ECE SEMINAR: Deep Learning Based Medical Image Analysis: Challenges and New Approaches
This event has passed.
Monday, May 9, 2022 - 12:00pm to 1:00pm
Danny Chen - Professor, Department of Computer Science and Engineering, University of Notre Dame
New technologies for acquiring large amounts of medical image data give rise to an ever increasing demand for effective approaches for medical image analysis tasks. Recently, deep learning (DL) methods have yielded remarkably high quality solutions for many medical imaging applications, largely outperforming traditional image analysis methods. Comparing to natural scene images, medical image analysis faces several different challenges. Commonly, DL methods rely on lots of annotated data for model training. While natural scene images are usually 2D, medical images can be 2D, 3D, and even higher dimensional. In particular, 3D medical images are widely used in basic research and clinical practice. 3D medical image analysis presents big challenges to deep learning methods. (1) 3D medical images are often of very large sizes (e.g., billions of voxels), and thus incur high computation costs. But, current GPUs are of limited memory for implementing 3D DL models. (2) Few efficient automatic techniques for annotating 3D images are available. Furthermore, generally only trained medical experts can annotate medical images effectively, hence making medical image annotation a highly costly and labor-intensive process (even for 2D images). Therefore, how to attain sufficient good quality annotated image data for DL model training while significantly reducing annotation effort for medical experts is a big bottleneck to the successful development and deployment of DL methods for medical imaging applications.
In this talk, we present new DL-based sparse annotation approaches for 3D medical image segmentation problems. Image segmentation aims to identify target objects in images, and is a central problem in medical image analysis. We show that it is often not enough to simply apply DL methods alone to tackle medical imaging problems. Our new approaches thus are based on combinations of DL methods and algorithmic techniques. For example, our new sparse annotation schemes judiciously select the most representative or valuable samples to annotate. Moreover, the problem of finding an optimal subset of samples to cover or represent an entire image dataset as sparse training data is an NP-hard problem which can be solved approximately with guaranteed good quality. Our approaches achieve strong performance with very small annotation ratios (e.g., 3%-15%). Notably, when utilizing full annotation (100%), our approaches outperform state-of-the-art full annotation segmentation methods on a number of 3D medical image segmentation tasks. We illustrate experimental results on various datasets to demonstrate the practical applicability of our new sparse annotation approaches for 3D medical image segmentation.
Danny Z. Chen received the B.S. degrees in Computer Science and in Mathematics from the University of San Francisco, California in 1985, and the M.S. and Ph.D. degrees in Computer Science from Purdue University, West Lafayette, Indiana in 1988 and 1992, respectively. He is a Professor in the Department of Computer Science and Engineering, the University of Notre Dame. His main research interests include computational biomedicine, biomedical imaging, computational geometry, algorithms and data structures, machine learning, data mining, and VLSI. He has worked extensively with biomedical researchers and practitioners, published many papers in these areas, and holds 7 US patents for technology development in biomedical applications. He received the NSF CAREER Award in 1996, a Laureate Award in the 2011 Computerworld Honors Program for developing “Arc-Modulated Radiation Therapy” (a new radiation cancer treatment approach), and the 2017 PNAS Cozzarelli Prize of the US National Academy of Sciences. He is a Fellow of IEEE and a Distinguished Scientist of ACM.