Abstract
Methods
Diffusion-based methods evaluated in DM4CT. Columns under Technique refer to implementation choices (e.g., latent-space diffusion or DDIM-based sampling). Columns under Reconstruction Strategy denote how measurement conditioning is incorporated, including data consistency gradient steering (DC-grad), separate optimization steps (DC-step), plug-and-play priors, and use of approximate pseudoinverse solutions. A
| Method | Year | Latent | DDIM | DC-grad | DC-step | Plug-and-Play | Pseudo Inv | Variational Bayes |
|---|---|---|---|---|---|---|---|---|
| MCG | 2022 | ✗ | ✗ | ✗ | ✗ | ✗ | ✓* | ✗ |
| DPS | 2023 | ✗ | ✗ | ✓ | ✗ | ✗ | ✗ | ✗ |
| PSLD | 2023 | ✓ | ✗ | ✓ | ✗ | ✗ | ✗ | ✗ |
| PGDM | 2023 | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | ✗ |
| DDS | 2024 | ✗ | ✓ | ✗ | ✓‡ | ✗ | ✗ | ✗ |
| Resample | 2024 | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✗ |
| DMPlug | 2024 | ✗ | ✓ | ✗ | ✓ | ✓ | ✗ | ✗ |
| Reddiff | 2024 | ✗ | ✓ | ✓ | ✗ | ✗ | ✓ | ✓ |
| HybridReg | 2025 | ✗ | ✓ | ✓ | ✗ | ✗ | ✓ | ✓ |
| DiffStateGrad | 2025 | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✗ |
Main Visualization
Reconstruction results of diffusion-based and other established methods. Top: medical dataset (config iv, 80 angles with noise & ring artifacts); middle: industrial dataset (config ii, 20 angles with mild noise); bottom: real-world synchrotron dataset (60 angles). Red and green boxes show zoom-in regions. PSNR and SSIM appear in the top-left and top-right of each image. A dash (–) indicates that the method exceeded the 40 GB GPU memory limit for single-slice reconstruction and is therefore not executed. Images are consistently linear rescaled across methods to improve contrast.
Some Interesting Results
Tradeoff between Prior and Data Consistency & Reconstruction Uncertainty
(a) Impact of data consistency step size η on PSNR and data fit in DPS. Moderate values improve both, while large η disrupts denoising and causes collapse. Visual examples in the plot highlight the transition from prior-dominated to noise-dominated reconstructions. (b) Mean and standard deviation of ten MCG reconstructions conditioned on the same real measurement. Note that the real measurement used in (b) is different from the one used for (a).
Prior Contribution and Consistency: A Null Space Perspective
Decomposition of reconstructions into range and null space components for different data consistency strategies with config i). For each method, the full reconstruction is shown on the left, with zoomed-in red insets of the range component in the center and the corresponding null component on the right. The top-left of each null component indicates its relative L2 energy as a percentage of the total reconstruction, reflecting the extent of content introduced by the prior. Zoom in for details.
Data Consistency for Latent Diffusion: Gradient or Optimization?
Reconstruction results of latent diffusion methods using only data consistency gradients (PSLD) versus additional optimization steps (ReSample) under noise-free (40 projections, no noise) and noisy (80 projections) scenarios. ADMM-PDTV serves as a classical model-based baseline that applies data consistency optimization with heuristic prior. Red insets show magnified regions.
Computation Efficiency
(a) Reconstruction time and GPU memory. The time is counted on medical dataset. (b) Training time and GPU memory of pixel diffusion, latent diffusion and SwinIR.
Fine Tuning Existing Natural Image Encoders (SDXL)
Comparison of autoencoder reconstruction, unconditional diffusion generation, and CT reconstruction across different autoencoders. The VQ-VAE used in our benchmark produces consistently superior representations and reconstructions, while SDXL AutoencoderKL variants exhibit reduced stability and quality.
Comparison between Training Stages on Reconstruction Performance
Visualization of unconditional genration and CT reconstruction using different stages of the trained diffusion models. The early-stage model produces noisy unconditional generations, while it yields the sharpest structures and the best fine-detail recovery for CT reconstruction.
Acknowledgments
The authors acknowledge financial support by the European Union H2020-MSCA-ITN-2020 under grant agreement no. 956172 (xCTing). JS is also supported by grant from Dutch Research Council under grant no. ENWSS.2018.003 (UTOPIA) and no. NWA.1160.18.316 (CORTEX). The computation in this work is supported by SURF Snellius HPC infrastructure under grant no. EINF-15060. Synchrotron data acquisition was financially supported by the Dutch Research Council, project no. 016.Veni.192.23.
BibTeX
@inproceedings{
shi2026dmct,
title={{DM}4{CT}: Benchmarking Diffusion Models for Computed Tomography Reconstruction},
author={Shi, Jiayang and Pelt, Dani{\"e}l M and Batenburg, K Joost},
booktitle={The Fourteenth International Conference on Learning Representations},
year={2026},
url={https://openreview.net/forum?id=YE5scJekg5}
}