Cancer Screening and Prevention

Identifying people who are at high-risk of cancer is critical for implementing personalized cancer screening protocols and forming preventive strategies. We are developing algorithms to analyze imaging data and integrate non-imaging information (genetics, blood, EHR) to more comprehensively predict who is at high risk for developing cancer. We pioneered the investigation of imaging biomarkers for breast cancer risk using digital breast tomosynthesis (DBT) and emerging 3D X-ray imaging modality for breast cancer screening. 

We are leveraging deep learning to quantitatively estimate percent breast density, a known risk factor for breast cancer. We are also working on developing radiomics-based and artificial intelligence approaches to quantitatively characterize the overall breast parenchymal pattern, shown to provide additional information about breast cancer risk to that of breast density alone. Having more accurate, quantitative breast density measures can improve individualized risk assessment to implement personalized and/or supplemental screening and prevention strategies. We have also been working on breast MR imaging, developing novel quantitative methods for characterizing the functionally active parenchymal tissue in the breast, known as background parenchymal enhancement (BPE). This is increasingly shown to be an emerging, stronger biomarker of cancer risk, capable of also indicating changes in breast cancer risk due to the effect of risk reduction interventions. Our study was the first to propose a fully-automated method for estimating MRI-BPE, showing that this can be used to identify women with BRCA1/2 mutations likely to respond to risk-reducing Salpingo-Oophorectomy. 

We are also working on developing novel radiomic and deep learning methods to characterize lung parenchymal phenotypes from screening low-dose computed tomography (LDCT) as biomarkers of lung cancer risk that could ultimately help better guide personalized screening recommendations for lung cancer. Our hypothesis is that these novel imaging markers can characterize the heterogeneity of the lung parenchyma, capturing conditions that may pre-dispose to lung cancer, such as inflammation, fibrosis, and obstruction of the airways and early/asymptomatic stages of diffuse lung diseases.