Cone-Beam CT Image Quality Enhancement Using A Latent Diffusion Model Trained with Simulated CBCT Artifacts

Naruki Murahashi; Mitsuhiro Nakamura; Megumi Nakao

eess.IVMar 27, 2026

Cone-Beam CT Image Quality Enhancement Using A Latent Diffusion Model Trained with Simulated CBCT Artifacts

Authors: Naruki Murahashi, Mitsuhiro Nakamura, Megumi Nakao

Organizations: Graduate School of Medicine, Kyoto University, Kyoto, 606-8507, JAPAN

Abstract

Cone-beam computed tomography (CBCT) images are problematic in clinical medicine because of their low contrast and high artifact content compared with conventional CT images. Although there are some studies to improve image quality, in regions subject to organ deformation, the anatomical structure may change after such image quality improvement. In this study, we propose an overcorrection-free CBCT image quality enhancement method based on a conditional latent diffusion model using pseudo-CBCT images. Pseudo-CBCT images are created from CT images using a simple method that simulates CBCT artifacts and are spatially consistent with the CT images. By performing self-supervised learning with these spatially consistent paired images, we can improve image quality while maintaining anatomical structures. Furthermore, extending the framework of the conditional diffusion model to latent space improves the efficiency of image processing. Our model was trained on pelvic CT-pseudo-CBCT paired data and was applied to both pseudo-CBCT and real CBCT data. The experimental results using data of 75 cases show that with our proposed method, the structural changes were less than 1/1000th (in terms of the number of pixels) of those of a conventional method involving learning with real images, and the correlation coefficient between the CT value distributions of the generated and reference images was 0.916, approaching the same level as conventional methods. We also confirmed that the proposed framework achieves faster processing and superior improvement performance compared with the framework of a conditional diffusion model, even under constrained training settings.

Figures & tables

Fig. 1: Difference between CT and CBCT images. CBCT images are classified into two main types according to image characteristics: Type 1 is characterized by decreased CT values, while Type 2 is marked by poor contrast. (a) CT image, (b) CBCT image, (c) artifact location on CBCT image. The arrows indicate radial artifacts, and CT values are reduced around the blue dashed line and in the area of the red line.

Fig. 2: Procedure for creating a pseudo-CBCT image. Background colors indicate the assumed factors of artifacts modeled. The processes shown in red boxes have several parameters that modify the image features. The generated images have the imaging features found in the real CBCT images, while their anatomical structures are consistent with those of the original CT images.

Fig. 3: Three examples of real and pseudo-CBCT images. The arrows indicate image characteristics common to both images.

Fig. 4: Structure and data flow of the proposed method. Pretrained VQVAE is used for the encoder $\mathcal{E}$ and the decoder $\mathcal{D}$ .

Fig. 5: Enlarged coronal views of volumes generated with different initial noise settings. (a) Different initial noise for each slice, (b) The same initial noise for all slices. Using different initial noise introduces inter-slice discontinuities.

	MAE [HU]	SSIM
VQVAE (256)	$2.70$	$0.998$
VQVAE (128)	$6.08$	$0.988$
VQVAE (64)	$12.0$	$0.961$

TABLE I: Reconstruction accuracy of VQVAE at each compression factor.

Fig. 6: Comparison of reconstruction results of VQVAEs. The upper panel shows the original image and the reconstructed image at each compression factor. The lower panel provides magnified regions for assessment of fine detail.


			Computation time
	MAE [HU]	SSIM	Training	Generation
			[min/epoch]	[sec/slice]
Pseudo CBCT	$74.0$	$0.902$
Palette	$69.0$	$0.684$	$52$	$140$
CLDM (256)	$19.5$	$\bf{0.955}$	$15$	$40$

TABLE II: Comparison of quantitative evaluation results for image quality improvement of pseudo-CBCT images and computation time.

