Paper Info Reviews Author Feedback Meta-Review Back to top
List of Papers Browse by Subject Areas Author List

Abstract

Unpaired Cone-beam CT (CBCT)-to-CT translation is pivotal for radiotherapy planning, aiming to synergize CBCT’s clinical practicality with CT’s dosimetric precision. Existing methods, limited by scarce paired data and registration errors, struggle to preserve anatomical fidelity—a critical requirement to avoid incorrect diagnosis and inadequate treatments. Current CycleGAN-derived approaches risk structural distortions, while diffusion models oversmooth high-frequency details vital for dose calculation in the reverse diffusion. In this paper, we propose the Anatomy-Conserving Schrödinger Bridge (ACSB), a novel unpaired medical image translation framework leveraging entropy-regularized optimal transport to disentangle modality-specific artifacts from anatomy. We incorporate a carefully designed generator, Anatomy-Conserving vision transformer (AC-ViT) to integrate multi-scale anatomical priors via attention-guided feature fusion. We further adopt frequency-aware optimization targeting radiotherapy-critical spectral components. Extensive experiments on the dataset demonstrate the superiority of the proposed ACSB, showcasing excellent generalization over different anatomically distinct regions.Code: https://github.com/Lalala-iks/ACSB

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/5303_paper.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to the Code Repository

https://github.com/Lalala-iks/ACSB

Link to the Dataset(s)

N/A

BibTex

@InProceedings{ShiKe_AnatomyConserving_MICCAI2025,
        author = { Shi, Ke and Ouyang, Song and Liu, Gang and Luo, Yong and Su, Kehua and Liang, Zhiwen and Du, Bo},
        title = { { Anatomy-Conserving Unpaired CBCT-to-CT Translation via Schrödinger Bridge } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
        year = {2025},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15963},
        month = {September},

}

Reviews

Review #1

  • Please describe the contribution of the paper

    The paper proposes a method for generating CT images from CBCT images. The method addresses anatomical structure preservation via a modified interpolation mechanism from CBCT to CT. Additionally, the model is trained using a weighted loss that includes: frequency awareness loss, optimal transport loss, semantic information loss (InfoNCE), and an adversarial discriminator loss. The performance is evaluated on a private dataset consisting of paired CT and CBCT images from the chest and head & neck (H&N) regions. The evaluation is done within the same regions (e.g., chest -> chest) and across regions (e.g., chest -> H&N). The proposed method outperforms the baselines, especially in the cross-region evaluation.

  • Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.

    1. Strong performance: Benchmarks are well chosen, including classical methods like Pix2Pix and CycleGAN, specialized methods like ResViT and SynDiff, and a similar method to the proposed one (UNSB). The benchmarks are outperformed in the cross-region evaluation (chest -> H&N and H&N -> chest).

    2. Clinically well-motivated method: This approach is highly valued by clinicians, as CBCT scans offer lower radiation exposure and reduced cost compared to CT scans, but lack the same level of dosimetric accuracy.

  • Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
    1. Details on datasets are missing. For example:
      • Patient demographics
      • Settings of CT and CBCT
      • Pre-processing of the images

    2. No statistial significance or confidence measure was reported for the performance metrics.

    3. No details on tuned hyperparameter choices and sensitivity: The method is complex, balancing multiple losses. Optimizing such frameworks can be very time consuming. It would therefore be desirable to mention which hyperparameters have been tested for the proposed method versus the benchmarks.

    4. The claim “We pioneer the adaptation of Schrödinger Bridge theory” appears too strong. The primary modification lies in the interpolation mechanism (IPM), yet no theoretical justification or motivation is provided for why this adaptation would be beneficial. Furthermore, the adapted IPM is not compared to the conventional version in the ablation study.
  • Please rate the clarity and organization of this paper

    Satisfactory

  • Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.

    The authors claimed to release the source code and/or dataset upon acceptance of the submission.

  • Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2025/en/REVIEWER-GUIDELINES.html

    Missing reference for positional encoding: “we adopt spectral-aware positional embedding based on FFT to suppress grid artifacts and preserve key diagnostic details.”

    A few typos:

    • “generalization capability is visually in Fig. 3,” -> visualized
    • “synergizing spectral grouding” -> grouping or what is meant here ?
    • “Despite these above, …” -> Despite these advances, …
  • Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.

    (4) Weak Accept — could be accepted, dependent on rebuttal

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    The proposed method yields strong performance on cross-region evaluation and is clinically well-motivated. However, the reproducibility is limited by the use of an in-house dataset with limited information shared about acquisition and demographics.

  • Reviewer confidence

    Confident but not absolutely certain (3)

  • [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.

    N/A

  • [Post rebuttal] Please justify your final decision from above.

    N/A



Review #2

  • Please describe the contribution of the paper

    This paper introduces a novel method for unpaired CBCT-to-CT translation using optimal transport theory.

  • Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.

    The use of Schrodinger Bridge theory for this task is mathematically grounded. Anatomy conservation and frequency-aware optimization are also good choices for this task. The use of dual-phase evaluation protocol is notable. Visuals like Fig. 1 effectively explains the main idea.

  • Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.

    Both datasets are from the same institution, so it is difficult to say if the model can generalize to different vendors or protocols. It would be beneficial to add standard errors in the tables, especially when the numbers are close. Otherwise, it is unclear if the improvements are noise or indeed statistically significant.

  • Please rate the clarity and organization of this paper

    Good

  • Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.

    The authors claimed to release the source code and/or dataset upon acceptance of the submission.

  • Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2025/en/REVIEWER-GUIDELINES.html

    N/A

  • Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.

