DM4CT: Benchmarking Diffusion Models for Computed Tomography Reconstruction

Jiayang Shi; Daniel M. Pelt; K. Joost Batenburg

eess.IVFeb 20, 2026

DM4CT: Benchmarking Diffusion Models for Computed Tomography Reconstruction

Authors: Jiayang Shi, Daniel M. Pelt, K. Joost Batenburg

Organizations: LIACS, Leiden University · Centrum Wiskunde en Informatica

Abstract

Diffusion models have recently emerged as powerful priors for solving inverse problems. While computed tomography (CT) is theoretically a linear inverse problem, it poses many practical challenges. These include correlated noise, artifact structures, reliance on system geometry, and misaligned value ranges, which make the direct application of diffusion models more difficult than in domains like natural image generation. To systematically evaluate how diffusion models perform in this context and compare them with established reconstruction methods, we introduce DM4CT, a comprehensive benchmark for CT reconstruction. DM4CT includes datasets from both medical and industrial domains with sparse-view and noisy configurations. To explore the challenges of deploying diffusion models in practice, we additionally acquire a high-resolution CT dataset at a high-energy synchrotron facility and evaluate all methods under real experimental conditions. We benchmark ten recent diffusion-based methods alongside seven strong baselines, including model-based, unsupervised, and supervised approaches. Our analysis provides detailed insights into the behavior, strengths, and limitations of diffusion models for CT reconstruction. The real-world dataset is publicly available at zenodo.org/records/15420527, and the codebase is open-sourced at github.com/DM4CT/DM4CT.

Figures & tables

Figure 1: Overview of the DM4CT benchmark. (a) The reconstruction pipeline, where representative diffusion and baseline methods are applied to measured sinograms using the same forward model. (b) The datasets used in the benchmark, including two simulated CT datasets (medical and industrial) and one real-world dataset acquired at a synchrotron facility. (c) The five simulation configurations used to evaluate robustness to limited views, noise, and ring artifacts. Two example FBP reconstructions under noise and ring artifact conditions are shown. (d) The evaluation metrics, including both qualitative (visual) and quantitative (image quality and computational efficiency) criteria.

Method	Year	Technique		Reconstruction Strategy
Method	Year	Latent	DDIM	DC-grad	DC-step	Plug-and-Play	Pseudo Inv	Variational Bayes
MCG ( Chung et al., 2022 )	2022						✓ ∗
DPS ( Chung et al., 2023 )	2023			✓
PSLD ( Rout et al., 2023 )	2023	✓		✓
PGDM ( Song et al., 2023a )	2023						✓
DDS ( Chung et al., 2024 )	2024		✓		✓ ‡

Table 1: 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 ✓ ∗ indicates only a single-step update toward the pseudoinverse. A ✓ ‡ indicates methods that incorporate data fidelity via a conjugate-gradient solve rather than a direct pixel-space optimization step.

