List of Papers Browse by Subject Areas Author List
Abstract
Semi-supervised medical image segmentation (SSMIS) reduces reliance on costly manual annotations while providing pixel-level anatomical localization to support downstream clinical tasks. Consistency learning has emerged as a prominent paradigm in SSMIS, harnessing unlabeled data to improve segmentation fidelity. However, existing consistency-based methods predominantly enforce constraints at the pixel or feature level, overlooking the preservation of anatomical structural consistency essential for accurate delineation. This limitation is further compounded by prediction uncertainty, especially near lesion boundaries. Uncertain predictions pose a dilemma: naïvely enforcing consistency on them misleads training, while simply discarding them risks losing diagnostically valuable information embedded in these challenging regions. To address these issues, we propose an Uncertainty-Aware Hypergraph Consistency Learning (UAHC) framework. UAHC elevates consistency learning to a high-order hypergraph level, capturing complex anatomical dependencies to enforce structural coherence beyond pairwise modeling. Moreover, it introduces a dual-branch design that separately handles certain and uncertain regions via complementary consistency constraints. The Certain-Region Hypergraph Consistency (CHC) module constructs hypergraphs to align teacher-student representations on stable areas, while the Uncertain-Region Hypergraph Consistency (UHC) module enables effective structural learning on ambiguous regions without being overwhelmed by noise. A novel hyperedge-level uncertainty weighting mechanism further adaptively emphasizes diagnostically significant yet uncertain anatomical patterns. Extensive experiments on four public medical segmentation benchmarks show that UAHC consistently outperforms state-of-the-art methods.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/4517_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/BaoyaoGroup/UAHC
Link to the Dataset(s)
N/A
BibTex
@InProceedings{XieKex_UncertaintyAware_MICCAI2026,
author = { Xie, Kexin AND Yang, Baoyao AND Yin, Kejing AND Cheung, William K. AND Luo, Guibo},
title = { { Uncertainty-Aware Hypergraph Consistency Learning for Semi-supervised Medical Image 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
Main Contributions (Reviewer Perspective – Critical Tone)
This paper presents an Uncertainty-Aware Hypergraph Consistency Learning (UAHC) framework for semi-supervised medical image segmentation. The main contributions can be summarized as follows:
Extension of consistency learning to a hypergraph formulation. The method reformulates conventional pixel-/feature-level consistency into a hypergraph-based framework, aiming to capture higher-order relationships among groups of pixels. While this provides a more expressive structural modeling compared to pairwise approaches, the overall formulation largely builds upon existing consistency learning paradigms. Region-wise dual-branch design guided by uncertainty. The proposed CHC and UHC modules separately handle high-confidence and uncertain regions, respectively. This design introduces a structured way to incorporate uncertainty into consistency learning, although it remains conceptually related to prior uncertainty-aware semi-supervised frameworks. Hyperedge-level uncertainty weighting mechanism. The approach propagates pixel-level uncertainty to the hypergraph level and uses it to reweight consistency objectives. This can be seen as a reasonable extension of uncertainty modeling, though its advantage over alternative uncertainty estimation or weighting strategies is not fully explored. Empirical improvements on benchmark datasets. The method demonstrates consistent performance gains across several public datasets under semi-supervised settings. However, the evaluation is limited to 2D benchmarks, and additional validation (e. g. , on 3D data or with statistical significance analysis) would further strengthen the claims.
- 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.Structured extension of consistency learning to hypergraph modeling. The paper extends conventional pixel-/feature-level consistency learning to a hypergraph formulation, enabling modeling of high-order relationships among groups of pixels. Compared to pairwise graph-based methods, this design is conceptually more expressive and better aligned with the structural nature of anatomical regions, where dependencies are often non-local and multi-element. 2.Well-motivated integration of uncertainty into structural learning. The work explicitly addresses a key limitation of existing semi-supervised segmentation methods—namely, the negative impact of uncertain predictions on consistency learning. The proposed region-wise partition (certain vs. uncertain) combined with tailored consistency strategies provides a reasonable and intuitive mechanism to mitigate noise while still leveraging informative ambiguous regions. 3.Hyperedge-level uncertainty weighting. The propagation of pixel-wise uncertainty to the hypergraph level and its use in reweighting consistency objectives is a technically interesting design. It provides a more structured alternative to standard pixel-wise masking or thresholding strategies commonly used in prior work. 4.Comprehensive experimental validation across datasets. The method is evaluated on multiple public datasets spanning different modalities and tasks (e. g. , polyp, skin lesion, ultrasound, cardiac MRI), showing consistent improvements over prior methods under low-label regimes. The inclusion of ablation studies (CHC/UHC/NS) and hyperparameter analysis further strengthens the empirical evaluation.
