Grant Number:	1R01CA305931-01A1 Interpret this number
Primary Investigator:	Vachon, Celine
Organization:	Mayo Clinic Rochester
Project Title:	Radiomic Phenotypes for Invasive and Advanced Breast Cancer
Fiscal Year:	2026

Mammography-based risk factors hold promise for broad use in breast cancer (BC) risk prediction, given >75% of women have routine screening mammography in the U.S. Mammographic breast density, the proportion of fibroglandular breast tissue, is the most established mammography-based BC risk factor. Recently, high throughput extracted radiologic imaging features (known as radiomics), reflecting the intrinsic heterogeneity and complexity of fibroglandular tissue structure, have been shown to improve risk prediction. We recently identified and validated six reproducible “intrinsic patterns” or radiomic phenotypes based on features extracted from digital mammograms (2D DM) and found them associated with future risk of invasive BC and interval invasive BC (occurring after a negative mammogram and before the next screen) in Black and White women, independent of breast density. As breast screening in the US has rapidly transitioned from 2D DM to 3D digital breast tomosynthesis (DBT), which offers superior tissue visualization, there is potential to extract more accurate radiomic features to improve BC prediction. Given known differences in breast density and BC prognosis by race, it is imperative to define radiomic phenotypes across representative racial and ethnic groups and examine their associations with advanced BC (pathologic prognostic stage II or higher), which is a strong surrogate for BC mortality. Our goal is to extract radiomic features from screening-DBT exams; characterize and validate radiomic phenotypes from these features for all women and among racial and ethnic groups; and examine their association with incident invasive and advanced BC risk. SA 1 will characterize and validate radiomic phenotypes on 36,000 screening-DBT exams among a representative sample of US women, ages 40-74, sampled from four breast screening cohorts. We will extract over 2,000 radiomic features, classify, and independently validate radiomic phenotypes, for all women and within racial and ethnic groups. SA 2 will examine the association of radiomic phenotypes (from SA 1) with incident invasive and advanced BC risk among a representative sample of US women within a nested case-control study of 8,500 invasive BC and 17,000 matched controls. We will assess radiomic features on the earliest screening- DBT performed within five years prior to diagnosis; classify them into the validated radiomic phenotypes; and examine the phenotype association with future invasive and advanced BC risk by race and ethnicity, breast density and body mass index (SA 2.1). We will also examine differential associations of radiomic phenotypes with tumor characteristics and with polygenic risk scores to inform underlying etiologic mechanisms (SA 2.2). SA 3 will evaluate the contribution of the radiomic phenotypes to clinical BC risk models and FDA approved artificial intelligence BC algorithms for risk prediction of invasive and advanced BC. Elucidating and characterizing novel radiomic phenotypes from screening-DBT relevant to all US women will improve our ability to define groups of women at differential BC risk, for personalized screening and prevention strategies.





None