EUVIMEDAvasta EUVIMED
Open access · CC BY 4.0

Knowledge‐distilled diffusion models for improving cone‐beam CT image quality with meta‐learning under imbalanced data

Joonil Hwang, Sangjoon Park, Seungryong Cho, Jin Sung Kim

Medical Physics · 2026-08 · 53 (8)

DOI: 10.1002/mp.70530 ↗

At a glance

Journal article in Medical Physics, published online on 2026-07-31. Written by Joonil Hwang and 3 co-authors. Institution: Department of Radiation Oncology, Yonsei Cancer Center, Heavy Ion Therapy Research Institute, Yonsei University College of Medicine, Seoul, South Kor…. Major MEDLINE topics: Cone-Beam Computed Tomography, Image Processing, Computer-Assisted, Machine Learning. 26 references. The full text is freely available (open access, CC BY 4.0). Funded by: Yonsei University College of Medicine, Ministry of Health & Welfare, Republic of Korea, National Research Foundation of Korea.

AbstraktSubjectsBibliograafilised andmedCiteRelated studies

Abstract (original language)

Background

Adaptive radiation therapy (ART) relies on daily cone‐beam CT (CBCT), yet its limited image quality hinders accurate dose calculation, particularly under substantial anatomical changes.

Purpose

To overcome the clinical challenge of scarce paired planning CT (pCT) data versus abundant unpaired CBCTs, we propose a framework driven by two core components: knowledge distillation and gradient‐based meta‐guidance.

Methods

The knowledge distillation strategy enables the model to leverage the vast unpaired dataset. Complementing this, the meta‐guidance mechanism stabilizes training by dynamically updating the weight of each unpaired sample; it assigns higher importance to pseudo‐labels that align with trusted supervised gradients, effectively filtering out noise. Our method was evaluated on a cohort of 99 breast cancer patients (with 19 reserved for testing) and further evaluated on a public dataset to assess the generalization capability.

Results

The proposed approach demonstrated superior performance, achieving the best quantitative metrics (MAE 13.22 HU, SSIM 0.9516, PSNR 30.35 dB). It significantly outperformed representative supervised, unsupervised, and standard distillation baselines (). Ablation studies confirm that our method enhances image quality while preserving the patient's daily anatomy by minimizing feature hallucination.

Conclusions

By uniquely combining knowledge distillation with meta‐guidance, our method advances the frontier of high‐quality synthetic CT, enabling more robust and adaptive ART workflows.

Subjects

MeSH terms (MEDLINE)

Cone-Beam Computed TomographyImage Processing, Computer-AssistedMachine LearningHumansDiffusionQuality ControlBreast Neoplasms

Author keywords

cone‐beam computed tomographydiffusion modelknowledge distillationmeta‐guidance

Authors and institutions

  • Department of Radiation Oncology, Yonsei Cancer Center, Heavy Ion Therapy Research Institute, Yonsei University College of Medicine, Seoul, South Korea.
  • Yonsei Institute for Digital Health, Yonsei University, Seoul, South Korea.
  • Department of Nuclear and Quantum Engineering, KAIST, Daejeon, South Korea.
  • Medical Image and Radiotherapy Lab (MIRLAB), KAIST, Daejeon, South Korea.
  • Oncosoft, Seoul, South Korea.
  • Medical Physics and Biomedical Engineering Lab (MPBEL), Yonsei University College of Medicine, Seoul, South Korea.

Funding

  • Yonsei University College of Medicine
  • Ministry of Health & Welfare, Republic of Korea
  • National Research Foundation of Korea

Cite

APA

Hwang, J., Park, S., Cho, S., & Kim, J. S. (2026). Knowledge‐distilled diffusion models for improving cone‐beam CT image quality with meta‐learning under imbalanced data. Medical Physics, 53(8). https://doi.org/10.1002/mp.70530

Vancouver

Hwang J, Park S, Cho S, Kim JS. Knowledge‐distilled diffusion models for improving cone‐beam CT image quality with meta‐learning under imbalanced data. Medical Physics. 2026;53(8). doi:10.1002/mp.70530

BibTeX
@article{Hwang2026,
  author = {Hwang, Joonil and Park, Sangjoon and Cho, Seungryong and Kim, Jin Sung},
  title = {Knowledge‐distilled diffusion models for improving cone‐beam CT image quality with meta‐learning under imbalanced data},
  journal = {Medical Physics},
  year = {2026},
  volume = {53},
  number = {8},
  doi = {10.1002/mp.70530}
}
Start for free →