- 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.Limited methodological novelty relative to existing literature. While the hypergraph formulation is a meaningful extension, the overall framework builds heavily on established components in semi-supervised segmentation, including: Mean Teacher paradigm (Tarvainen & Valpola, NeurIPS 2017), consistency learning frameworks (e. g. , CPS, CVPR 2021), uncertainty-aware methods (e. g. , UA-MT, MICCAI 2019). Additionally, hypergraph-based modeling has been explored in medical segmentation (e. g. , Dong et al. , TNNLS 2020; Liu et al. , PR 2025). As such, the contribution is more of a careful integration and extension rather than a fundamentally new paradigm. 2.Insufficient justification of hypergraph design choices. The hypergraph construction relies on KNN-based hyperedges and feature similarity, but the paper does not provide: theoretical justification for this choice, comparison with alternative structural modeling approaches (e. g. , graph-based methods, attention mechanisms, or prototype-based representations). It remains unclear whether hypergraphs are essential, or simply one possible design choice. 3.Lack of 3D validation despite domain relevance. Although the method is motivated by medical imaging, all experiments are conducted on 2D slice-based datasets. Given that many clinically relevant segmentation tasks are inherently 3D, the absence of volumetric evaluation limits the practical impact and generalizability of the approach. 4.Training protocol deviates from common semi-supervised practice. The method applies supervised training for a large portion of epochs before introducing semi-supervised learning. This staged strategy differs from standard joint training pipelines and may influence performance. The paper does not provide sufficient justification or comparison to alternative training schedules. 5.Minor issues in clarity and presentation. The manuscript contains several minor issues, including typographical errors, duplicated figures, and occasionally unclear notation. While not critical, these issues affect readability and should be addressed. 6.Limited analysis of uncertainty modeling strategy. The uncertainty estimation is based on entropy and cross-model disagreement, which are standard techniques. The paper does not compare against more advanced uncertainty estimation approaches such as: Monte Carlo Dropout, deep ensembles, Bayesian methods.
- 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.
- 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 paper presents a structured extension of consistency learning for semi-supervised medical image segmentation by introducing a hypergraph-based formulation with uncertainty-aware modeling. The work is well-motivated, addressing two relevant challenges in the field: the lack of structural consistency modeling and the difficulty of leveraging uncertain regions effectively.
On the positive side, the proposed method introduces a hypergraph-based consistency framework that aims to capture higher-order anatomical relationships beyond pairwise interactions. The dual-branch design (CHC and UHC) is conceptually sound and provides a reasonable mechanism to balance learning from reliable and uncertain regions. In addition, the hyperedge-level uncertainty weighting is a technically interesting component that extends standard pixel-wise uncertainty handling to a more structured representation. The experimental evaluation is relatively comprehensive, covering multiple datasets and including ablation studies, with consistent performance improvements over prior methods.
However, there are several limitations that prevent a stronger recommendation. First, the overall methodological novelty is somewhat incremental, as the framework builds upon well-established paradigms such as Mean Teacher, consistency learning, and uncertainty-aware training. While the hypergraph formulation is a meaningful extension, it is not fully demonstrated that this design is essential compared to alternative structural modeling approaches. Second, the uncertainty modeling relies on relatively standard techniques (entropy and model disagreement), without comparison to more advanced methods. Third, all experiments are conducted on 2D datasets, which limits the demonstrated impact in the context of 3D medical imaging, a core focus of MICCAI. Additional validation on volumetric data would strengthen the contribution. Finally, the lack of statistical significance analysis and limited discussion of training protocol choices reduce the rigor of the empirical evaluation.
Overall, the paper represents a solid but incremental contribution with consistent empirical gains and a reasonably well-designed framework. I believe it is slightly above the acceptance threshold, although the final decision may depend on the strength of the rebuttal and comparison with competing submissions.
