Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Low performing pixel correction in computed tomography with unrolled network and synthetic data training

Hongxu Yang, Levente Lippenszky, Edina Timko, Lehel Ferenczi, Gopal Avinash

TL;DR

The paper tackles ring and streak artifacts from low performing CT detector pixels (LPP) by formulating LPP correction as a dual-domain inverse problem and training an unrolled ISTA-Net using synthetic data generated from natural images. By leveraging the intrinsic correlations between the sinogram and image domains via the forward CT model, the approach corrects LPP without real clinical data, solving a regularized optimization that combines data fidelity with a learned prior. The method demonstrates superior correction over state-of-the-art approaches on synthetic and public datasets, with strong generalization to different anatomies (e.g., RibFrac) even when trained synthetically. This supports software-based LPP artifact mitigation as a cost-effective alternative to hardware replacement across varied scanner configurations.

Abstract

Low performance pixels (LPP) in Computed Tomography (CT) detectors would lead to ring and streak artifacts in the reconstructed images, making them clinically unusable. In recent years, several solutions have been proposed to correct LPP artifacts, either in the image domain or in the sinogram domain using supervised deep learning methods. However, these methods require dedicated datasets for training, which are expensive to collect. Moreover, existing approaches focus solely either on image-space or sinogram-space correction, ignoring the intrinsic correlations from the forward operation of the CT geometry. In this work, we propose an unrolled dual-domain method based on synthetic data to correct LPP artifacts. Specifically, the intrinsic correlations of LPP between the sinogram and image domains are leveraged through synthetic data generated from natural images, enabling the trained model to correct artifacts without requiring any real-world clinical data. In experiments simulating 1-2% detectors defect near the isocenter, the proposed method outperformed the state-of-the-art approaches by a large margin. The results indicate that our solution can correct LPP artifacts without the cost of data collection for model training, and it is adaptable to different scanner settings for software-based applications.

Low performing pixel correction in computed tomography with unrolled network and synthetic data training

TL;DR

The paper tackles ring and streak artifacts from low performing CT detector pixels (LPP) by formulating LPP correction as a dual-domain inverse problem and training an unrolled ISTA-Net using synthetic data generated from natural images. By leveraging the intrinsic correlations between the sinogram and image domains via the forward CT model, the approach corrects LPP without real clinical data, solving a regularized optimization that combines data fidelity with a learned prior. The method demonstrates superior correction over state-of-the-art approaches on synthetic and public datasets, with strong generalization to different anatomies (e.g., RibFrac) even when trained synthetically. This supports software-based LPP artifact mitigation as a cost-effective alternative to hardware replacement across varied scanner configurations.

Abstract

Low performance pixels (LPP) in Computed Tomography (CT) detectors would lead to ring and streak artifacts in the reconstructed images, making them clinically unusable. In recent years, several solutions have been proposed to correct LPP artifacts, either in the image domain or in the sinogram domain using supervised deep learning methods. However, these methods require dedicated datasets for training, which are expensive to collect. Moreover, existing approaches focus solely either on image-space or sinogram-space correction, ignoring the intrinsic correlations from the forward operation of the CT geometry. In this work, we propose an unrolled dual-domain method based on synthetic data to correct LPP artifacts. Specifically, the intrinsic correlations of LPP between the sinogram and image domains are leveraged through synthetic data generated from natural images, enabling the trained model to correct artifacts without requiring any real-world clinical data. In experiments simulating 1-2% detectors defect near the isocenter, the proposed method outperformed the state-of-the-art approaches by a large margin. The results indicate that our solution can correct LPP artifacts without the cost of data collection for model training, and it is adaptable to different scanner settings for software-based applications.
Paper Structure (10 sections, 4 equations, 3 figures, 1 table)

This paper contains 10 sections, 4 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Method
  3. Theoretical Analysis of LPP Artifacts
  4. Problem Formulation
  5. Synthetic Data for Training
  6. Experiments and results
  7. Dataset
  8. Experiment settings
  9. Experiment results
  10. Conclusions

Figures (3)

  • Figure 1: Example CT and synthetic images with artifacts. Top: sinogram and corresponding CT with and without LPP. Bottom: sinogram and corresponding synthetic image with and without LPP. LPP: Zero-out lines along view axis.
  • Figure 2: Structure of ISTA-Net. The observed sinogram with LPP is reconstructed by FBP to generate observed CT, which is processed by k-layer ISTA-Net. In addition to the CT image, the detector LPP information is also provided as input to ISTA-Net characterizing the sinogram defect pattern. Output is the corrected CT without LPP artifacts.
  • Figure 3: Qualitative comparison for LPP artifact removal methods (a) FBP, (b) Interpolation, (c) AST, (d) DeepRAR, (e) NAFNet, (f) Riner, (g) Ours-Real (h) Ours-Syn, and (i) Ground truth. Top: images in HU value. Bottom: HU value differences to GT.