1. Home
  2. Project Highlights - 2023
  3. High-Performance Computing, Simulation, and Data Science - 2023
  4. Next-Generation Implicit Neural Representation for Ill-Posed and Dynamic Computed Tomography Reconstruction

Next-Generation Implicit Neural Representation for Ill-Posed and Dynamic Computed Tomography Reconstruction

Hyojin Kim | 22-ERD-032

Executive Summary

We will use implicit neural representations to develop a next-generation, training-free, computed tomography reconstruction approach to address extremely ill-posed reconstruction problems, such as limited-angle, few-view, and dynamic four-dimensional computed tomography of moving objects. This new capability will lead to significant improvements in non-destructive evaluations, detection of contraband and explosives, weapon inspections, and other mission-relevant applications.

Publications, Presentations, and Patents

K. A. Mohan, H. Kim, “4D Time-Space Reconstruction of Dynamic Objects using X-ray CT” (Presentation, DHS RDT&E Summit, TSA Headquarters, June, 9, 2023).

H. Kim, J. Liu, R. Anirudh, S. He, U. Kamilov, K. A. Mohan, J. Thiagarajan, “Generalizable diffusion-based machine learning inference for severely ill-posed CT reconstruction” (Presentation, DHS RDT&E Summit, TSA Headquarters, June 9, 2023).

H. Kim, K. Champley, “Differentiable Forward Projector for X-ray Computed Tomography” (Presentation, ICML Workshop - Differentiable Almost Everything, Livermore, CA, 2023).

H. Kim, K. Champley, “Livermore AI Projector (LEAP)” (Presentation, Center for Advanced Signal and Image Sciences (CASIS) 27th Annual Workshop, Livermore, CA, 2023).

J. Liu, R. Anirudh, J. J. Thiagarajan, K. A. Mohan, K. Champley, S. He, U. Kamilov, H. Kim, “Model-Based Probabilistic Diffusion Architecture for Limited-Angle CT Reconstruction” (Presentation, Conference on Computer Vision and Pattern Recognition (ICCV) (Accepted), Paris, France, 2023). 

H. Kim, K. Champley, “Livermore AI Projector (LEAP)” Open-source software release, 2023.

H. Kim, J. Liu, “DOLCE: A Model-Based Probabilistic Diffusion Framework for Limited-Angle CT Reconstruction (DOLCE).” v. 1.0, Open-source software release (Approved).

H. Kim, K. A. Mohan, “Continuous Projection With Implicit Neural Representation for Dynamic and Ill-Posed Computed Tomography.” Record of Invention, submitted August 28, 2023. (IL-13897).