- 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 proposes an Uncertainty-Aware Hypergraph Consistency Learning (UAHC) framework for semi-supervised medical image segmentation. It introduces a dual-branch hypergraph structure to model high-order anatomical dependencies, separately enforcing consistency on certain and uncertain regions, and uses hyperedge-level uncertainty weighting to exploit valuable information from ambiguous areas while suppressing noise.
- 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 paper addresses the important problem of structural consistency and uncertainty modeling in semi-supervised medical image segmentation.
2.The writing is clear and the technical pipeline is complete with reasonable motivation.
3.The proposed method is validated on four public datasets with sufficient quantitative and qualitative 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.The authors should include more recent state-of-the-art semi-supervised segmentation methods in the comparison, especially those designed for medical images with uncertainty or hypergraph-based learning, to fully demonstrate the superiority of the proposed UAHC framework.
2.A discussion on failure cases is strongly recommended. The authors should analyze typical failure examples, explain the main limitations of the proposed method, and clarify under what conditions the model may perform unsatisfactorily.
3.For the hyperparameter analysis in Figure 4, more evaluation metric curves should be provided. All curves can be plotted in the same figure for better visualization and more comprehensive information presentation.
4.To better validate the robustness and generalizability of the proposed components, the ablation study in Table 2 should be extended to at least one more dataset instead of only being reported on Kvasir-SEG.
- 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.
- 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 score is mainly based on the novel hypergraph-based consistency design, competitive performance on multiple medical datasets, and solid experimental validation. However, the work still needs improvements such as enriching comparisons with recent SOTAs, analyzing failure cases, and extending ablation studies to further enhance its quality.
- Reviewer confidence
Confident but not absolutely certain (3)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
Accept
- [Post rebuttal] Please justify your final decision from above.
The authors have adequately addressed my queries in the rebuttal. I maintain my original recommendation for acceptance.
Review #3
- Please describe the contribution of the paper
This paper proposes an Uncertainty-Aware Hypergraph Consistency Learning (UAHC) framework for semi-supervised medical image segmentation. The method extends conventional consistency learning from pixel/feature-level constraints to hypergraph-level structural alignment, with a dual-branch design that separately handles certain and uncertain regions. Specifically, the CHC module aligns stable hypergraph structures in certain regions, while the UHC module performs uncertainty-aware hypergraph consistency learning in ambiguous regions with density-guided noise suppression and hyperedge-level uncertainty weighting. The method is evaluated on four public datasets under 5% and 10% labeled settings, with additional ablation, efficiency, and hyperparameter analysis.
- 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 studies a relevant problem in semi-supervised medical image segmentation. The motivation of improving consistency learning by incorporating anatomical structure and uncertainty is well aligned with the practical challenges of medical segmentation, especially in low-contrast and boundary-ambiguous regions. The overall framework is reasonably coherent. The proposed UAHC design combines three intuitive ideas in a structured manner: hypergraph-based high-order modeling, region-wise decomposition into certain and uncertain areas, and uncertainty-aware weighting. The CHC/UHC split is easy to understand and gives the method a clear conceptual structure. The experimental evaluation is relatively broad. The method is tested on four public datasets from different imaging scenarios, and results are reported under both 5% and 10% labeled data settings. The paper also includes qualitative results, ablation studies, efficiency analysis, and hyperparameter analysis. Quantitatively, the method is competitive on multiple datasets and particularly strong on ACDC, where it reports the best Dice/Jaccard/ASD among the listed baselines. The computational overhead reported in the ablation table is also moderate relative to the baseline. The manuscript is generally readable at a high level, and the overview figure helps illustrate the intended role of the CHC and UHC modules.