Method

Medical

Industrial

Real-world

config i

config ii

config iii

config iv

config v

config i

config ii

config iii

config iv

config v

200 projs

100 projs

60 projs

FBP

26.98/0.69

9.89/0.03

12.78/0.09

14.50/0.13

24.14/0.64

13.73/0.19

10.65/0.09

15.01/0.25

13.21/0.18

14.23/0.18

27.76/0.56

26.35/0.41

24.95/0.30

SIRT

30.40/0.80

26.23/0.47

24.48/0.32

25.86/0.40

26.49/0.72

18.40/0.38

16.67/0.30

19.48/0.46

19.17/0.40

17.86/0.36

28.16/0.56

28.06/0.54

27.92/0.52

ADMM-PDTV

30.56/0.79

25.12/0.36

18.20/0.10

20.11/0.15

29.69/0.80

16.95/0.31

18.02/0.38

20.45/0.43

19.25/0.34

20.03/0.53

28.13/0.53

28.01/0.52

27.92/0.52

FISTA-SBTV

30.57/0.82

26.28/0.70

25.39/0.67

27.84/0.74

27.68/0.76

17.63/0.37

18.34/0.46

20.18/0.54

20.12/0.52

19.20/0.46

28.03/0.52

28.01/0.52

27.78/0.50

DIP

28.58/0.80

24.13/0.61

26.40/0.66

27.89/0.71

27.87/0.75

19.35/0.41

16.99/0.36

21.29/0.52

19.66/0.41

19.16/0.41

24.57/0.46

24.36/0.44

24.27/0.39

Table 2: Reconstruction performance (PSNR / SSIM) of different methods under various configurations for medical, industrial and synchrotron CT datasets. The highest score among diffusion-based methods is shown in bold , and the second highest is underlined . A dash (–) indicates that the method exceeded the 40 GB GPU memory limit for single-slice reconstruction and is therefore not executed.

Panel of Figure 3 — Figure 3: (a) Impact of data consistency step size $\eta$ (Equation 7 ) on PSNR and data fit in DPS. Moderate values improve both, while large $\eta$ 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).

Figure 4: 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.

Figure 5: 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.

Panel of Figure 6 — Figure 6: PSNR comparisons of method categories under (a) varying number of projection angles on the medical dataset (sparsity), and (b) different noise levels (average photon count) on the industrial dataset. Pixel diffusion refers to the six diffusion methods operating in image space; latent diffusion includes the other three latent-space methods (see Table 1 ). Network prior includes DIP and INR, classical methods include FBP and SIRT, and MBIR refers to ADMM-PDTV and FISTA-SBTV. Shaded regions indicate standard deviation of all methods in the category group.

Panel of Figure 7 — Figure 7: (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.

Appendix figures & tables25 assets

Supplementary material from the paper’s appendix.

Appendix

Configuration	average absorption (noise)	$I_{0}$ (noise)	$p_{\text{ring}}$ (ring)	$\sigma^{2}$ (ring)
i) 40 angles without noise	-	-	-	-
ii) 20 angles with mild noise	50%	10000	-	-
iii) 80 angles with more noise	50%	5000	-	-
iv) 80 angles with noise and ring artifacts	50%	10000	0.05	$0.25\cdot\sigma^{2}_{\bm{y}_{0}}$
v) 40 angles [0, $\frac{3}{4}\pi$ )	-	-	-	-

Appendix

Table 3: Simulation parameters used for generating noisy and artifact-corrupted sinograms.

I_{0}

controls the Poisson noise level based on the Beer–Lambert model (see Equation 11 ), while

p_{\text{ring}}

and

\sigma^{2}

define the severity and intensity of ring artifacts (see Equation 14 ).

\sigma_{\bm{y}_{0}}

is the standard deviation of original clean measurement

\bm{y}_{0}

. A value of “–” indicates no corruption of that type.

Sample	# Proj.	Pixel Size ( $\mu$ m)	Exposure (s)	Filter	Energy (keV)	# Dark Fields	# Flat Fields	Center (pre crop)	Center (post crop)	Crop Size
F3_1 (train)	1200	9	4	None	24	10	10	-61	-40.5	$679\times 768$
F3_2 (test)	1200	9	4	None	24	10	10	-62	-41.5	$679\times 768$

Appendix

Table 4: Acquisition parameters of the real-world synchrotron CT dataset used in this benchmark. Both rocks are scanned using the same setup under parallel-beam geometry.

Figure 8: Example projection images (before cropping) of the synchrotron dataset at different angles.

Figure 9: Reference reconstructions of three slices from the training and test rocks using all 1200 angles.

Method	FDK	SIRT	INR	MCG	DPS	DMPlug	Reddiff
PSNR/SSIM	19.48 / 0.16	23.11 / 0.43	24.77 / 0.33	29.12 / 0.64	29.10 / 0.61	23.65 / 0.48	27.32 / 0.60

Appendix

Table 5: Reconstruction from raw projections (40 angles) under value range misalignment, corrected using an empirical linear mapping. Results are approximate and intended to demonstrate feasibility rather than optimal performance. PSNR and SSIM are computed slice-wise in 2D and averaged across slices, using the corresponding physical value range of each reference slice.

