Quantitative MR Imaging Markers For Breast Cancer Risk Prediction

Abstract

Breast cancer risk assessment can help early detection and intervention planning, lower mortality, and improve care outcomes. Many current tools utilize risk factors such as age, menopausal status, personal biopsy history, and family history of breast cancer. Although these characteristics are associated with risk, the tools incorporate minimal information from imaging studies such as mammograms or MRIs. Even when negative for disease, analysis of these images could provide rich information on breast tissue biology that might also capture risk. There remains much unexplored potential in leveraging imaging markers as part of risk assessment models for prevention and early prediction. In this work, we aim to develop and assess a new risk prediction model that incorporates MRI-derived features along with clinical risk factors. In Part 1, I collected relevant clinical risk factors for predicting cancer within 5 years from the MRI for a cohort of elevated risk patients. I built a REDcap data repository to store their clinical risk factors obtained from electronic health records, such as select demographic information, menopausal status, BRCA mutation status, and family history of breast cancer. In Part 2, I identified predictive quantitative breast MRI markers. I extracted quantitative markers of background parenchymal enhancement using an image processing pipeline, which I helped develop, and assessed their ability to predict 5-year cancer risk. In Part 3, I performed multimodal combination of imaging markers and clinical risk factors to predict 5-year cancer risk. I also explored the performance of these combinations for several cohort subsets. My findings showed that multimodal combination with imaging markers can improve the predictive performance of a conventional risk assessment tool.