Speaker
Dr. Zhiling Gu, Postdoctoral Associate at Yale School of Public Health, Yale University
Title
Statistics Seminar Series
Subtitle
Building Confidence in AI-Generated Biomedical Images through Advanced Statistical Inference
Physical Location
Allen 14
Abstract:
Generative artificial intelligence (AI) has transformed biomedical imaging through image synthesis, helping to address challenges related to data availability, privacy, and population diversity. In this talk, I will present a novel nonparametric method within the functional data analysis framework to identify significant differences in both the mean and covariance structures of original and synthetic biomedical imaging data, with the goal of improving the fidelity and utility of synthetic data. To account for complex spatial heterogeneity, the proposed approach employs triangulated spherical splines. A key methodological contribution is the construction of simultaneous confidence regions that enable rigorous uncertainty quantification of original-synthetic differences. We establish the asymptotic properties of these confidence regions, demonstrating exact coverage probabilities and theoretical equivalence to inference procedures derived from noise-free imaging data. Simulation studies further validate the coverage properties and assess the size and power of the associated hypothesis tests. We apply the proposed methodology to original and synthetic brain imaging data from the Human Connectome Project, revealing systematic differences between the two. We further show that a simple transformation can substantially reduce these discrepancies, thereby enhancing the reliability and downstream utility of synthetic images for biomedical research. Finally, I will place this work in the context of several related ongoing projects that use multimodal data to study brain aging and mental health risk assessment.
Note:
Contact Prof. JZ at jzhang@math.msstate.edu for additional information.