Quantitative multi-spectral (or multi-energy) CT remains an active research topic in the CT field. Algorithms for accurate image reconstruction play a key role in quantitative multi-spectral CT imaging. In the presentation, using quantitative dual-energy CT (QDECT) as an example, I discuss our recent research on optimization-based algorithms for accurate image reconstruction in QDECT. The inherently non-linear data model in QDECT leads to a non-convex optimization problem that needs new algorithms to solve. The presentation discusses the non-convex primal-dual (NCPD) algorithm that we have developed for image reconstruction through solving a non-convex optimization problem in QDECT. Evidence will be provided to show that the NCPD algorithm can numerically accurately solve the non-convex optimization problem and thus reconstruct images in QDECT, and that the algorithm can also be exploited for enabling innovative QDECT scanning configurations of practical application significance.

If time allows, I will also discuss the claim in literature that machine learning (ML), neural network (NN), deep learning (DL) or artificial intelligence (AI) can solve an inverse problem in CT. Specifically, I will share with the audience recent results of ML/NN/DL studies of colleagues around the world on a basic inverse problem in CT, which yield no evidence to support the claims made in literature.

Short Bio: Xiaochuan Pan is a faculty member in the Department of Radiology, Department of Radiation & Cellular Oncology, the Committee on Medical Physics, the Comprehensive Cancer Center, and the College at The University of Chicago. His research centers on physics, algorithms, and engineering underpinning tomographic imaging and its biomedical and clinical applications.