- This event has passed.
Ph.D. Defense: Jun Bai
April 17, 2023 @ 12:00 pm - 1:00 pm EDT
Title: Applying Deep Learning in Biomedical Image Analysis
Ph.D. Candidate: Jun Bai
Major Advisor: Dr. Sheida Nabavi
Associate Advisors: Dr. Clifford Yang, Dr. Jinbo Bi, Dr. Caiwen Ding
Committee Members: Dr. Qian Yang, Dr. Dongjin Song
Date/Time: Monday, April 17th, 2023, 12:00 pm
Location: WebEx and In-person
In-person location: HBL 1102
Meeting link: https://uconn-cmr.webex.com/uconn-cmr/j.php?MTID=m80266b7d3aafcff89a45558a262e71d1
Meeting number: 2622 521 1520
Password: RKhiMMYf572
Abstract:
The impact of cancers and diseases on the general population has caused immense anxiety and concern. It is essential that we have in place accurate and early detection methods to effectively combat the burden of these illnesses. The advancements in X-ray imaging technology have been nothing short of transformative in the field of cancer and disease diagnostics, resulting in a significant improvement in survival rates. Despite this progress, the low prevalence of these conditions among the screened population and the complexity of X-rays poses a significant challenge for radiologists, increasing the risk of missed or incorrect diagnoses. However, the advent of deep learning has profoundly affected the interpretation of biomedical images and imaging diagnostics. This presents us with a unique opportunity to merge X-ray imaging with deep learning, further enhancing the validity of diagnoses and improving patient outcomes.
Despite these achievements, there are still obstacles to be conquered, such as a lack of data, high-resolution data, small tumors, and subtle abnormalities. To address these challenges, we must continue to develop innovative deep-learning models. In this presentation, I will discuss our research on novel deep learning models for disease diagnosis, focusing on three innovative areas: (1) a comprehensive deep learning framework to predict the probability of breast cancer using 2D mammograms with prior mammograms; (2) a graph-based, training-efficient model for predicting the probability of 3D mammogram breast cancer; and (3) un-supervised breast abnormal variation localization of using patients current and prior 2D mammograms.
