Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Clinical Metadata Guided Limited-Angle CT Image Reconstruction

Yu Shi, Shuyi Fan, Changsheng Fang, Shuo Han, Haodong Li, Li Zhou, Bahareh Morovati, Dayang Wang, Hengyong Yu

Abstract

Limited-angle computed tomography (LACT) offers improved temporal resolution and reduced radiation dose for cardiac imaging, but suffers from severe artifacts due to truncated projections. To address the ill-posedness of LACT reconstruction, we propose a two-stage diffusion framework guided by structured clinical metadata. In the first stage, a transformer-based diffusion model conditioned exclusively on metadata, including acquisition parameters, patient demographics, and diagnostic impressions, generates coarse anatomical priors from noise. The second stage further refines the images by integrating both the coarse prior and metadata to produce high-fidelity results. Physics-based data consistency is enforced at each sampling step in both stages using an Alternating Direction Method of Multipliers module, ensuring alignment with the measured projections. Extensive experiments on both synthetic and real cardiac CT datasets demonstrate that incorporating metadata significantly improves reconstruction fidelity, particularly under severe angular truncation. Compared to existing metadata-free baselines, our method achieves superior performance in SSIM, PSNR, nMI, and PCC. Ablation studies confirm that different types of metadata contribute complementary benefits, particularly diagnostic and demographic priors under limited-angle conditions. These findings highlight the dual role of clinical metadata in improving both reconstruction quality and efficiency, supporting their integration into future metadata-guided medical imaging frameworks.

Clinical Metadata Guided Limited-Angle CT Image Reconstruction

Abstract

Limited-angle computed tomography (LACT) offers improved temporal resolution and reduced radiation dose for cardiac imaging, but suffers from severe artifacts due to truncated projections. To address the ill-posedness of LACT reconstruction, we propose a two-stage diffusion framework guided by structured clinical metadata. In the first stage, a transformer-based diffusion model conditioned exclusively on metadata, including acquisition parameters, patient demographics, and diagnostic impressions, generates coarse anatomical priors from noise. The second stage further refines the images by integrating both the coarse prior and metadata to produce high-fidelity results. Physics-based data consistency is enforced at each sampling step in both stages using an Alternating Direction Method of Multipliers module, ensuring alignment with the measured projections. Extensive experiments on both synthetic and real cardiac CT datasets demonstrate that incorporating metadata significantly improves reconstruction fidelity, particularly under severe angular truncation. Compared to existing metadata-free baselines, our method achieves superior performance in SSIM, PSNR, nMI, and PCC. Ablation studies confirm that different types of metadata contribute complementary benefits, particularly diagnostic and demographic priors under limited-angle conditions. These findings highlight the dual role of clinical metadata in improving both reconstruction quality and efficiency, supporting their integration into future metadata-guided medical imaging frameworks.

Paper Structure

This paper contains 26 sections, 10 equations, 12 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Problem Formulation and Framework Overview
  4. Data Curation
  5. Metadata-guided Diffusion for LACT Generation
  6. Metadata-conditioned Diffusion Sampling
  7. Model Architecture
  8. Data Consistency Regularization
  9. Supervised Fine-Tuning (SFT)
  10. Experiments Setup
  11. Competing Methods
  12. Implementation Details
  13. Evaluation Metrics
  14. Results
  15. CTRATE Dataset
  16. ...and 11 more sections

Figures (12)

  • Figure 1: Overview of the proposed two-stage, metadata-guided diffusion framework for LACT reconstruction.
  • Figure 2: Internal architecture of the conditional UNet Encoder/Decoder Block in the proposed diffusion framework.
  • Figure 3: Visual comparison of metadata-conditioned image generation with ground-truth chest CT cases. Each column pair shows the synthesized image from the metadata-guided diffusion model (bottom) and the full-view reference (top).
  • Figure 4: Qualitative comparison for representative baselines in fan-beam LACT on the CTRATE dataset. (a)–(d) correspond to 120°, 90°, 60° (chest), 60° (abdomen) cases, respectively. Each subgraph shows reconstructions by ADMM-TV, FBPConvNet, DOLCE, DDS, proposed Stage-I, proposed complete framework, and the full-angle reference. Boxes indicate magnified ROIs with obvious artifacts or hallucinations, and arrows highlight noticeable artifacts or local inconsistencies within the ROIs. All slices are displayed with a $[-1000, 1000]$ Hounsfield unit (HU) display window. PSNR/SSIM are annotated.
  • Figure 5: Qualitative comparison with representative baselines for fan-beam LACT under an unseen 75° scan on the CTRATE dataset. (a) and (b) show two representative slices. In (a), the red-box indicates a magnified high-contrast bony ROI; in (b), the blue-box indicates a magnified a mediastinal soft-tissue ROI containing major vessels and airway structures. Each subfigure includes reconstructions from ADMM-TV, FBPConvNet, DOLCE, DDS, proposed Stage-I, proposed complete framework (Stage-II), and the full-angle reference. Arrows highlight noticeable differences in the ROIs. The display window is $[-1000, 1000]$ HU for the full slices and the bone ROIs, and $[-700, 700]$ HU for the zoomed-in mediastinal ROI. PSNR/SSIM are annotated.
  • ...and 7 more figures