Fig. 7: Comparison of image quality improvement for pseudo-CBCT images. The display window range was set to [ $-$ 1000 HU, 1000 HU] for case 001, and [ $-$ 300 HU, 150 HU] for cases 005 and 006. Red arrows indicate residual artifacts, and blue arrows show examples of blurred areas. Except for CLDM (128), there are some errors in CT values and insufficient improvement.

Fig. 8: Improvement results for real CBCT images and color maps of overcorrected areas. The annotations and arrows indicate residual artifacts or structural changes. First column: the real CBCT image of each case. Second to fourth columns: images generated by the comparison method. Fifth column: images generated by the proposed method. Sixth and seventh columns: areas of overcorrection with absolute error of 600 HU or more between the real CBCT and the generated image.


	RMSE	Number of	Mean $\pm$ SD	Correlation	Average within ROI [HU]
	[HU]	errors [pixels]	[HU]	coefficient	ROI1	ROI2	ROI3	ROI4	ROI5
Real CBCT	-	-	$-170.2\pm 262$	$0.421$	$-169.5$	$887.1$	$177.9$	$64.1$	$-97.3$
GAN-based (Real)	$26.2$	$598$	$-87.0\pm 306$	$0.908$	$-58.0$	$977.0$	$489.0$	$26.1$	$8.5$
GAN-based (Pseudo)	$1.21$	$\bf{3}$	$-72.5\pm 272$	$0.822$	$-42.1$	$947.0$	$255.8$	$67.9$	$-109.4$
CLDM (Real)	$112$	$3919$	$-81.9\pm 315$	$0.934$	$-97.1$	$198.9$	$69.5$	$-972.3$	$-984.7$

TABLE III: Quantitative evaluation results for improvement of the image quality of real CBCT images. Average values of eight cases for evaluating structural changes (errors) between real CBCT and synthetic CT (root mean squared error and number of errors in the field of view) and evaluating the distribution of CT values in the entire image (mean

\pm

standard deviation, correlation coefficient). Examples of average CT values in the region of interest.

Fig. 9: Example histogram distributions of CT values for real CBCT, CLDM (Pseudo), CLDM (Real), and reference CT.

Fig. 10: Comparisons of image quality improvement of real CBCT images when changing the pseudo-CBCT image creation procedure. The display window range was set to [ $-$ 300 HU, 150 HU]. In the images obtained by methods other than the proposed method, the CT values were reduced and artifacts remained in the areas indicated by the arrows.


	Mean $\pm$ SD [HU]	Correlation coefficient
Real CBCT	$-170.2\pm 262$	$0.421$
w/o warp	$-69.6\pm 278$	$0.839$
w/o contrast	$-67.8\pm 288$	$0.862$
w/o mask1	$-119.5\pm 274$	$0.682$
w/o mask2	$-91.5\pm 283$	$0.903$

TABLE IV: Results of the evaluation of CT value distributions (mean

\pm

standard deviation, correlation coefficient) for different procedures of the pseudo-CBCT image creation process.

Naruki Murahashi received the bachelor’s degree in human health sciences from Kyoto University, Kyoto, Japan, in 2024, and the master’s degree in human health sciences in 2026. She is currently pursuing the Ph.D. degree with the Department of Human Health Sciences. Her research interests include biomedical engineering and machine intelligence in clinical medicine.

Table 15

Mitsuhiro Nakamura received the Ph.D. degree in medicine from Kyoto University, Kyoto, Japan, in 2010. He is currently a Professor with the Medical Physics Laboratory, Graduate School of Medicine, Kyoto University. His research interests include medical physics and high precision radiotherapy. His awards include the Young Scientist Award in Medical Physics of the International Union of Pure and Applied Physics (IUPAP), International Organization for Medical Physics, in 2019.

Table 16

Megumi Nakao received the Ph.D. degree in informatics from Kyoto University, Kyoto, Japan, in 2003. He was with the Graduate School of Medicine, Kyoto University, and the Graduate School of Information Science, Nara Institute of Science and Technology, Nara, Japan. He is currently a Professor with the Biomedical Engineering and Intelligence Laboratory, Graduate School of Medicine, Kyoto University. His research interests include biomedical engineering and machine intelligence in clinical medicine.