    (4) Weak Accept — could be accepted, dependent on rebuttal

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    The method is technically sound and it integrates several components from image-to-image translation. Although the empirical gains can be explained in a more convincing way (e.g., statistical tests), the paper could be accepted dependent on rebuttal.

  • Reviewer confidence

    Somewhat confident (2)

  • [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.

    N/A

  • [Post rebuttal] Please justify your final decision from above.

    N/A



Review #3

  • Please describe the contribution of the paper

    Authors proposed the Anatomy-Conserving Schrodinger Bridge (ACSB), a novel method for unpaired CBCT-to-CT translation. Schrodinger Bridge framework directly maps CBCT images to CT without acquiring paired images, which preserve anatomical structures. Interpolation mechanism gradually refines intermediate image states, balancing structure preservation and noise. Anatomy-conserving vision transformer (AC-ViT) incorporates multi-scale anatomical features to maintain crucial details in generated CT images.

  • Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.

    This paper introduces ACSB which is the first to adapt Schrodinger Bridge optimal transport for unpaired medical image translation. The introduction of AC-ViT is a substantial methodological innovation, preserving essential anatomical details across different scales. It enhances the fidelity of synthetic CT images, making it highly suitable for precise clinical applications. Evaluations on multiple anatomical regions and quantitative comparisons demonstrate consistent and significant improvements and the paper convincingly demonstrates generalization and clinical relevance.

  • Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.

    Authors do not provide detailed runtime analysis or discuss computational practicality for real-time workflow integration. Clinical assessments or evaluations on actual patient treatments would strengthen the real-world impact claims.

  • Please rate the clarity and organization of this paper

    Good

  • Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.

    The authors claimed to release the source code and/or dataset upon acceptance of the submission.

  • Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2025/en/REVIEWER-GUIDELINES.html

    N/A

  • Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.

    (5) Accept — should be accepted, independent of rebuttal

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    The proposed model is novel and innovative for unpaired CBCT-to-CT image translation. The use of the Schrodinger Bridge framework combined with the proposed AC-ViT, IPM, and frequency-aware optimization shows a substantial methodological advancement. The robust quantitative evaluation across multiple anatomical sites clearly demonstrates superior performance compared to strong baseline methods.

  • Reviewer confidence

    Somewhat confident (2)

  • [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.

    N/A

  • [Post rebuttal] Please justify your final decision from above.

    N/A



Author Feedback

We greatly thank all the reviewers for their constructive comments on this work. We strive to keep the response concise. Response to Reviewer #1 W1: Runtime analysis and clinical validation. We sincerely appreciate this suggestion. Our current focus was establishing the foundational ACSB framework to ensure anatomical fidelity and cross-domain generalization in CBCT-to-CT translation, as evidenced by our extensive quantitative evaluations (Tables 1-2) and visual comparisons (Figs 2-3) across intra-/corss-region scenarios.Future work will benchmark computational efficiency (inference time&GPU memory) and collaborate with clinics to validate ACSB’s impact on radiotherapy planning workflows. Response to Reviewer #2 W1: Dataset details. Thank you for highlighting the need for clearer dataset description. Due to the length limitation, we simplified the description. (1) H&N dataset: 40 patients (2021-2023). CT (GE Discovery, 512×512, 0.625×0.625 mm², 2.75 mm slice thickness); CBCT (Elekta XVI). (2) Chest dataset: 49 patients (same acquisition time as H&N), CT (GE Discovery, 512×512, 0.625×0.625 mm², 5 mm slice thickness); CBCT (Elekta XVI). (3) Preprocessing. We leverage the open-source Advanced Normalization Tools (ANTs) for affine registration to ensure alignment between each CBCT and CT pair for model testing purposes. The image data from both datasets are transformed using a uniform distribution of Hu values of [-1000, 2200]. W2: Statistical significance. We appreciate this critical feedback. While rebuttal guidelines restrict adding new experimental results, we emphasize that all metrics in Tables 1-2 were averaged over 3 independent runs, with standard deviations <0.5% . Our code release upon acceptance will enable full reproducibility. W3: Hyperparameter analysis. We thank the reviewer for raising this concern. While deeper hyperparameter exploration could further optimize performance, our priority here was to demonstrate the framework’s viability. To ensure reproducibility and alignment with prior works (e.g., UNSB [15]), the loss weights (λot=0.1, λadv=1, λpatchNCE=1) were inherited from established baselines to maintain consistency in framework validation. The newly introduced loss (λfreq=1) was empirically set to balance its impact, as validated by our ablation study (Table 3). W4: IPM analysis. We thank the reviewer for this critique. Our IPM modifies the SB framework by integrating anatomy-aware interpolation (Eqs. 2-3), which explicitly balances noise injection and anatomical preservation. This is theoretically justified via the entropy-regularized OT formulation. We’ll revise “pioneer” to “novel adaptation” to better reflect incremental contributions. Additional: We appreciate these detailed comments. We’ll carefully revise the article, add necessary references and descriptions to improve readability, and carefully correct grammar errors. Response to Reviewer #3 W1: Single-institution data. We appreciate for raising this concern. The current single-institution origin was intentionally chosen to control for protocol variability during the framework’s initial technical validation, allowing to isolate the impact of anatomical fidelity and cross-region generalization(Section 3.2). According to the rebuttal guidelines, it is not allowed to add new experimental results. Nevertheless, we ensure that all metrics in Table1-2 were averaged over 3 independent runs, with standard deviations <0.5% . Our code release upon acceptance will enable full reproducibility.



Meta-Review

Meta-review #1

  • Your recommendation

    Provisional Accept

  • If your recommendation is “Provisional Reject”, then summarize the factors that went into this decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. You do not need to provide a justification for a recommendation of “Provisional Accept” or “Invite for Rebuttal”.

    N/A



back to top