EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

3DAD: Super-Resolution Image Synthesis from Anisotropic CT Images Using a Three-Dimensional Adversarial Diffusion Model

Jianliang Lu, Ho Ming Cheng iD, Benjamin Xin Hao Fang iD, Chun On Anderson Tsang, Sarah Yu, Wai-Kay Seto, Philip Leung Ho Yu iD, Keith Wan-Hang Chiu iD

Bioengineering · 22.05.2026 · 13 (6) · 595

DOI: 10.3390/bioengineering13060595 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Bioengineering, online erschienen am 22.05.2026. Verfasst von Jianliang Lu und 7 weiteren Autor:innen. Beteiligte Einrichtung: Department of Medicine, School of Clinical Medicine, The University of Hong Kong, Hong Kong. Die Arbeit nennt 29 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Research Grant Council, HKSAR Government, General Research Fund, Research Grant Council, HKSAR Government, Innovation and Technology Fund, HKSAR Government.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

High-resolution thin-slice computed tomography (CT) images are often compressed into lower-quality thick-slice images for long-term storage, necessitating synthesis for medical diagnosis. In this paper, we propose a novel 3D adversarial diffusion model (3DAD) for high-fidelity synthesis of thin-slice CT from compressed thick-slice CT. 3DAD is composed of a generator and a discriminator for synthesizing denoised thin-slice images from random noise and source images and distinguishing between noised samples from real and denoised synthetic thin-slice images. Specific models were trained on two-slice to six-slice scenarios for abdominal data, using thick-slice CT compressed from real thin-slice CT as the source. 3DAD was evaluated at the time of HCC diagnosis, at the observation and patient levels, using real thin-slice and synthetic thin-slice CT, with DeLong’s test to compare the similarity of receiver operating characteristic (ROC) curves. We further evaluated 3DAD on real-world data with both thin and thick images, with the synthetic image quality assessed by radiologists and in radiomics feature analysis. Based on the external dataset with 548 samples, the achieved mean squared error (MSE), peak signal-to-noise ratio (PSNR), and structural similarity index measure (SSIM) values were 81.374, 29.478, and 0.916, respectively, for the five-slice scenarios at the portal venous phase. The Areas Under Curves (AUCs) achieved were 0.896 on synthetic thin-slice images compared with 0.889 on real thin-slice images at the observation level (p = 0.028) and 0.854 versus 0.846, correspondingly, at the patient level (p = 0.055). For evaluation on the real-world testing dataset after fine-tuning at the portal venous phase, the MSE, PSNR, and SSIM were 70.435, 30.243, and 0.94, respectively. Radiologist evaluation confirmed the high quality of the synthetic image, with no significant difference in the majority of cases across all five parameters, except for radiologist 2, in realistic and consistent situations, under which at least 41 of 43 synthetic images were assessed as equal to or above grade 3. Our 3DAD enabled the synthesis of thick-slice CT images into high-resolution thin-slice images, facilitating high-fidelity volume image application in HCC diagnosis.

Schlagwörter

Stichwörter der Autor:innen

computed tomography (CT)diffusion modelsyntheticthick slicethin slice

Autor:innen und Institutionen

  • Department of Medicine, School of Clinical Medicine, The University of Hong Kong, Hong Kong.
  • Department of Diagnostic & Interventional Radiology, Hong Kong Sanatorium & Hospital, Hong Kong.
  • Wessex Neurological Centre, Southampton General Hospital, University of Southampton, Southampton SO16 6YD, UK.
  • Department of Medicine, The University of Hong Kong-Shenzhen Hospital, Shenzhen 518000, China.
  • State Key Laboratory of Liver Research, The University of Hong Kong, Hong Kong.
  • Department of Mathematics and Information Technology, The Education University of Hong Kong, Hong Kong.

ORCID-Profile: Ho Ming Cheng, Benjamin Xin Hao Fang, Philip Leung Ho Yu, Keith Wan-Hang Chiu

Förderung

  • Research Grant Council, HKSAR Government
  • General Research Fund, Research Grant Council, HKSAR Government
  • Innovation and Technology Fund, HKSAR Government

Zitieren

APA

Lu, J., Cheng, H. M., Fang, B. X. H., Tsang, C. O. A., Yu, S., Seto, W. K., Yu, P. L. H., & Chiu, K. W. (2026). 3DAD: Super-Resolution Image Synthesis from Anisotropic CT Images Using a Three-Dimensional Adversarial Diffusion Model. Bioengineering, 13(6), 595. https://doi.org/10.3390/bioengineering13060595

Vancouver

Lu J, Cheng HM, Fang BXH, Tsang COA, Yu S, Seto WK, et al. 3DAD: Super-Resolution Image Synthesis from Anisotropic CT Images Using a Three-Dimensional Adversarial Diffusion Model. Bioengineering. 2026;13(6):595. doi:10.3390/bioengineering13060595

BibTeX
@article{Lu2026,
  author = {Lu, Jianliang and Cheng, Ho Ming and Fang, Benjamin Xin Hao and Tsang, Chun On Anderson and Yu, Sarah and Seto, Wai-Kay and Yu, Philip Leung Ho and Chiu, Keith Wan-Hang},
  title = {3DAD: Super-Resolution Image Synthesis from Anisotropic CT Images Using a Three-Dimensional Adversarial Diffusion Model},
  journal = {Bioengineering},
  year = {2026},
  volume = {13},
  number = {6},
  pages = {595},
  doi = {10.3390/bioengineering13060595}
}
Kostenlos starten →