Abstract

Medical image analysis faces significant challenges in data sharing due to privacy regulations and complex institutional protocols. Dataset distillation offers a solution to address these challenges by synthesizing compact datasets that capture essential information from real, large medical datasets. Trajectory matching has emerged as a promising methodology for dataset distillation; however, existing methods primarily focus on terminal states, overlooking crucial information in intermediate optimization states. We address this limitation by proposing a shape-wise potential that captures the geometric structure of parameter trajectories, and an easy-to-complex matching strategy that progressively addresses parameters based on their complexity. Experiments on medical image classification tasks demonstrate that our method improves distillation performance while preserving privacy and maintaining model accuracy comparable to training on the original datasets. Our code will be made publicly available.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/3876_paper.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to the Code Repository

https://github.com/Bian-jh/HoP-TM

Link to the Dataset(s)

N/A

BibTex

@InProceedings{DonLe_HighOrder_MICCAI2025,
        author = { Dong, Le and Bian, Jinghao and Hou, Jingyang and Hu, Jingliang and Shi, Yilei and Dong, Weisheng and Zhu, Xiao Xiang and Mou, Lichao},
        title = { { High-Order Progressive Trajectory Matching for Medical Image Dataset Distillation } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
        year = {2025},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15973},
        month = {September},
        page = {280 -- 290}
}


Reviews

Review #1

  • Please describe the contribution of the paper

    This paper focuses on the efficient sharing of medical image data under privacy constraints and proposes a high-order progressive trajectory matching method for dataset distillation. Based on the trajectory matching strategy, the method retains intermediate parameter evolution information during training and introduces a Shape-wise Potential and an Easy-to-Complex Matching Strategy. The effectiveness of the proposed method is validated on two public medical image classification datasets: PathMNIST and COVID19-CXR.

  • 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.This paper proposes and applies several innovative concepts. The shape-wise potential captures parameter dynamics more precisely from an angular perspective, while the easy-to-complex matching strategy alleviates the optimization difficulty caused by hard-to-match parameters. 2.The utility of the synthesized data is evaluated on two public medical datasets, and ablation studies are conducted to verify the effectiveness of the two proposed modules. 3.Cross-model experiments are conducted on the PathMNIST dataset to validate the generalization ability of the proposed method.

  • 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.In the comparative experiments, the method should be compared not only with trajectory matching-based approaches, but also with the latest generative distillation methods, such as those based on GANs or diffusion models. 2.The authors should validate the proposed method on datasets such as CIFAR-10 and Tiny ImageNet to demonstrate its generalization ability. 3.I have some concerns regarding the applicability of the proposed data distillation method to other downstream tasks,such as object detection or semantic segmentation. 4.Although image visualizations are provided, there is no further analysis of the semantic meaning of the synthesized images or their contribution to the final classification. 5.The article does not introduce some important works related to medical data distillation techniques, such as 1) Compressed Gastric Image Generation Based on Soft-Label Dataset Distillation for Medical Data Sharing, and 2) Dataset Distillation for Medical Dataset Sharing.

  • 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

    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.

    (3) Weak Reject — could be rejected, dependent on rebuttal

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

    The paper focuses on the efficient sharing of medical image data under the context of privacy protection and proposes a higher-order progressive trajectory matching method for dataset distillation. However, the paper lacks a literature review and comprehensive experimental validation.

  • Reviewer confidence

    Very confident (4)

  • [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 proposes a high-order trajectory matching framework that leverages a shape-wise potential to capture rich geometric structures of parameter trajectories, and introduces an easy-to-complex matching strategy to progressively align parameters based on their complexity. Experiments demonstrate the effectiveness and state-of-the-art performance of the proposed method.

  • 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. The methods are well-motivated and grounded in clear insights about trajectory optimization.
    2. The paper is well written, with a clear and logical presentation of ideas and experimental results.
  • 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. In the preliminary section, Equations (1) and (2) need further clarification to provide sufficient background on data distillation. For example, the loss L_{ce} in Equation (1) should be explicitly defined. Additionally, it is unclear how D_s is optimized in Equation (2), and whether Equations (1) and (2) are optimized alternately.
    2. Figure 2 should be presented as a vector graphic to improve clarity and readability.
    3. Is there any other way to demonstrate the sufficiency of the distilled data beyond classification performance?
  • 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?

    Overall, the method proposed in this paper is well-motivated and demonstrates novelty in both the shape-wise potential and the easy-to-complex matching strategy. The paper is clearly written and the experimental results are comprehensive. However, some technical details in the preliminary section could be better clarified

  • 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 #3

  • Please describe the contribution of the paper

    This paper aims to improve exising trajectory matching methods for medical image dataset distillation. Based on the observation that exising approaches focus solely on terminal states and parameter matching difficulty varies, two key solutions are presented in this paper: (1) matching trajectories with a high-order potential and (2) progressively matching parameters from easy to complex.

  • 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 present critical observation that existing trajectory matching methods focus solely on terminal states. The motivition to improve existing methods is strong and convincing.

    1. This paper presents a novel solution of high-order trajectory matching to capture geometric structure of parameter trajectories.
    2. The proposed solution achieves state-of-the-art performances on two medical image datasets. And the ablation studies also support the usages of high-order trajectory matching and easy-to-complex matching strategies.
  • 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.

    The methods for comparison are general methods and are originally evaluated on natural image datasets. And the proposed solution seems not to explicitly utilize specific properties of medical images. This paper should add discussion on existing efforts specific to medical images and explanation why the proposed solution works well on medical images.

  • 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
    1. Fig 1 is a bit confusing. The angle is selected at t+2 for real data and t+3 for distilled data, which could be misunderstood as randomly picking the angle instead of picking at middle of the trajectories.
    2. For table 1 and table 2, besides IPC, it would be better to also show the percentage of the original dataset size.
    3. For figure 2, it would be better to show some original dataset images for comprehensive understanding.
  • 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?

    This paper is overall strongly motivated. The proposed solution tackles the issues of existing methods and achieves state-of-the-art performances on two medical image datasets. The major concern is that the properties of medical images should be emphasized in method description and experiment result analysis.

  • 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




Author Feedback

We sincerely thank the reviewers for their thoughtful, detailed, and constructive feedback. We appreciate the time they invested in evaluating our work and will thoroughly address all concerns raised in both the meta-reviews and reviews in the final version.




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”.

    The paper presents a well-written and novel contribution to the field of medical dataset distillation. Its publication would help stimulate further discussion in this emerging area. Therefore, I recommend a Provisional Accept. However, the current version lacks discussion or comparison with several important works related to medical dataset distillation. I strongly encourage the authors to include the following relevant papers in the revised manuscript:

    1)Soft-Label Anonymous Gastric X-ray Image Distillation 2)Compressed Gastric Image Generation Based on Soft-Label Dataset Distillation for Medical Data Sharing 3)Dataset Distillation for Medical Dataset Sharing



back to top