Speaker: Mirabela Rusu, PhD, Department of Radiology, Stanford Medicine
Date: Monday, November 13th, 2023
Time: 11:00AM Central Time
Location: Webex
Title: “Multimodal Artificial Intelligence Methods to Improve the Interpretation of Radiology Images”
Abstract: Radiology imaging is an essential non-invasive tool for cancer care, facilitating diagnosis and treatment planning. Yet, despite technological advancements in imaging, radiologists still miss 12-36% of aggressive prostate cancers due to their subtle radiologic features. Some patients undergo surgery as cancer treatment and thus have rich data comprised of high-resolution pathology images showing the extent, heterogeneity, and aggressiveness of cancer. In this context, my team develops advanced deep learning methods to assist radiologists in their image interpretation. Our approaches are split into three categories. First, we developed registration methods that align radiology and pathology images, e.g., in the prostate or breast, using both traditional and deep learning-based registration methods. Second, we developed radiology-pathology fusion strategies by constructing pathology-derived radiology (rad-pathomic) biomarkers. When included in deep learning models, these biomarkers improved by 7% the detection of aggressive vs indolent cancers on prostate MRI or kidney CT. Third, we focus on using the ubiquitous but noisy b-mode ultrasound images to detect and localize prostate cancer, in approaches that often rely on MRI during the training of the models, but only use b-mode ultrasound images at inference time in new patients, rendering then useful to guide biopsy or local treatment procedures.
Bio: Dr. Mirabela Rusu received her MS and PhD in Computational Biomedicine from University of Texas, Houston, and focused her research on the fusion of biomolecular structural data from different sources (i.e., cryo-electron microscopy and X-ray crystallography). Her postdoctoral training at Rutgers and Case Western Reserve University was focused on developing computational methods for the fusion of medical images, i.e., to register radiology or pathology images, or create population atlases for prostate cancer studies. Following postdoctoral training, Dr. Rusu joined Industry as an Image Analysis Scientist/Lead Engineer. Currently, Dr. Rusu is an Assistant Professor of Radiology, and by courtesy, Urology and Biomedical Data Science, at Stanford University, where she leads the Personalized Integrative Medicine Laboratory (http://pimed.stanford.edu). Dr. Rusu’s team focuses on developing analytic methods to improve the interpretation of radiology images by taking advantage of existing high-resolution information during training but only needing lower resolution radiology images during inference (e.g., when applied in new patients).