Abstract
Background Parenchymal phenotypes reflect the intrinsic heterogeneity of both tissue structure and distribution on mammograms. Purpose To define parenchymal phenotypes on the basis of radiomic texture features derived from full-field digital mammography (FFDM) in breast screening populations and assess associations of parenchymal phenotypes with future risk of breast cancer and masking (false-negative [FN] findings or interval cancers), beyond breast density, and by race and ethnicity Materials and Methods A two-stage study design included a retrospective cross-sectional study of 30 000 randomly selected women with four-view FFDM (mean age, 57.4 years) and a nested case-control study of 1055 women with invasive breast cancer (151 Black and 893 White women) matched to 2764 women without breast cancer (411 Black and 2345 White women) (mean age, 60.4 years) sampled from April 2008 to September 2019 from three diverse breast screening practices. Radiomic features (n = 390) were extracted and standardized using an automated pipeline and adjusted for age and practice. Variation was classified using hierarchical clustering and principal component (PC) analysis. The resulting clusters and PCs were examined for association with invasive breast cancer risk, FN findings on mammograms, and symptomatic interval cancers beyond radiologist-reported Breast Imaging Reporting and Data System (BI-RADS) breast density using conditional logistic regression and likelihood ratio tests. Discrimination for breast cancer was assessed with area under the receiver operating characteristic curve (AUC). Results Six clusters and six PCs were defined, replicated, and associated with a higher risk of invasive breast cancer (P = .01 and P < .001, respectively) after adjustment for age, body mass index (calculated as weight in kilograms divided by height in meters squared), and BI-RADS breast density. PCs showed similar associations among Black and White women (P = .23). PCs were also positively associated with FN findings (P = .004) and symptomatic interval cancers (P = .006). AUC improved for all breast cancer end points when incorporating PCs, with the greatest improvement shown in prediction of FN findings (AUC with vs without PCs, 0.73 [95% CI: 0.68, 0.78] vs 0.66 [95% CI: 0.61, 0.71] , respectively; P = .004) and symptomatic interval cancers (AUC with vs without PCs, 0.77 [95% CI: 0.71, 0.82] vs 0.68 [95% CI: 0.62, 0.74], respectively; P = .006). Conclusion Parenchymal phenotypes based on radiomic features extracted from FFDM were associated with a higher risk of invasive breast cancer, specifically for FN findings and symptomatic interval cancer. © RSNA, 2025 Supplemental material is available for this article. See also the editorial by Mesurolle and El Khoury in this issue.