Statistics Seminar: Boosting Biomedical Imaging Analysis via Distributed Functional Regression and Synthetic Surrogates
Presenter: Guannan Wang, Associate Professor, The College of William & Mary
Description: Generative AI has emerged as a powerful tool for synthesizing biomedical images, offering new solutions to challenges such as data scarcity, privacy constraints, and modality imbalance. However, the reliable use of synthetic images in scientific analysis requires principled statistical frameworks that can assess fidelity and rigorously quantify uncertainty. In this talk, I present a distributed functional data analysis approach for comparing original and AI- generated biomedical images through their mean and covariance structures. Using spline-based representations on complex imaging domains, we construct simultaneous confidence regions, enabling formal inference on original-synthetic differences and providing statistical safeguards for downstream analyses. Building on this foundation, I demonstrate how synthetic images can
be safely incorporated into functional regression models to learn spatially varying covariate effects when key imaging modalities are partially observed. Applications to large-scale neuroimaging studies illustrate how integrating generative AI with rigorous statistical inference enhances the reliability, interpretability, and scientific value of modern biomedical imaging analyses.
Bio: Guannan Wang is a Diamond Term Distinguished Associate Professor in the Department of Mathematics at William & Mary. She received a Ph.D. in Statistics and an M.S. in Computer Science from the University of Georgia in 2015. Her research focuses on the statistical foundations of generative AI, distributed and federated learning, and spatial and functional data analysis, with applications to neuroimaging, public health, and environmental and social sciences. She has published over 30 peer-reviewed articles in leading statistical journals, including JASA, JCGS, Statistica Sinica, Biometrics, and JMLR, and her work has been supported by the NIH, NSF, and the Simons Foundation.
Hosted by: Statistics Department
Zoom link: https://ucsc.zoom.us/j/92479478035?pwd=S6b9SNtCorApA04sISbDwWqaF3wyPZ.1