Table 17

Cone-Beam CT

Cone-beam computed tomography (CBCT) reconstructs three-dimensional anatomy from X-ray projections for dental, interventional, and radiotherapy imaging.

See all 47 papers

Anatomy-preserving unpaired cone-beam CT refinement for image-guided radiotherapy using pseudo-label guided diffusion
Oct 5, 2026Qi Lai, Yutong HeCone-Beam CT Diffusion Guidance
Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, 518055,China. · Southern university of science and technology, Shenzhen, 518055,China.
EPC-3D-Diff: Equivariant Physics Consistent Conditional 3D Latent Diffusion for CBCT to CT Synthesis
May 19, 2026Alzahra Altalib, Chunhui Li, Haytham Ahmad Alewaidat +3Physics-Informed Diffusion Models Cone-Beam CT
School of Science and Engineering, University of Dundee UK · Faculty of Applied Medical Sciences, Jordan University of Science and Technology · Experia Healthcare, Jordan +1
Toward CT-Equivalent Image Quality in Low-Dose Radiotherapy Planning: Conditional Diffusion-Based CBCT-to-CT Synthesis and the Impact of CBCT Input Representation
Aug 9, 2026Alzahra Altalib, Chunhui Li, Christopher Hamill Taylor +2Cone-Beam CT Medical Image Synthesis
Faculty of Applied Medical Sciences, Jordan University of Science and Technology, Irbid, Jordan · School of Science and Engineering, University of Dundee, Scotland, UK · NHS Ninewells Dundee, Scotland, UK +1

Oct 5, 2026J/K move · Enter open · S save

Anatomy-preserving unpaired cone-beam CT refinement for image-guided radiotherapy using pseudo-label guided diffusion

Qi Lai, Yutong He

Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, 518055,China. · Southern university of science and technology, Shenzhen, 518055,China.

Cone-Beam CT Diffusion Guidance Unpaired Image-to-Image Translation

Open paper AskPDF arXiv

Cone-beam computed tomography (CBCT) is widely used in image-guided radiotherapy, but scatter, beam hardening, noise, truncation, and other artifacts limit image quality and CT number accuracy. Paired CBCT and CT data are difficult to obtain clinically because of motion, anatomical changes, and acquisition mismatch. We present RefineCBCT, an unpaired CBCT refinement framework that uses pseudo-label guidance and short-step diffusion to reduce artifacts while preserving patient-specific anatomy. RefineCBCT was trained and evaluated on unpaired CBCT and planning CT data from public LUNG TCIA and PELVIC TCIA datasets and compared with representative GAN and diffusion based methods. On LUNG TCIA, it achieved the best results across all metrics, with MAE 19.411, RMSE 62.758, PSNR 30.845 dB, and SSIM 0.931. On PELVIC TCIA, it achieved the best MAE, PSNR, and SSIM, with values of 14.905, 36.671 dB, and 0.876. The refined images showed fewer streaking and shading artifacts, clearer anatomical boundaries, and improved soft tissue uniformity, with line profile and ROI analyses showing closer agreement with planning CT. These results suggest that RefineCBCT provides efficient and effective CBCT refinement under clinically realistic unpaired training conditions and may support more reliable CBCT use in image-guided radiotherapy workflows. Code is publicly available on GitHub, and the evaluated datasets are available from The Cancer Imaging Archive.

Abstract

Cone-Beam CT

Explore similar work

Anatomy-preserving unpaired cone-beam CT refinement for image-guided radiotherapy using pseudo-label guided diffusion

EPC-3D-Diff: Equivariant Physics Consistent Conditional 3D Latent Diffusion for CBCT to CT Synthesis

Toward CT-Equivalent Image Quality in Low-Dose Radiotherapy Planning: Conditional Diffusion-Based CBCT-to-CT Synthesis and the Impact of CBCT Input Representation