Abstract

Morphological characteristics of vessels serve as important biomarkers for numerous diseases, making accurate vessel segmentation fundamental to computer-aided diagnosis. However, cross-center domain shifts often lead to substantial segmentation performance degradation in real-world deployment. Continual Test-Time Adaptation (CTA) addresses this challenge by updating models online using streams of unlabeled test data, while mitigating error accumulation and catastrophic forgetting. Nevertheless, when applied to 3D vessel segmentation, CTA often introduces topology-breaking artifacts, such as vessel disconnections and spurious branches. To address these challenges, we propose DyCo-CTA, a dynamic collaborative continual test-time adaptation framework tailored for 3D vessel segmentation. Unlike conventional CTA methods with fixed teacher–student roles, we recognize that the teacher is not always more reliable than the student for complex 3D vessel structures, which may introduce bias and reduce adaptation quality. To mitigate this issue, we design a dynamic collaborative network that enables the two models to adaptively exchange teacher–student roles based on their relative reliability. In addition, we introduce a pseudo-break transformation to construct complementary teacher–student inputs that explicitly encourage structural integrity during adaptation. To further preserve global vessel connectivity, we propose a topological regularization loss that aligns topological critical points beyond pixel-level consistency. Extensive experiments on 3D cerebrovascular and cardiovascular segmentation benchmarks demonstrate that DyCo-CTA outperforms state-of-the-art methods, achieving improvements of 3.05% and 2.66%, respectively.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/0111_paper.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to the Code Repository

https://github.com/lixiangcog/DyCo-CTA

Link to the Dataset(s)

N/A

BibTex

@InProceedings{LiXia_Dynamic_MICCAI2026,
        author = { Li, Xiang AND Fang, Yuqi AND Zhang, Dan AND Zhang, Jiong AND Shan, Caifeng},
        title = { { Dynamic Collaborative Continual Test-Time Adaptation for 3D Vessel Segmentation } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 16881},
        month = {September},
        page = {pending}
}


Reviews

Review #1

  • Please describe the contribution of the paper

    The authors propose a dynamic teacher–student framework that incorporates topology-aware constraints and a “pseudo-break transformation” (a form of data augmentation via inpainting) to address the challenge of vessel segmentation under domain shift. The framework is evaluated on six cerebral vessel segmentation datasets and is shown to perform better than models trained only on the source domain or static teacher–student models.

  • 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.
    • Extensive comparison against not only source-only models but also static teacher–student setups
    • An ablation study demonstrating the impact of each component in the framework
    • Framework code is made publicly available
  • 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.

    Mainly related to clarity of explanations:

    • The term “cognitive bias” in the context of the static teacher–student framework is potentially misleading as it seems to refer to an incorrect assumption of the researchers that the teacher should always perform better, which often does not hold, rather than specific perception bias in the teacher model itself.
    • It would be helpful to briefly explain what is meant by “pseudo-break transformations” early ion the text as they are referenced multiple times before Section 2.2.
    • From the text and Figure 1, it is not entirely clear at which points pseudo-break transformations are incorporated into the M1 and M2 model training.
  • 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 has provided an anonymized link to the source code, dataset, or any other dependencies.

  • Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?

    N/A

  • 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/2026/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?

    Experimental results demonstrate that the framework improves performance on vessel segmentation tasks under domain shift.

  • 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 #2

  • Please describe the contribution of the paper

    The paper presents a continual test-time adaptation framework for 3D vessel segmentation that explicitly addresses the structural fragility of vascular topology under domain shift. Its main contribution is a dynamic collaborative teacher-student design, together with pseudo-break augmentation and topology-aware regularization, to improve adaptation robustness and preserve vessel connectivity during online deployment.

  • 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.Continual test-time adaptation for 3D vessel segmentation is an important yet underexplored setting, especially given the practical challenges of domain shift in medical imaging and the structural sensitivity of vascular segmentation. The paper identifies a meaningful gap between existing generic CTA methods and the specific demands of vessel analysis. 2.Continual test-time adaptation for 3D vessel segmentation is an important yet underexplored setting, especially given the practical challenges of domain shift in medical imaging and the structural sensitivity of vascular segmentation. The paper identifies a meaningful gap between existing generic CTA methods and the specific demands of vessel analysis. 3.The pseudo-break transformation is a meaningful way to simulate topology-damaging perturbations and encourages the model to learn structurally robust adaptation. Similarly, the topology regularization goes beyond pixel-level consistency and explicitly constrains topological stability, which is highly relevant for vascular anatomy.

  • 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.While the teacher and student roles are reassigned online based on entropy, the actual optimization in Sec. 2.3 still becomes a one-way supervision scheme within each iteration, where the lower-entropy model is treated as a fixed teacher. Since lower entropy does not necessarily imply a more correct prediction under domain shift, the method may still inherit overconfident teacher errors. This point deserves deeper analysis and clearer empirical validation. 2.The method involves several task-specific settings, such as the binarization threshold for pseudo masks, the dilation setting in Eq. (4), the patch size and Gaussian scale in the pseudo-break transformation, and the persistence threshold used in Eq. (8). While these choices are reasonable, the paper would benefit from a bit more discussion or sensitivity analysis to help readers better understand their impact on performance and robustness. 3.The paper claims improvements in preserving vessel topology and reducing structural errors such as disconnections or false branches. As a minor suggestion, the visual comparison could be further strengthened by highlighting representative regions, so that the qualitative advantages are easier to appreciate.

