Talk: In Pursuit of Medical Foundation Models

LIVE STREAM: https://uwmadison.zoom.us/j/97442799109?pwd=T0g0RFg1NXBFRzNkamQ0enhBYzlJZz09

Abstract: Recent advancements in foundation models like GPT-4 have undoubtedly signified a transformative era in AI. However, their integration into medical imaging remains challenging due to the limited dataset access and domain complexities. This presentation will first review the evolution of deep-learning-based solutions, focusing on seminal models like U-Net and nnUNet, as well as their successful applications in medical imaging. Nonetheless, these specialized models often necessitate extensive retraining for new applications. Subsequently, I will introduce Transformer-based foundation models, exemplified by our pioneering TransUNet - the first Transformer-based model in medical imaging, as a solution not only for advancing medical vision representation but also for opening up possibilities for large medical AI models. Following this, I will discuss our recent efforts to further push the boundary by developing versatile medical AI models that offer unlimited flexibility across data, modalities and tasks, and can quickly adapt to new tasks with minimal training, Lastly, I will share insights into the safe deployment of these models in real-world learning systems, ensuring fairness, trustworthiness, and timely medical feedback across various medical settings, including hospitals, long-term care facilities, and home care environments.

Bio: Yuyin Zhou is an Assistant Professor of Computer Science and Engineering at the University of California, Santa Cruz, and a former Postdoctoral Fellow at Stanford University. She earned her Ph.D. in Computer Science from Johns Hopkins University under the supervision of Bloomberg Distinguished Professor Alan Yuille. Her contributions have been recognized with several accolades, including a finalist for the MICCAI Young Scientist Publication Impact Award in 2022 and the Best Paper Honorable Mention at DART 2022. Dr. Zhou is also actively involved in the broader scientific community, organizing over 15 workshops and tutorials for leading conferences such as ICML, MICCAI, ML4H, ICCV, CVPR, and ECCV, and her work has been featured in media outlets like ICCV Daily, Computer Vision News, the ``Best of CVPR'' series, and National Republic Radio (NPR). Additionally, she serves as a Senior Program Committee member for IJCAI 2021 and AAAI 2022, and as an Area Chair for CVPR 2024, ICLR 2024, MICCAI 2022-2024, and CHIL 2022-2024. She will serve as a DC chair for WACV 2025.