stacey J. Winham, PhD, and colleagues.
The study used a two-stage design. In the first stage, mammograms from more than 30,000 women at 3 screening centers were analyzed to define and validate parenchymal phenotypes using 390 radiomic features. In the second stage, the findings were applied to a case-control sample of 1,055 women who developed invasive breast cancer and 2,764 matched controls, noted Winham, of the Department of Quantitative Health Sciences at the Mayo Clinic in Rochester, MN, and colleagues.
Researchers identified six phenotype clusters and six principal components (PCs). Both were associated with an increased risk of invasive breast cancer after adjusting for age, body mass index, and breast density.
PC1 and PC2 were linked to higher cancer risk, while PC3 was associated with lower risk. These associations were consistent across both Black and White women, with slightly stronger predictive performance observed in Black women.
The study also examined false-negative findings—cancers missed during screening—and interval cancers, which develop between screenings. PCs improved the detection of both types compared with current methods.
Adding PCs to existing models increased the accuracy of identifying false-negative findings from 66% to 73%. For symptomatic interval cancers, accuracy rose from 68% to 77%.
“These imaging phenotypes may improve breast cancer prediction for Black women,” according to investigators. Breast density–based models may be less accurate in this population.
Even women with nondense breasts—typically considered lower risk—had elevated cancer risk if their parenchymal phenotypes showed certain patterns. PC3 was strongly protective in nondense breasts but not in dense breasts.
"Beyond breast density, more information linked to the genotypic and phenotypic characteristics of tissues is embedded within the radiologic images and is invisible to the eye of the radiologist but can be extracted using advanced texture and shape analysis," wrote Benoit Mesurolle, MD, and Mona EL Khoury, MD, in an accompanying editorial.
All mammograms analyzed were taken before cancer diagnosis, allowing researchers to assess how imaging features relate to future cancer development. An automated pipeline was used to ensure consistent radiomic analysis across study sites.
This study is among the largest to date evaluating mammographic texture features for breast cancer risk prediction and detection of missed cancers. The authors suggest that integrating these imaging biomarkers into clinical models may enhance personalized screening strategies.
Full disclosures can be found in the published study.
Source: Radiology