- 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 novelty is somewhat limited for a MICCAI paper. While the combination of hypergraph consistency learning and uncertainty-aware region handling is reasonable, the contribution appears more incremental than fundamentally new. The manuscript positions hypergraph modeling as a key novelty, but the cited literature already includes structurally related graph/hypergraph work, including GraphCL and a 2025 hypergraph-based SSMIS paper (MFHS). However, these most relevant methods are not directly compared in the experiments. The paper overstates its empirical conclusions. In the results section, the manuscript claims that UAHC “consistently outperforms” prior methods across all datasets and metrics. However, Table 1 does not fully support this statement. For example, on BUSI with 5% labeled data, ABD reports higher Dice/Jaccard than UAHC. This inconsistency between the narrative and the table weakens confidence in the paper’s claims. Several methodological details are unclear. In Eq. (2), the definitions of certain and uncertain regions do not appear to cover all pixels; some pixels may satisfy neither condition, but the paper does not explain how such pixels are handled. In addition, the dynamic threshold schedule for τ is not specified, and the construction of the “shared adjacency structure” between teacher and student is not sufficiently described. There are potential formula inconsistencies. In the certain-region hypergraph construction, the paper states that KNN is built based on feature similarity, but the edge weight is written as a plain Euclidean distance. In Eq. (6), the affinity term uses exp(||zi-zj||^2 / τ) without a negative sign, which would imply larger weights for more distant nodes. This may be a typo, but as written it is inconsistent with the stated intuition. The experimental protocol raises fairness questions. The paper states that training runs for 100 epochs, with only labeled data used for the first 75 epochs and the UAHC framework introduced only in the last 25 epochs. This is unusual for semi-supervised segmentation and requires stronger justification. It is also not fully clear whether all baselines were carefully re-implemented under the same protocol or whether some results were taken from prior work. Reproducibility is not yet strong enough. Although many training details are given, several core implementation choices remain ambiguous, and the paper does not provide enough clarity to fully reproduce the proposed method. The manuscript still has some signs of insufficient polishing. For example, the references contain duplicated entries for Mutual Consistency Learning, and the text includes several typos and phrasing issues. While these are not fatal on their own, they contribute to the impression that the paper is not yet fully mature.
- 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 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
The paper addresses an interesting and relevant problem, and the overall framework is reasonably well motivated. To strengthen the work, I encourage the authors to do three things. First, please improve the technical clarity of the method, especially the exact handling of pixels not covered by Eq. (2), the schedule for the dynamic threshold, and the construction of the shared hypergraph adjacency. Second, please better position the work against the most closely related graph/hypergraph semi-supervised segmentation methods and include direct comparisons where possible. Third, please revise the empirical claims to match the actual numbers in Table 1 and avoid overstatements such as “consistently outperforms” when the table does not fully support that conclusion.
- 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?
I recommend weak rejection. The paper addresses a relevant problem in semi-supervised medical image segmentation and proposes a reasonably structured uncertainty-aware hypergraph consistency framework. The experimental section is also broader than average, covering four datasets, two label ratios, ablation, efficiency, and hyperparameter analysis. However, I am not yet convinced that the contribution is strong enough for MICCAI in its current form. The novelty appears somewhat incremental, especially given the existence of related graph/hypergraph semi-supervised segmentation literature that is cited but not directly compared. In addition, several methodological details remain unclear, including the exact region partitioning behavior, the dynamic thresholding scheme, and the hypergraph construction details. I am also concerned by inconsistencies between the paper’s claims and Table 1, where the manuscript states that the method consistently outperforms prior work across all datasets and metrics, but the reported numbers do not fully support that statement. Overall, the paper has potential, but the current version is not yet sufficiently convincing in terms of novelty, rigor, and empirical positioning.
- 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
Novelty (R1, R3): Our work addresses an underexplored question in SSMIS: how to jointly model higher-order anatomical structures while effectively leveraging informative cues from uncertain regions.
- Why existing methods fall short: (1) Pixel-/feature-level consistency methods cannot capture group-wise anatomical dependencies and are vulnerable to noisy pseudo-labels. (2) Uncertainty-aware methods suppress or discard uncertain regions, ignoring diagnostically critical boundaries.
- Our approach tightly couples the two: (1) We extend consistency learning from pairwise to hypergraph-based higher-order structural modeling. (2) We route the uncertain areas through a dedicated UHC module and reweight them to preserve informative boundaries while suppressing noise.
- Distinction from prior hypergraph works: Dong et al. (TNNLS’20) applies hypergraphs to fully supervised brain segmentation (no SSL, no uncertainty). Liu et al. (PR’25, MFHS) uses hypergraphs on a SAM foundation backbone, a heavyweight setting different from our lightweight framework. Neither addresses uncertainty-aware hypergraph consistency in SSMIS.