Figure 10: Reconstruction from raw projections (40 angles) under value range misalignment, corrected using an empirical linear mapping. Results are approximate and intended to demonstrate feasibility rather than optimal performance. PSNR and SSIM are shown in the lower-left and lower-right corners of each reconstruction.

Figure 11: Effect of noise model mismatch on DDS. Gaussian and Poisson noise are simulated to produce similar FBP baselines (roughly PSNR/SSIM). DDS reconstructs fine details under Gaussian noise but degrades significantly under Poisson noise, both visually and quantitatively.

Method	Dice / IoU
FBP	0.603 / 0.550
SIRT	0.819 / 0.735
ADMM-PDTV	0.781 / 0.707
FISTA-SBTV	0.418 / 0.359
DIP	0.345 / 0.283
INR	0.707 / 0.633

Appendix

Table 6: Segmentation performance (Dice / IoU) on the medical dataset with 40 projections.

Figure 12: Segmentation masks obtained using SAM for different reconstruction methods. Dice and IoU scores are shown in the upper-left and upper-right corners of each mask, respectively. These visualizations highlight how variations in reconstruction quality influence the resulting anatomical segmentation.

Autoencoder type	config	training epochs	autoencoder reconstruction	CT reconstruction
AutoencoderKL (SDXL)	w/o fine tuning	0	37.12/0.88	21.34/0.42
AutoencoderKL (SDXL)	w fine tuning	5	37.13 / 0.90	23.97/0.66
AutoencoderKL (SDXL)	from scratch	200	34.70/0.88	23.79 / 0.72
VQ-VAE (this benchmark)	from scratch	200	39.30 / 0.92	25.52 / 0.70

Appendix

Table 7: Representation (autoencoder reconstruction) and CT reconstruction quality for fine-tuned natural-image encoders. Values are PSNR/SSIM.

Figure 13: 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.

config iii)
	Pixel iters
		25	50	100	200
Latent iters	100	27.91/0.78	28.79/0.79	29.23/0.79	28.90/0.77
Latent iters	200	27.90/0.78	28.76/0.79	29.20/0.79	28.87/0.77
config iii - more noise)
Latent iters	100	27.64/0.77	28.40/0.77	28.23/0.75	27.77/0.72

Appendix

Table 8: Average PSNR/SSIM for different combinations of pixel and latent optimization iterations under three noise conditions.

Figure 14: Effect of pixel and latent optimization iterations on reconstruction quality. Fewer iterations lead to oversmoothing, while excessive iterations cause overfitting to noise. The best-performing settings often occur near the transition between these two regimes.

Stage	Early	Mid	Final
PSNR/SSIM	30.68 / 0.75	28.46 / 0.72	28.71 / 0.73

Appendix

Table 9: Comparison of CT reconstruction using early stage, mid stage and final stage trained pixel diffusion models.

Figure 15: Visualization of unconditional generation and CT reconstruction using different stages of the trained diffusion models.

Method	Medical					Industrial					Real-world
Method	config i	config ii	config iii	config iv	config v	config i	config ii	config iii	config iv	config v	200 projs	100 projs	60 projs
FBP	30388.56	12799.31	12493.05	7770.67	1996.67	5732.75	6223.18	7191.11	7039.36	5404.93	534.30	614.72	808.90
SIRT	65.31	1178.44	3440.96	2140.96	297.28	361.80	389.71	937.56	878.33	449.79	244.96	131.53	71.76
ADMM-PDTV	11.79	1074.95	3025.14	1196.18	84.25	42.94	196.32	749.10	1406.61	161.64	304.55	226.74	183.23
FISTA-SBTV	209.77	836.49	2252.61	1680.66	218.40	588.70	365.39	1154.75	1515.28	589.53	307.44	204.25	145.94
DIP	659.49	985.68	2726.06	1802.99	565.09	677.67	439.45	1228.74	1714.36	531.00	1023.01	559.77	395.69

Appendix

Table 10: L2 data fit between reconstructed projections and noisy measurements (

\|\bm{A}\hat{\bm{x}}-\widetilde{\bm{y}}\|_{2}

) for all benchmarked methods under different CT configurations. Lower values in noise-free settings indicate better consistency with the measurement, while in noisy cases, overly low values may reflect overfitting rather than accurate reconstruction.

