Title: Safe Breast Cancer Diagnosis Resilient to Mammographic Adversarial Samples

Speaker: Degan Hao

Abstract: Adversarial data can lead to malfunction of deep learning applications. It is essential to develop deep learning models that are resilient to adversarial data while accurate on standard, clean data. In this study, we focus on building safe breast cancer diagnosis models against mammographic adversarial samples. We proposed a novel adversarially robust feature learning (ARFL) method to facilitate adversarial training using both standard data and adversarial data, where a feature correlation measure is incorporated as an objective function to encourage learning of robust features and restrain spurious features. To show the efficacy of ARFL for robust breast cancer diagnosis, we built and evaluated deep learning diagnosis models using two independent clinically collected breast imaging datasets, comprising a total of 9,548 mammogram images. We performed extensive experiments showing that the ARFL method outperformed several state-of-the-art methods. ARFL can serve as an effective method to enhance adversarial training, towards building safe breast cancer diagnosis against adversarial attacks in clinical set- tings. The code repository of this study is publicly available at GitHub: https://github.com/usernamesafeai/ARFL.

Bio: Degan Hao is a Ph.D. candidate in the Intelligent Systems Program at the School of Computing and Information and a machine learning research scientist at the School of Health and Rehabilitation Sciences, University of Pittsburgh. Degan’s research focuses on developing trustworthy AI solutions for medical imaging and electronic health records, utilizing advanced techniques in generative AI, deep learning, computer vision, and natural language understanding to tackle critical challenges in healthcare. His work has been published in leading conferences and journals, including the Journal of Biomedical and Health Informatics, Artificial Intelligence in Medicine, and Frontiers in Oncology. Degan also serves as a reviewer for top AI conferences, such as MICCAI and AAAI, and holds patents from his internship at Genentech.

Title: Knowledge-guided multi-task learning for breast cancer diagnosis using longitudinal mammogram images

Speaker: Zhengbo Zhou

Abstract: In clinical detection and diagnosis of breast cancer, tempo- ral analysis of mammogram images plays a crucial role. This study introduces a knowledge-guided multi-task learning approach that aims to elevate the accuracy of breast cancer diagnosis by incorporating the analysis of breast density changes over time. While breast cancer diagnosis is our main task, we construct the classification of breast density evolution over time as an auxiliary task that leverages crucial longitudinal tissue changes to enhance breast cancer diagnosis. Specifically, we propose a novel architecture that employs cross- view mechanisms and context-guided triplet loss to capture temporal changes and facilitate more effective learning. Our contributions include a customized multi-task model that integrates breast density categorization to capture temporal feature changes, and the application of context-guided triplet loss for faster convergence. We performed experiments on a co- hort of 580 breast cancer patients (each has two sequential mammogram exams) in a case-control setting, and the pro- posed method achieved an AUC of 0.745, outperforming the several compared methods.