Hypergraph Design Justifications (R1, R3): Hypergraphs are necessary. Unlike graphs that model only pairwise relations, hyperedges explicitly represent group-wise anatomical dependencies among multiple co-varying pixels (e.g., along a lesion boundary). Compared to attention, hyperedges provide an explicit group structure, allowing our hyperedge-level uncertainty aggregation in Eq. 8.Compared to prototype-based representations, hyperedges preserve local feature variation within a group rather than collapsing it to a single mean, critical for boundary structures. For constructing hypergraph, we follow HGNN (AAAI19) to use KNN with feature similarity.
3D validation (R1, R3): UAHC can be extended to 3D, requiring only computing KNN in volumetric feature space, with no architectural change. The ACDC dataset itself is a clinically 3D modality, evaluated slice-wise following standard SSMIS protocols. We will include volumetric evaluations in our future work.
Training Protocol and Fairness (R1, R3): We respectfully clarify that our training protocol (75 epochs labeled-only warm-up, 25 epochs SSL) follows prior SSMIS work, UARC [25]. The rationale is that: UAHC’s hyperedge construction and uncertainty estimation depend on teacher features being stable and semantically meaningful; applying hypergraph consistency on noisy early features yields unreliable hyperedges. Thus, we adopt this warm-up training paradigm from prior work. All baselines use the same protocol to ensure fairness.
Typographical Errors (R3, R3): We thank the reviewer for pointing out the issues: 1) Formula typos: Eq. (2) should use max(Us(i), Ut(i)) (instead of min) for uncertain-region selection. Eq. 6 missed a negative sign and should be exp(-||zi-zj||^2/tau), where tau follows UARC’s dynamic strategy (Sec 2.1). We confirm that implementations used the correct ones, and all pixels are covered; the results are unaffected. 2) Table error (BUSI-5% in Table 1): The reported performance for ABD model under 5% labeled data unexpectedly (and mistakenly) exceeded that under 10% labeled data. We identify a transcription error after a careful inspection: The corrected results for ABD BUSI-5% should be Dice/Jaccard=71.34/57.22.Our conclusion that the proposed method consistently outperforms existing methods holds. We will open-source codes and training/testing logs upon acceptance for reproducibility.
Comparison to Uncertainty Estimation Methods (R1): MC Dropout requires 5-50x forward passes at inference and Deep Ensembles require training multiple full models, limiting practicality for SSMIS.
Comparison to More Recent SOTAs (R2): MFHS is a more recent method built on SAM foundation model. For reference under similar protocols, MFHS reports 85.67 Dice on ACDC 5% in its paper, versus our 89.39.We will include relevant discussions in the camera-ready.
Meta-Review
Meta-review #1
- Your recommendation
Invite for Rebuttal
- 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 reviewers raised several concerns about this paper. Specifically, they highlighted limited methodological novelty, insufficient justification for hypergraph design choices, and a lack of 3D validation. These constrain the contribution’s impact and generalizability. Concerns were also raised regarding the atypical training protocol, incomplete comparisons with relevant baselines, unresolved clarity issues in both methodology and presentation, and limited analysis of uncertainty modeling strategies.
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Accept
- Please justify your recommendation.
The reviewers acknowledge the contributions of the paper, particularly its hypergraph-based consistency framework and competitive empirical performance. Key concerns include limited methodological novelty, lack of direct comparison with closely related structural methods, absence of validation on 3D datasets, and insufficient statistical rigor. Overall, while the framework is interesting and the results promising, the incremental nature of the work and incomplete evaluation methods suggest that the paper is slightly above the acceptance threshold, contingent on the rebuttal’s strength.
Meta-review #2
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Accept
- Please justify your recommendation.
The rebuttal adequately addresses the reviewers’ concerns. There are still some concerns including writing issues. But the strengths overall outweigh the weaknesses of the paper. I consider this paper a borderline accept. I strongly suggest that the authors make the necessary corrections including eqs (2) and (6), and Table 1 results.
Meta-review #3
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Accept
- Please justify your recommendation.
This paper was initially invited for rebuttal following the primary review period, where reviewers raised several concerns regarding limited novelty, insufficient justification, and a lack of 3D validation. After carefully evaluating the authors’ rebuttal, the reviewers’ revised feedback, and re-examining the paper, I conclude that the authors have effectively addressed the key concerns raised, and the work now meets MICCAI’s publication standards.