Method	Medical					Industrial					Real-world
Method	config i	config ii	config iii	config iv	config v	config i	config ii	config iii	config iv	config v	200 projs	100 projs	60 projs
FBP	0.59	0.57	1.09	1.04	0.37	0.65	0.76	0.63	0.78	0.63	0.20	0.30	0.37
SIRT	0.41	1.16	0.64	0.57	0.35	0.64	0.73	0.53	0.56	0.63	0.42	0.40	0.41
ADMM-PDTV	0.38	0.63	0.93	0.84	0.25	0.62	0.70	0.51	0.56	0.40	0.46	0.49	0.50
FISTA-SBTV	0.40	0.52	0.52	0.46	0.40	0.72	0.75	0.66	0.66	0.72	0.56	0.56	0.57
DIP	0.38	0.50	0.46	0.40	0.32	0.50	0.59	0.41	0.46	0.50	0.39	0.39	0.41

Appendix

Table 11: LPIPS for all benchmarked methods under different CT configurations. Lower values mean better perceptual alignment.

Figure 16: Range-null space decompositions of four reconstruction methods with different data consistency strategies, evaluated on the industrial dataset (config iv: 80 projections with noise and ring artifacts). PSNR and SSIM are shown in the top-left and bottom-left corners of the reconstruction images. Energy percentages of the range and null space components are shown in their respective top-left corners.

Figure 17: Visual comparison of all benchmarked methods on the real-world synchrotron CT dataset under three different sparse-view scenarios (200, 100, and 60 projections). PSNR and SSIM values are shown in the top-left and top-right corners of each image, respectively. Red and green boxes mark zoom-in regions for structural detail comparison. The reference image is reconstructed using all 1200 projections, median flat-field correction, and additional ring artifact suppression.

Figure 18: Visual comparison of all benchmarked methods on the simulated industrial CT dataset under four different configurations: (i) 40 projections without noise, (ii) 20 projections with mild noise, (iii) 80 projections with stronger noise, and (iv) 80 projections with noise and ring artifacts. PSNR and SSIM are shown in the top-left and top-right corners. Red and green insets highlight zoomed-in regions.

Figure 19: Visual comparison of all benchmarked methods on the simulated medical CT dataset under four different configurations: (i) 40 projections without noise, (ii) 20 projections with mild noise, (iii) 80 projections with stronger noise, and (iv) 80 projections with noise and ring artifacts. PSNR and SSIM are shown in the top-left and top-right corners. Red and green insets highlight zoomed-in regions.

Parameter	Medical CT	Industrial CT	Synchrotron CT
Input channels	1	1	1
Output channels	1	1	1
Sample size	512(pixel)/128(latent)	512(pixel)/128(latent)	768(pixel)/192(latent)
Activation function	silu	silu	silu
Dropout	0.0	0.0	0.0
Normalization groups	32	32	32

Appendix

Table 12: UNet configurations used for different CT datasets.

Parameter	Medical CT	Industrial CT	Synchrotron CT
Input channels	1	1	1
Output channels	1	1	1
Sample size	512	512	768
Activation function	silu	silu	silu
Block output channels	[128, 256, 512]	[128, 256, 512]	[192, 384, 768]
Layers per block	2	2	2

Appendix

Table 13: VQ-VAE configurations used in the benchmark for different CT datasets.

Parameter	Medical CT	Industrial CT	Synchrotron CT
Image size	512×512	512×512	768×768
Embed dim	128	128	128
Depths	[2, 2, 2, 2]	[2, 2, 2, 2]	[1, 1, 1, 1]
Num heads	[2, 2, 2, 2]	[2, 2, 2, 2]	[1, 1, 1, 1]
MLP ratio	2.0	2.0	1.0
Window size	8	8	8

Appendix

Table 14: SwinIR configurations used in the benchmark for different CT datasets.

