Developing Imaging Biomarkers for Breast Cancer by Leveraging Image Transformation and Deep Learning

Juhun Lee, PhD
Assistant Professor
Department of Radiology
University of Pittsburgh

Abstract:  Among many traits leading to possible breast cancer, radiologists focus on two aspects when they read screening mammogram of women. These two aspects are: 1) temporal changes from previous mammogram to current one, and 2) lateral tissue difference from left – right breast mammogram pair. In this seminar, I will introduce two machine learning approaches to analyze the ways in which radiologists read screening mammogram by focusing on temporal changes and lateral differences. The first approach is to highlight suspicious but subtle breast cancer signal by using Radon Cumulative Distribution Transform (RCDT). The second approach is to simulate normal and personalized mammograms using Conditional Generative Adversarial Network (CGAN). I will show how each approach is developed and how they are used as imaging biomarkers for identifying mammographically-occult breast cancer and estimating short-term breast cancer risk.

Bio:  Dr. Juhun Lee is an Assistant Professor in the Department of Radiology, School of Medicine at the University of Pittsburgh. He received his PhD in Electrical and Computer Engineering at The University of Texas at Austin in 2014. Dr. Lee joined the Imaging Research Laboratory of the Department of Radiology in 2014 as a Postdoctoral Associate. He was promoted to Research Instructor in 2017, and then to Assistant Professor in 2020. His research focuses on breast imaging with an emphasis on using quantitative information extracted from breast computed tomography (CT) images, mammograms, and digital breast tomosynthesis to detect and characterize breast lesions, estimate breast cancer risk, and optimize image reconstruction. He is currently Principal Investigator (PI) of NIH R37 grant on developing novel imaging biomarkers for estimating near-term breast cancer risk.