Deep Learning Image Analysis of Benign Breast Disease to Identify Subsequent Risk of Breast Cancer

Abstract

AbstractBackgroundNew biomarkers of risk may improve breast cancer risk prediction. We developed a computational pathology method to segment benign breast disease (BBD) whole slide images (WSIs) into epithelium, fibrous stroma, and fat. We applied our method to the BBD breast cancer nested case-control study within the Nurses’ Health Studies to assess whether computer-derived tissue composition or a morphometric signature was associated with subsequent risk of breast cancer.MethodsTissue segmentation and nuclei detection deep-learning networks were established and applied to 3795 WSIs from 293 cases who developed breast cancer and 1132 controls who did not. Percentages of each tissue region were calculated and 615 morphometric features were extracted. Elastic net regression was used to create a breast cancer morphometric signature. Associations between breast cancer risk factors and age-adjusted tissue composition among controls were assessed using analysis of covariance. Unconditional logistic regression, adjusting for the matching factors, BBD histological subtypes, parity, menopausal status, and BMI evaluated the relationship between tissue composition and breast cancer risk.ResultsAmong controls, BBD subtypes, parity, and number of births were differentially associated with all three tissue regions (p< 0.05); select regions were associated with childhood body size, BMI, age of menarche, and menopausal status (p< 0.05). Higher proportion of epithelial tissue was associated with increased breast cancer risk (OR = 1.39, 95% CI 0.91–2.14 comparing highest and lowest quartiles; p-trend< 0.05). No morphometric signature was associated with breast cancer.ConclusionThe amount of epithelial tissue may be incorporated into risk assessment models to improve breast cancer risk prediction.