  • 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 has provided an anonymized link to the source code, dataset, or any other dependencies.

  • Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?

    N/A

  • 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/2026/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 paper addresses a meaningful and underexplored problem in 3D vessel segmentation, and the proposed framework is well aligned with the structural challenges of continual adaptation. My positive score is mainly based on the task relevance, the coherent method design, and the promising empirical results, although some design choices and result presentation could still be further clarified.

  • 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

    This paper studies continual test-time adaptation (CTA) for 3D vessel segmentation under domain shift, a setting that is both practically important and technically challenging because 3D vascular structures are highly topology-sensitive. The authors propose DyCo-CTA, which departs from the conventional fixed teacher-student paradigm by dynamically switching the two roles according to relative prediction reliability. In addition, the paper introduces a pseudo-break transformation to simulate topology-breaking perturbations during adaptation and a topological regularization objective to preserve global vessel connectivity beyond pixel-level consistency. To the best of my knowledge, this is the first work to explicitly formulate and evaluate CTA in the context of 3D vessel segmentation, and the paper demonstrates clear gains on both cerebrovascular and cardiovascular benchmarks.

  • 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.

    Important and timely problem setting. The paper addresses continual test-time adaptation for 3D vessel segmentation, which is highly relevant for real-world medical deployment under cross-center and cross-device distribution shifts. This is a meaningful extension of CTA to a clinically relevant structured segmentation problem. Clear technical novelty. The proposed method is not a minor variant of existing CTA pipelines. The dynamic collaborative mechanism challenges the common assumption that the teacher is always more reliable than the student, which is particularly reasonable in complex 3D vascular segmentation. This is a well-motivated and interesting design choice. Topology-aware formulation is well aligned with the task. The pseudo-break transformation and the topological regularization are both tailored to the structural characteristics of vessels. Rather than relying only on generic entropy minimization or pixel-level consistency, the method explicitly targets connectivity preservation and the reduction of topology-breaking artifacts such as disconnections and spurious branches. Strong empirical evaluation. The paper provides experiments on both cerebrovascular and cardiovascular datasets and reports consistent improvements over a broad range of CTA baselines. The long-term adaptation experiment is also valuable, as it directly addresses error accumulation and catastrophic forgetting, which are central concerns in CTA. Good overall paper quality. The method is conceptually coherent, the ablation study supports the contribution of each component, and the submission also provides an anonymized code link, which further strengthens reproducibility.

  • 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 paper is strong overall, and my concerns are relatively minor. First, the computational overhead and memory/runtime cost of DyCo-CTA are not analyzed in sufficient depth. Since CTA is an online adaptation setting, efficiency is important for practical deployment, especially in 3D medical imaging. Second, although the paper argues that this is the first CTA work for 3D vessel segmentation, the final version could better position itself with respect to prior topology-aware tubular segmentation and topology-enhanced adaptation literature, for example recent topology-preserving tubular segmentation and topology-aware TTA works. A sharper discussion of what is fundamentally new here beyond extending teacher-student CTA to 3D vessel data would make the contribution even clearer. Third, some implementation details could be explained more thoroughly, such as the sensitivity to the pseudo-break construction strength, the choice of the persistence threshold, and the robustness of the dynamic switching criterion. These points mainly affect completeness and reproducibility, rather than the validity of the core idea. Overall, these weaknesses do not undermine the main contribution or the empirical strength of the paper.

  • 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 has provided an anonymized link to the source code, dataset, or any other dependencies.

  • Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?

    N/A

  • 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/2026/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.

    (6) Strong Accept — must be accepted due to excellence

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

    I recommend a strong accept because the paper addresses an important and underexplored problem, proposes a technically coherent and clearly motivated solution, and demonstrates consistent gains across multiple challenging 3D vessel segmentation benchmarks. The combination of dynamic teacher-student role switching, pseudo-break transformation, and topology-aware regularization is well matched to the structural properties of vascular segmentation and goes beyond a routine application of existing CTA ideas. I also appreciate the inclusion of long-term adaptation experiments, which directly evaluate robustness to error accumulation over time. While I would welcome a stronger analysis of computational cost and a more explicit discussion of the relationship to prior topology-aware work, these issues are secondary and do not detract from the main contribution. Overall, I view this paper as a high-quality and impactful contribution that should be accepted.

  • 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



Author Feedback

We thank the reviewers and the Area Chair for their time and helpful comments. The comments on terminology, method clarification, implementation details, sensitivity analysis, computational cost, and relation to prior topology-aware work are very useful. We will revise the final version carefully to address these points and make the paper clearer and more complete. Thank you again for the constructive feedback and for supporting our work.




Meta-Review

Meta-review #1

  • Your recommendation

    Provisional Accept

  • Please justify your decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. In case of an invitation for rebuttal, clarify which points are important to address in the rebuttal.

    This paper addresses an important and underexplored problem, proposes a technically coherent and clearly motivated solution, and demonstrates consistent gains across multiple challenging 3D vessel segmentation benchmarks. The work is technically solid and likely to attract interest from researchers working on test-time adaptation, domain shift, and structure-preserving medical image segmentation.



back to top