Quantitative Parenchyma Descriptor as an Imaging Biomarker of Breast Cancer Risk

点击翻译按钮获取中文摘要

DESCRIPTION (provided by applicant): Quantitative Parenchyma Descriptor as an Imaging Biomarker of Breast Cancer Risk Project Summary/Abstract Breast cancer remains one of the leading causes of death among women at the age of 40 and older. Mammography has been used as a low-cost screening tool for breast cancer. The recent controversy on breast cancer screening recommendations has increased public awareness and interests for informed counseling of screening and health care options based on individualized estimates of risk. The goal of this proposed project is to develop a computerized image-based biomarker to assess the breast cancer risk of individual patients in the screening population. The innovation of our approach lies in the fact that the quantitative breast parenchyma descriptor (q-BPD) will be designed to take into account not only the percentage of dense tissue (PD) but also the stromal and epithelial structural pattern of an individual's breast that is complementary to, rather than a surrogate of, the breast density. The q-BPD is obtained by a joint analysis of the complexity of the parenchymal distribution pattern (mammographic parenchymal pattern, MPP) and the amount of dense tissue (PD) as they are imaged on full-field digital mammograms (FFDMs). We hypothesize that the proposed q-BPD is an independent risk factor for breast cancer and will have a stronger predictive power than previous approaches such as PD or BI-RADS density categories alone. To test the hypothesis, we have the following specific aims: (1) to collect a matched case-control data set of 500 breast cancer cases and 2000 matched controls with 5 years of prior FFDMs (prior to cancer diagnosis for the case group). We will split the entire data set into independent subsets for training and validation; (2) to design a q-BPD by using advanced machine learning and computer vision techniques to maximize the discriminatory power at the personal level; (3) to investigate the association of developed q-BPD with breast cancer risk in comparison with commonly used density descriptors, such as radiologist's estimates of BI-RADS density categories and interactive PD on FFDMs by case-control studies and statistical analyses, taking into account other confounding risk factors. When fully developed, the automated q-BPD can be incorporated as a part of routine breast cancer screening. It will not only be useful for breast cancer risk prediction for individual patients but also for monitoring of risk regression or progression over time due to treatment or other factors. The new risk prediction tool is expected to play a key role in personalized breast cancer screening for women at different risk levels, thereby reducing health care costs while benefiting high risk women. The success of this project will lay the foundation for future large-scale clinical trials to address the limitations and investigate the clinical utilities of the proposed q-BPD for breast cancer risk prediction.