“Learned Alternating Minimization Algorithm for Sparse-View CT Reconstruction: A Multi-Modal Approach”
Wednesday, Nov. 15 at 3:00pm
NEB 589

Abstract

We propose a novel Learned Alternating Minimization Algorithm (LAMA) for dual-domain sparse-view CT image reconstruction. LAMA is naturally induced by a variational model for CT reconstruction with learnable non-smooth non-convex regularizers, which are parameterized as composite functions of deep networks in both image and sinogram domains. To minimize the objective of the model, we incorporate the smoothing technique and residual learning architecture into the design of LAMA. We show that LAMA substantially reduces network complexity, improves memory efficiency and reconstruction accuracy, and is provably convergent for reliable reconstructions. Extensive numerical experiments demonstrate that LAMA outperforms existing methods by a wide margin on multiple benchmark CT datasets.

Paper link: https://arxiv.org/pdf/2306.02644.pdf
Github link: https://github.com/chrisdcs/LAMA-Learned-Alternating-Minimization-Algorithm

Biography

Returning to CNEL! Chi Ding, the math enthusiast, is back in the ECE department. After a Master’s degree from ECE, he has returned for a refreshed adventure. Currently a third-year Mathematics PhD student from the University of Florida, working on optimization and deep learning research, with a focus on medical imaging.