Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Advancing Limited-Angle CT Reconstruction Through Diffusion-Based Sinogram Completion

Jiaqi Guo, Santiago Lopez-Tapia, Aggelos K. Katsaggelos

TL;DR

This work tackles limited-angle CT reconstruction by shifting the problem to sinogram inpainting. It introduces MR-SDE to complete missing projection data in the sinogram domain, followed by a one-step diffusion distillation with RNSD-based rectification and a post-processing stage after FBP to refine the image. The approach delivers state-of-the-art perceptual and fidelity metrics on LACT benchmarks, while dramatically reducing inference time compared to image-domain diffusion methods. The combination of sinogram-domain diffusion, data-consistency rectification, and efficient post-processing positions this method as a practical solution for rapid, accurate LACT reconstructions in clinical and scientific settings.

Abstract

Limited Angle Computed Tomography (LACT) often faces significant challenges due to missing angular information. Unlike previous methods that operate in the image domain, we propose a new method that focuses on sinogram inpainting. We leverage MR-SDEs, a variant of diffusion models that characterize the diffusion process with mean-reverting stochastic differential equations, to fill in missing angular data at the projection level. Furthermore, by combining distillation with constraining the output of the model using the pseudo-inverse of the inpainting matrix, the diffusion process is accelerated and done in a step, enabling efficient and accurate sinogram completion. A subsequent post-processing module back-projects the inpainted sinogram into the image domain and further refines the reconstruction, effectively suppressing artifacts while preserving critical structural details. Quantitative experimental results demonstrate that the proposed method achieves state-of-the-art performance in both perceptual and fidelity quality, offering a promising solution for LACT reconstruction in scientific and clinical applications.

Advancing Limited-Angle CT Reconstruction Through Diffusion-Based Sinogram Completion

TL;DR

This work tackles limited-angle CT reconstruction by shifting the problem to sinogram inpainting. It introduces MR-SDE to complete missing projection data in the sinogram domain, followed by a one-step diffusion distillation with RNSD-based rectification and a post-processing stage after FBP to refine the image. The approach delivers state-of-the-art perceptual and fidelity metrics on LACT benchmarks, while dramatically reducing inference time compared to image-domain diffusion methods. The combination of sinogram-domain diffusion, data-consistency rectification, and efficient post-processing positions this method as a practical solution for rapid, accurate LACT reconstructions in clinical and scientific settings.

Abstract

Limited Angle Computed Tomography (LACT) often faces significant challenges due to missing angular information. Unlike previous methods that operate in the image domain, we propose a new method that focuses on sinogram inpainting. We leverage MR-SDEs, a variant of diffusion models that characterize the diffusion process with mean-reverting stochastic differential equations, to fill in missing angular data at the projection level. Furthermore, by combining distillation with constraining the output of the model using the pseudo-inverse of the inpainting matrix, the diffusion process is accelerated and done in a step, enabling efficient and accurate sinogram completion. A subsequent post-processing module back-projects the inpainted sinogram into the image domain and further refines the reconstruction, effectively suppressing artifacts while preserving critical structural details. Quantitative experimental results demonstrate that the proposed method achieves state-of-the-art performance in both perceptual and fidelity quality, offering a promising solution for LACT reconstruction in scientific and clinical applications.

Paper Structure

This paper contains 12 sections, 9 equations, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Method
  3. MR-SDE for Sinogram Inpainting
  4. Distillation with RNSD-based Rectification
  5. Experiment
  6. General Experiment Setting
  7. Dataset and Preprocessing
  8. Basic Experiment Setup
  9. Ablation Study
  10. Evaluation and Comparison
  11. Visual Evaluation and Runtime Efficiency
  12. Conclusion & Discussion

Figures (2)

  • Figure 1: Illustration of our proposed LACT reconstruction pipeline. (a) MR-SDE is employed to restore missing angular information via a diffusion process. (b) Sinogram pairs $(\mathbf{x}_\mathbf{T}, \mathbf{x}_\mathbf{0})$ generated by MR-ODE are used for model distillation. During testing, the distilled model is sampled $N$ times to produce multiple inpainted sinograms $\{\mathbf{x}^i_0\}^N_{i=0}$. (c) The inpainted sinograms are back-projected using FBP, and their mean and standard deviation are computed and fed to the post-processing module $\tau_{\psi}$.
  • Figure 2: Visualization of reconstructions generated by RN-SDE and our proposed method under three LACT scenarios. The number at the bottom right corner shows the corresponding PSNR relative to the Ground Truth. (first column)