Abstract

Source-free domain adaptation (SFDA), where only a pre-trained source model is available to adapt to the target domain, has gained widespread application in the medical field. Most existing methods overlook low-quality pseudo-labels, i.e., pseudo-labels with boundary semantic confusion, when learning target domain-specific knowledge, leading to the loss of crucial boundary information. Furthermore, focusing solely on the specific knowledge can drive the model shifts in an uncontrollable direction, resulting in model degradation. To address these issues, we propose Dual Knowledge-aware Guidance (DKG), a novel SFDA method that integrates domain-specific knowledge with domain-invariant knowledge to improve transfer performance. Specifically, the pseudo-label calibration scheme is proposed to reduce semantic bias in high-uncertainty pixels, preserving the boundary information of target domain-specific knowledge. To ensure stable training, we propose a domain-invariant knowledge-based loss strategy, leveraging a confidence-guided mechanism and a consistency constraint. Additionally, we also introduce a dynamic balancing loss to address class imbalance. Extensive experiments on cross-domain fundus image segmentation show that DKG achieves state-of-the-art performance. Code is available at https://github.com/Hanshuqian/DKG.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/1789_paper.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to the Code Repository

https://github.com/Hanshuqian/DKG

Link to the Dataset(s)

N/A

BibTex

@InProceedings{CheYu_Dual_MICCAI2025,
        author = { Chen, Yu and Wang, Hailing and Wu, Chunwei and Cao, Guitao},
        title = { { Dual Knowledge-Aware Guidance for Source-Free Domain Adaptive Fundus Image Segmentation } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
        year = {2025},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15965},
        month = {September},
        page = {185 -- 195}
}


Reviews

Review #1

  • Please describe the contribution of the paper

    The authors propose a Dual Knowledge-Aware Guidance (DKG) SFDA method for the segmentation of the optic cup and optic disc. This includes an Anchor Point-Driven Pseudo-Label Calibration mechanism and a domain-invariant knowledge-based loss, as well as a balancing loss designed to learn domain-specific knowledge. Experiments demonstrate the effectiveness of DKG, surpassing classic medical SFDA methods such as DPL.

  • 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 authors propose a dual knowledge-aware guidance framework, which introduces a collaborative modeling strategy that incorporates both domain-specific knowledge and domain-invariant knowledge. This design aims to capture intra-class consistency within the target domain and enhance cross-domain generalization, thereby improving model robustness and adaptability under pseudo-label supervision. In addition, the proposed Anchor Point-Driven Pseudo-Label Calibration module leverages the concept of nearest anchor point consistency, combined with local feature similarity, to effectively alleviate boundary ambiguity and reduce uncertainty accumulation in the pseudo labels.

  • 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 authors propose a “Dual Knowledge-Aware Guidance” (DKG) framework to address pseudo-label boundary ambiguity and model degradation in source-free medical image segmentation, demonstrating promising performance. However, DKG still has the following limitations:

    1.The invariant knowledge-aware loss consists of two components: the BIM loss and the prototype distance-based KL loss. Since both rely on uncertainty and feature similarity, there may be overlapping optimization objectives. It is recommended to conduct ablation studies on each loss component independently to verify whether their combination provides complementary benefits.

    2.The current experiments assume a fixed amount of target domain training data. The robustness of the model under varying data scales has not been analyzed. A sensitivity study (e.g., using 25% to 100% of target data) is suggested to examine DKG’s performance stability and boundary of effectiveness under limited data conditions.

    3.The proposed method is only validated on the optic disc/cup binary segmentation task. It remains unclear whether the approach is effective for more complex scenarios, such as multi-structure or multi-disease segmentation (e.g., macular lesions or diabetic retinopathy grading).

    4.Various anchor-based pseudo-label refinement methods already exist. To better highlight the advantage of the proposed anchor mechanism in terms of structural accuracy and boundary localization, comparisons with existing anchor-based approaches are recommended.

    5.The method involves uncertainty estimation (e.g., multiple forward passes) and prototype matching, which may introduce additional computational overhead. It would be helpful to discuss whether this affects inference speed and deployment feasibility in real clinical settings.

    6.To improve the reproducibility of the results and benefit the community, it is strongly recommended that the authors release the source code.

  • 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 submission does not mention open access to source code or data but provides a clear and detailed description of the algorithm to ensure reproducibility.

  • 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?

    This paper proposes a novel medical image segmentation method—Dual Knowledge-Aware Guidance (DKG)—designed for the source-free domain adaptation setting where source data is unavailable. The method is conceptually sound, structurally concise, and supported by comprehensive experiments. It demonstrates strong effectiveness, particularly in addressing pseudo-label boundary ambiguity and model degradation. Given the current research landscape of source-free domain adaptation, this work holds practical significance and application potential.

  • 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
    1. This paper reveals that high-accuracy samples preserve more domain-shared features, facilitating the learning of domain-specific features. This investigation can benefit the DA problem.
    2. The authors proposed the Anchor Point-Driven Pseudo-Label Calibration scheme, to effectively capture the semantic boundary information of pseudo-labels.
    3. The experiments on three fundus segmentation datasets show the advantage 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 authors investigate that high-accuracy samples preserve more domain-shared features, facilitating the learning of domain-specific features with provided evidence. This investigation is interesting.

    2. The authors consider semantic distribution consistency between reference regions (anchor points) and high-uncertainty pixels in high-dimensional feature space for constructing representative anchors. This idea is reasonable.

    3. The improvement of the proposed method over the existing works is significant, especially for the Optic Cup Segmentation dataset.

  • 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. The improvement on Optic Disc Segmentation is very marginal. More analysis is needed.
    2. The Anchor Point Refinement mechanism is not new, lacks a discussion with the existing work that focuses on learning representative anchors [1,2,3].

    [1] Beyond prototypes: Semantic anchor regularization for better representation learning [2] Exploring prototype-anchor contrast for semantic segmentation [3] Geometry-Aware Guided Loss for Deep Crack Recognition

  • 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 submission does not mention open access to source code or data but provides a clear and detailed description of the algorithm to ensure reproducibility.

  • 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?

    Although this paper lacks a comprehensive discussion on the existing works in terms of anchor learning, overall, this paper is well-written with a clear contribution and solid results. Thus, I recommend acceptance.

  • 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 #3

  • Please describe the contribution of the paper

    The paper presents a novel source-free domain adaptation method that employs a calibration scheme to mitigate semantic bias in high-uncertainty pixels, ensuring comprehensive capture of target domain knowledge. Experimental results demonstrate state-of-the-art performance.

  • 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 paper is well-written and clearly explains the proposed method. It addresses the limitations of state-of-the-art approaches by introducing an Anchor Point-Driven Pseudo-Label Calibration scheme to effectively capture the semantic boundary information of pseudo-labels, and by improving training stability through a loss strategy based on domain-invariant knowledge.

  • 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. The text in Figures 2 and 3 is barely legible.
    2. The dataset descriptions lack information about image resolution, which is a critical factor in evaluating segmentation accuracy.
    3. The paper focuses exclusively on the fundus segmentation task and does not investigate the method’s applicability to other medical imaging tasks.
  • 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 submission does not mention open access to source code or data but provides a clear and detailed description of the algorithm to ensure reproducibility.

  • 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

    Please refer the comments of No. 7.

  • 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 method is clearly presented, with well-justified design choices. The experiments demonstrate its effectiveness through both quantitative and qualitative evaluations, highlighting the method’s efficiency and suitability for computer-aided image processing.

  • 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

N/A




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