Abstract

Accurate pancreas segmentation is critical for early cancer diagnosis, where annotation scarcity necessitates Semi-Supervised Learning (SSL). However, due to significant inter-sample morphological variability, existing SSL methods face severe generalizability limitations under sparse supervision, leading to the Supervision Bias problem. To address this, we propose Structural Consensus-based KAN Prototype Learning (SCKAN), which constructs the first cross-sample structural consensus learning with Kolmogorov-Arnold Networks (KANs), to achieve more generalizable and accurate segmentation. Specifically, SCKAN contains two key designs: Structure-constrained Prototype Consistency Learning (SPCL), which prompts unbiased structural representation by enforcing cross-sample consistency via prototype-level contrastive optimization, and Consensus-based Kolmogorov-Arnold Fusion (CKaF), which reduces morphology-specific bias by aggregating stable consensus and filtering sample-wise noise via KAN’s adaptive B-spline nonlinearity. Extensive experiments on two public pancreas datasets demonstrate the effectiveness of SCKAN. Code is at https://github.com/rhodaliu17/SCKAN.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/0226_paper.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to the Code Repository

https://github.com/rhodaliu17/SCKAN

Link to the Dataset(s)

NIH-PAN: https://www.cancerimagingarchive.net/collection/pancreas-ct/ MSD-PAN: http://medicaldecathlon.com/

BibTex

@InProceedings{LiuYuq_SCKAN_MICCAI2026,
        author = { Liu, Yuqi AND Chen, Yufei AND Fu, Wei AND Yue, Xiaodong AND Li, Shuo},
        title = { { SCKAN: Structural Consensus-Based KAN Prototype Learning for Semi-supervised Pancreas Segmentation } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 16880},
        month = {September},
        page = {pending}
}


Reviews

Review #1

  • Please describe the contribution of the paper

    This paper proposes the first structural consensus-based Kolmogorov-Arnold Network (KAN) prototype learning framework (SCKAN). Through anatomy-guided feature consistency alignment (SPCL) and KAN-enhanced cross-sample denoising fusion (CKaF), it effectively overcomes the supervision bias problem caused by sample morphological diversity in semi-supervised pancreas segmentation.

  • 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.By introducing position-aware weighting and decorrelation regularization, it elegantly resolves the conflict between cross-sample structural alignment and intra-sample semantic collapse. However, an ablation study could be conducted on different weights. 2.Replacing standard MLPs with KANs for cross-sample fusion is a strong methodological choice. 3.The framework connects prototype extraction and pixel-level prediction via “Consensus-Guided Synergetic Learning,” using high-dimensional consensus to refine segmentation in a self-consistent loop.

  • 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 method applies K-means directly on unfiltered pseudo-labels from the teacher model. Lacking a confidence-weighting or rejection mechanism, erroneous masks (e. g. , fragmented boundaries or false positives) can easily propagate structural noise into the CKaF module and contaminate the global consensus. 2 While anatomically motivated (head, body, tail), applying spatial K-means with a fixed K=3 does not guarantee alignment with true biological boundaries, especially for deformed or pathological organs. The paper lacks a crucial ablation study on K (e. g. , 2, 4) to justify this strong assumption empirically. 3.The framework relies on numerous intertwined hyperparameters (e. g. , $\tau$, $\alpha$, $\lambda_{div}$, and Gaussian warm-up parameters). With only $\lambda_{div}$ ablated, the heavy tuning burden raises concerns about the method’s out-of-the-box generalization to new or cross-modality datasets.

  • 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.

  • 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.

    (4) Weak Accept — marginally above the acceptance threshold, but would not mind if rejected, dependent on rebuttal

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

    This work is distinguished by a methodologically sound and self-consistent design. In particular, the strategic use of KAN’s local filtering for cross-sample prototype aggregation and the novel position-aware contrastive learning scheme underscore the authors’ deep insight into the complexities of semi-supervised medical image segmentation and their strong command of algorithmic design.

  • 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.It leverages KANs to construct a cross-sample structural consensus learning framework, capturing anatomical knowledge shared across samples of varying morphologies, thereby enhancing the model’s generalization capability and segmentation accuracy.

