May 15, 2025 Webinar
An Overview of Bayesian Image Analysis in Transformed Space
John Kornak, PhD
Abstract:
Bayesian image analysis offers a powerful way to improve image quality and highlight important features by combining prior knowledge with probabilistic modeling. This talk will provide an overview of Bayesian Image Analysis in Transformed Space (BITS), a new approach that uses transformed domains, like Fourier and wavelet spaces, to make analysis more efficient and flexible. I'll describe how Bayesian Image Analysis in Fourier Space (BIFS) takes advantage of frequency-space structure to support rich prior models, and how Data-Driven BIFS (DD-BIFS) further extends this by learning priors from data. I'll also introduce Bayesian Image Analysis in Wavelet Space, which captures patterns at multiple scales and handles non-stationary features common in real-world images. By broadening the kinds of prior knowledge that can be used-and doing so efficiently-BITS opens up new possibilities for Bayesian image analysis. I will illustrate with examples from neuroimaging using arterial spin labeling (ASL) and functional MRI (fMRI).
Short Bio:
John Kornak is a Professor of Biostatistics at the University of California, San Francisco (UCSF), where he has been a faculty member since 2002. He earned his undergraduate degree in Mathematics and PhD in Statistics from the University of Nottingham. Dr. Kornak began his UCSF career in the Department of Radiology and Biomedical Imaging before transitioning to Epidemiology and Biostatistics. In addition to his faculty role, he serves as Head of the Health Data Science Program and Director of the Biostatistics Consulting Unit within UCSF's Clinical and Translational Sciences Institute (CTSI). His research focuses on applying statistical and computational methods to medical imaging, emphasizing applications to the study of neurodegenerative diseases such as Alzheimer's disease and frontotemporal dementia.
Hope to see you at the webinar!