Method	Parameters	Grid Search Range	Medical CT	Industrial CT	Synchrotron CT
MCG	step size $\eta$	$[1\times 10^{-4},1\times 10^{3}]$	0.01	0.1	0.01
DPS	step size $\eta$	$[1\times 10^{-4},1\times 10^{3}]$	10	1	0.05
PSLD	factor on latent error $\gamma$		0.2/0.2/0.2/0.1	0.5/0.5/0.5/0.6	0.993
PSLD	factor on masurement consistency error $\omega$	-	$1-\gamma$	$1-\gamma$	$1-\gamma$
PGDM	step size $\eta$	$[1\times 10^{-4},1\times 10^{3}]$	1/0.01/0.01/0.01	1/0.01/0.01/0.01	0.3/0.1/0/1
ReSample	pixel optimization learning rate	[ $1\times 10^{-5}$ ,1]	$1\times 10^{-4}$ / $1\times 10^{-4}$ / $1\times 10^{-3}$ / $1\times 10^{-2}$	$1\times 10^{-4}$	-

Appendix

Table 15: Search ranges and selected hyperparameters for each reconstruction method in the DM4CT benchmark. Parameters are optimized via grid search on ten randomly selected training images by minimizing average mean squared error with respect to reference reconstructions.

Explore similar work

Toward a Foundation Plug-and-Play Prior for Computed Tomography Reconstruction via a Multimodal Diffusion Model
Aug 24, 2026Haley Duba-Sullivan, Patxi Fernandez-Zelaia, Obaidullah Rahman +1Computed Tomography Diffusion Models
Oak Ridge National Laboratory Oak Ridge, TN, USA
Efficient Flow Matching for Sparse-View CT Reconstruction
Feb 27, 2026Jiayang Shi, Lincen Yang, Zhong Li +3Flow Matching Sparse-View CT Reconstruction
Centrum Wiskunde en Informatica · LIACS, Leiden University · Great Bay University +1
Cross-Modal Guidance for Fast Diffusion-Based Computed Tomography
Mar 1, 2026Timofey Efimov, Singanallur Venkatakrishnan, Maliha Hossain +2Computed Tomography Diffusion Model Guidance
Carnegie Mellon University, Pittsburgh, USA · Oak Ridge National Laboratory, Oak Ridge, USA · Purdue University, West Lafayette, USA

Aug 24, 2026J/K move · Enter open · S save

Toward a Foundation Plug-and-Play Prior for Computed Tomography Reconstruction via a Multimodal Diffusion Model

Haley Duba-Sullivan, Patxi Fernandez-Zelaia, Obaidullah Rahman, Amirkoushyar Ziabari

Oak Ridge National Laboratory Oak Ridge, TN, USA

Computed Tomography Diffusion Models Multimodal Diffusion Models

Open paper AskPDF arXiv

Computed tomography (CT) throughput is limited by scan time, which grows with both the number of projections acquired and the detector integration time for each projection. Reconstructing high-quality volumes from sparse-view or low-dose measurements therefore depends on using an informative prior, typically a neural network trained for one specific scan setting and retrained whenever the modality, geometry, or material changes. We investigate whether a single diffusion model trained across several imaging domains can instead serve as a reusable prior for heterogeneous CT reconstruction problems. We evaluate the proposed method using the same diffusion visual transformer model and normalized denoising strength on three datasets that differ in modality, beam geometry, material, and degradation type, spanning additively manufactured metal parts and concrete microstructure imaged with cone-bean X-ray CT and parallel-beam neutron CT respectively. The proposed method improves upon analytic reconstructions in all three cases, demonstrating transferability across the evaluated problems and providing a step toward a reusable foundation prior for heterogeneous CT reconstruction.

Abstract

Explore similar work

Toward a Foundation Plug-and-Play Prior for Computed Tomography Reconstruction via a Multimodal Diffusion Model

Efficient Flow Matching for Sparse-View CT Reconstruction

Cross-Modal Guidance for Fast Diffusion-Based Computed Tomography