Understanding the biological basis for the association between parenchymal texture features and breast cancer risk

Breast composition is a potential breast biomarker, but its utility has been limited by measurement methods. Visually-assessed qualitative scales capture within-breast heterogeneity but are subjective and lack reproducibility. In contrast, quantitative automated assessments of global breast density are reproducible, but contain no information about within-breast variation. Limitations of both of these approaches can be overcome with the measurement of parenchymal texture features. Texture features are quantitative measures that estimate complex characteristics of pixel density in the breast image, ranging from descriptive statistics to higher order statistics that describe spatial relationships and structural patterns. Prior studies have shown that texture features independently predict breast cancer risk. However, little is known about the biological mechanisms driving that risk relationship. The objective of this study is to identify the biological processes associated with parenchymal texture features. The rationale is that direct evidence that texture features reflect specific biological properties will provide the basis for development of texture features as a dynamic marker of breast cancer risk and prognosis. This study will pursue three aims. Using a case-control analysis, Aim 1 will identify the texture features that are independently associated with newly-diagnosed breast cancer among women attending breast cancer screening. Aim 2 will evaluate how the texture features that were associated with breast cancer in this population vary with estrogen levels, through (i) cross-sectional analysis of texture features and 15 urinary estrogens and estrogen metabolites, and (ii) analyses of longitudinal change in texture features among breast cancer patients treated with anti-estrogenic therapy. Aim 3 will evaluate associations between texture features and breast histologic characteristics (tissue composition, benign breast disease/LCIS, measures of lobular involution) among women with a benign biopsy. Analyses will draw on existing mammograms, biopsy specimens, and electronic health records from women participating in mammography at the University of North Carolina; urine will be collected prospectively. Texture features will be measured using a novel lattice-based grid method developed and validated by members of the study team that allows information from the whole breast to inform the texture measurements. These analyses will establish: the magnitude of the relationship between lattice-based texture features and breast cancer in a general screening population (Aim 1); the extent to which texture features may act as biosensors of breast estrogen/anti-estrogen activity (Aim 2); and whether texture features can serve as a radiologic surrogate of histologic characteristics that have known associations with breast cancer risk (Aim 3). These results will clarify the potential role of parenchymal texture features as predictors of breast cancer risk and therapeutic response; such new uses have the potential to identify new prevention targets and reduce unnecessary procedures and treatments for women at risk for and being treated for breast cancer.