November 26, 2024

Diffusion Model based Super-resolution for Medical Images

High-resolution medical imaging is critical for precise diagnosis and optimal patient care. Yet, obtaining such images often necessitates prolonged scanning durations, leading to discomfort for patients and increasing the risk of scan failures due to patient movements. To address these challenges, super-resolution techniques were proposed and offer a promising avenue for enhancing the quality of medical images. Traditional super- resolution methods often fall short in preserving intricate details inherent to medical images. This talk introduces deep learning approaches, particularly the diffusion models, to enhance the resolution of medical images. We begin by discussing the fundamental challenges in medical image super-resolution and the limitations of conventional methods. Then, we talk about the deep learning approaches for image super-resolution. In particular, we delve into the principles of diffusion models and shed light on why they are promising in this application. Experimental results will be showcased, comparing our diffusion model-based approach with other SOTA super-resolution methods. The outcomes reveal notable improvements in preserving fine-grained details and critical structures, making it an invaluable tool for clinicians.

About the speaker

Dr. Xianqi Li is currently a faculty member in the Department of Mathematics and Systems Engineering at Florida Institute of Technology. His principal research interests now lie in the areas of Artificial Intelligence, Optimization, and Statistical Inference. Previously, he was a postdoctoral research fellow in the Department of Radiology, Harvard Medical School. He received his Ph.D. from the Department of Mathematics and MS from ECE both at UFL.