When
Where
Title
Ideal Observer Computation in the Age of Machine Learning
Abstract
The Bayesian ideal observer (IO) is widely recognized as an indispensable tool in imaging and vision research. The IO performs optimally in statistical inference tasks and measures the total amount of task-relevant information in images. It is widely accepted that IO performance in clinically relevant tasks (e.g., tumor detection) should be used as a figure of merit for objectively assessing and optimizing medical imaging systems. The IO is also highly useful for predicting and explaining human performance in perceptual tasks (e.g., visual search). However, because the IO relies on complete knowledge of the statistics of image data, it cannot be determined analytically in most cases. Machine learning is profoundly impacting a wide range of research areas and has achieved significant success in performing image-based tasks, such as image generation and classification. In this talk, I will present machine learning methods for computing the IO for various tasks. Specifically, I will describe supervised learning-based methods that utilize artificial neural networks (ANNs) to approximate the IO for signal detection and localization tasks. I will also discuss a method combining generative adversarial networks (GANs) and Markov-chain Monte Carlo (MCMC) that can be readily employed to compute the IO for more complex estimation tasks. Finally, I will demonstrate the use of reinforcement learning to approximate the foveated IO, which generates optimal eye movement strategies for visual search tasks. Together, these machine learning methods represent a set of feasible computational tools for approximating the IO and hold great potential for evaluating emerging imaging technologies and addressing fundamental questions in vision science.
Bio
Weimin Zhou is an Assistant Professor with appointments in the Wyant College of Optical Sciences (OSC) and the Department of Medical Imaging (DMI) at the University of Arizona (UA), where he leads the Computational Imaging and Visual Intelligence Laboratory (CIVIL). He earned his Ph.D. in Electrical Engineering from Washington University in St. Louis (WashU) in 2020 and served as a Postdoctoral Scholar in the Department of Psychological & Brain Sciences at the University of California, Santa Barbara (UCSB) from 2020 to 2022. Before joining UA in 2025, he was an Assistant Professor at Shanghai Jiao Tong University (SJTU) from 2022 to 2024. Dr. Zhou is the recipient of the SPIE Community Champion Award and the SPIE Medical Imaging Cum Laude Award. He also served as a Program Committee Member for SPIE Medical Imaging, an Area Chair for the Conference on Health, Inference, and Learning (CHIL), an Area Chair for the Machine Learning for Health (ML4H) Symposium, and a reviewer for a wide variety of journals. His research aims to facilitate the objective assessment and optimization of image quality, as well as to understand and improve how humans and machines interpret medical images to support radiological decision-making. He has been active in publishing research articles in peer-reviewed journals, such as IEEE Transactions on Medical Imaging, Medical Physics, Optics Express, Journal of Biomedical Optics, and Journal of Medical Imaging.
Can't Join Us In Person?
Register for the Zoom Webinar!
Subscribe to Upcoming Colloquium Announcements
Visit our website for future lecture dates and speaker information