Abstract:  Machine learning and artificial intelligence are driving forces in medical research, including radiology. These technologies heavily rely on data, often millions of meticulously curated medical images, for deep neural network development and optimization. Our lab focuses on innovative ways to maximize existing data for deep algorithm performance. In this seminar, we'll discuss recent studies from our lab. The first part explores using breast laterality features to improve breast cancer detection in mammograms through the Radon Cumulative Distribution Transform (RCDT), which can highlight subtle bilateral tissue differences in breast mammograms. We will then discuss how simulated mammograms via Generative Adversarial Networks (GANs) could improve the performance of cancer detection algorithms. I will conclude the seminar with cautions on when to use GANs for simulating mammograms, as this can significantly affect the performance of the detection algorithm.

This seminar is hosted by Purdue’s Institute for Cancer Research and the Weldon School of Biomedical Engineering.