    2.It proposes SPCL, which utilizes anatomical knowledge to partition the pancreas into three distinct regions—the head, body, and tail—thereby facilitating the model’s acquisition of unbiased structural feature representations.

    3.It designs CKaF, which exploits the powerful nonlinear fitting capabilities of the adaptive B-spline functions within the KAN architecture to effectively filter out morphological noise unique to each individual sample.

  • 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 novel deep learning modules and deeply integrates them with prior knowledge derived from clinical anatomy.

    1.While previous studies typically focused on learning the pancreas as a single entity, the proposed SPCL framework explicitly decomposes pancreatic features into three distinct sub-regions—the head, body, and tail—thereby mitigating issues related to supervisory bias. Furthermore, this study marks the first instance of incorporating KANs into semi-supervised pancreatic segmentation tasks; this approach effectively filters out morphological noise unique to individual samples, thereby enhancing the model’s generalization capability when encountering samples with previously unseen morphologies.

    2.The evaluation methodology is rigorous, featuring stress tests conducted under scenarios with extremely limited annotations; these tests validate the robustness of the “structural consensus” mechanism against interference, while the study also includes a comprehensive series of ablation experiments.

    3.Pancreatic segmentation constitutes a critical component in the clinical diagnosis of early-stage cancer; however, significant morphological variations often exist across different patients, frequently leading to issues of over-segmentation or under-segmentation. The framework proposed in this paper effectively captures the core structural features of the organ and demonstrates superior performance in terms of boundary adherence, thereby offering substantial clinical value.

  • 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.

    To extract the prototypes of the pancreatic head, body, and tail, a spatial-coordinate-based K-means clustering approach was employed; however, this method appears overly heuristic and hard-coded.As early as the 2015 paper on Spatial Transformer Networks, it was demonstrated that utilizing learnable spatial attention mechanisms or anatomical landmark detection to dynamically localize organ sub-regions offers greater robustness and theoretical novelty compared to static coordinate clustering.

  • 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.

    (4) Weak Accept — marginally above the acceptance threshold, but would not mind if rejected, dependent on rebuttal

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

    The primary reason for recommending “Accept” is that this paper accurately addresses a critical pain point: the poor generalization capability under low-labeling conditions resulting from the highly variable morphology of the pancreas. It innovatively leverages the anatomical characteristic of “structural consensus.” Furthermore, the introduction of KAN for cross-sample feature fusion—aimed at filtering out morphology-specific noise—is supported by a sound theoretical rationale and preliminary validation through ablation studies. The reason for recommending “Weak Accept” stems from the forced application of K-means clustering with k=3 to delineate the head, body, and tail of the pancreas; this constitutes a hard-coded approach that relies heavily on absolute spatial coordinates.

  • 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 tried to propose a semi-supervised pancreas segmentation method based on Kolmogorov-Arnold networks, and two modules, Structure-constrained Prototype Consistency Learning and Consensus-based Kolmogorov-Arnold Fusion, are used to strengthen the segmentation performance. Experiments show that it can realize the pancreas segmentation.

  • 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.
    • Good topic selection
    • The number and scale of the experiments are adequate.
  • 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.
    • Unclear motivation. The authors raise the issue of supervision bias, but fail to provide an in-depth analysis. Furthermore, they do not clearly justify the urgency of addressing this problem, nor do they explain why the proposed network is necessary to solve it.
    • Unclear figures. The figures are not clear enough to support the texts. For example, Figure 1 is hard to understand.
    • Insufficient method description. Many critical contents and components are missing. In addition, the rationale behind the design of each module is not adequately reflected.
    • Lack of rigor. For instance, references are missing for many of the comparison methods.
  • Please rate the clarity and organization of this paper

    Poor

  • 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 provide sufficient information for reproducibility.

  • 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.

    (3) Weak Reject — marginally below the acceptance threshold, but would not mind if 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 is with the problems stated in ‘the weaknesses of the paper’, but I desire to provide a rebuttal opportunity for the authors to explain them.

  • 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

N/A




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.

    The authors provide a highly innovative and methodologically sound approach by leveraging KANs and explicit anatomical priors to address supervision bias. Reviewers recognize the clear clinical value of using KAN’s local filtering to manage the severe morphological variability of the pancreas, acknowledging the framework’s self-consistent design despite minor concerns regarding spatial clustering.



back to top