The `intrinsic' heterogeneity of breast tissue, reflected in texture and spatial composition on the mammogram, may provide independent but complementary information to breast density for the assessment of both risk of breast cancer (BC) and masking that can lead to a missed BC on screening mammography. This may be especially important for the 40-50% of women with dense breasts who need improved risk stratification. We have developed automated methods to quantitatively measure parenchymal complexity features from full field digital mammograms (FFDM) using an innovative lattice-based approach to comprehensively characterize parenchymal tissue heterogeneity on the mammogram. Using unsupervised clustering applied to features measured from 2000 screen-FFDM, we found evidence for four reproducible `intrinsic' parenchymal complexity phenotypes that independently contributed to BC risk, accounting for breast density. In this proposal, we will expand this set of parenchymal features, classify and validate parenchymal phenotypes generalizable to multiple racial/ethnic groups, and examine their association with BC risk and masking. In AIM1, we will characterize and validate parenchymal complexity phenotypes reflecting the `intrinsic' heterogeneity of the breast parenchyma. We will use established automated algorithms to measure features representing statistical and structural properties of parenchymal heterogeneity on 36,000 screen-FFDM sampled from three large multi-ethnic mammography cohorts. We will use hierarchical clustering methods, and a split-sample approach, to first classify, and then independently validate a robust set of distinct parenchymal phenotypes among all breast density categories and specifically for dense breasts. In AIM 2, we will examine the association of parenchymal complexity phenotypes with risk for invasive BC. We will measure these parenchymal features on screen-FFDM performed within five years prior to diagnosis from 3817 incident invasive cancer cases and 7634 matched controls, and classify them into the parenchymal phenotypes from Aim 1. We will examine their association with BC (both across all levels of density and dense breasts only) adjusting for established risk factors and breast density. Finally, in AIM 3, we will examine the contribution of parenchymal complexity phenotypes to masking invasive BC. We will examine whether parenchymal phenotypes are associated with interval vs. screen-detected cancers, compared to true-negative controls, using the case-control study in AIM 2. AIMS 1 and 2 will also be tested within a subset of women with available digital breast tomosynthesis (DBT) exams (N=300 invasive BC), to inform the translation of our results to the emerging DBT technology. Our proposal capitalizes on experienced investigators, productive collaborations, novel algorithms, and established, well-characterized cohorts and will elucidate novel parenchymal phenotypes that can improve our ability to define subsets of women at differential BC risk and increased risk of missed BC. Our study will ultimately pave the way for more effective, tailored BC screening and prevention approaches.

In addition to the overall quantity of breast density, patterns and texture features on the mammogram show promise for discrimination of future risk and breast cancer detection. Using the largest and most diverse set of digital screening mammograms to date, we will define a set of parenchymal complexity phenotypes, representing `intrinsic' heterogeneity of breast tissue, that may better discriminate invasive breast cancer risk and who is at risk for interval cancers. We will also begin to examine these radiomic phenotypes in the emerging digital tomosynthesis setting. Our findings are especially important for women with dense breasts who comprise 40-50% of women undergoing breast screening, where refining the risk using additional parenchymal features could result in tailored screening, thereby reducing unnecessary imaging and health care costs.