Abstract

Accurate segmentation of the carotid artery from computed tomography angiography (CTA) images is essential to manage atherosclerotic vascular disease. However, studies specifically targeting carotid artery segmentation in CTA imaging remain relatively limited, and existing 3D segmentation methods face significant challenges when applied to this task, including discontinuous lesion distribution, extreme foreground scarcity, inadequate capture of the vessel wall’s low-contrast subtle double-edge boundaries and irregular calcification contours. Moreover, these methods suffer from insufficient spatial and semantic complementarity modeling in encoder-decoder feature alignment. To address these limitations, we propose a novel Multi-View Dual-Alignment Network (MVDA-Net) that integrates three key innovations: (1) a Spatially-Aware Orthogonal Projection (SAOP) module that enhances global context modeling and representation of small vascular structures through multi-view projection across orthogonal anatomical planes with spatial attention; (2) a Hierarchical Inverted Attention Refinement (HIAR) module that propagates boundary-enhancing inverted attention from deep features to shallow layers for progressive boundary refinement; and (3) a Synergistic Feature Alignment (SFA) module that suppresses redundancy and enhances complementarity through feature differences while modeling semantic relationships via summation for effective spatial-semantic alignment. Comprehensive experiments on one in-house carotid artery CTA dataset (named CA) and three public benchmarks demonstrate that our proposed MVDA-Net achieves state-of-the-art performance. The code for MVDA-Net is publicly available at: https://github.com/Vi-cai/MVDA-Net.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/1321_paper.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to the Code Repository

https://github.com/Vi-cai/MVDA-Net

Link to the Dataset(s)

N/A

BibTex

@InProceedings{WenYan_MVDANet_MICCAI2026,
        author = { Wen, Yang AND Wang, Xiaoyao AND Song, Wenfeng AND Shi, Wuzhen AND Fu, Huazhu AND Zhu, Lei AND Sheng, Bin},
        title = { { MVDA-Net: Multi-View Dual-Alignment Network for 3D Carotid Artery Segmentation } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 16884},
        month = {September},
        page = {pending}
}


Reviews

Review #1

  • Please describe the contribution of the paper

    The paper proposes the Multi-View Dual-Alignment Network (MVDA-Net) for 3D carotid artery segmentation in CTA images, achieving state-of-the-art performance via three modules: SAOP for multi-view projection, HIAR for inverted attention-based boundary refinement, and SFA for spatial–semantic feature alignment.

  • 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 proposes MVDA-Net for 3D carotid artery segmentation in CTA, addressing challenges such as discontinuous lesions and severe foreground scarcity.
    2. It introduces three complementary modules: SAOP for multi-view contextual modeling, HIAR for boundary refinement via inverted attention, and SFA for spatial–semantic feature alignment.
    3. The method is evaluated on a newly curated clinical dataset and multiple public benchmarks, achieving state-of-the-art performance.
  • 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 use of multi-view pooling and axis projection for handling foreground–background imbalance and capturing global context is not new, as similar ideas appear in prior works (e. g. , APAUNet, LW-CTrans). Likewise, attention mechanisms for boundary refinement have already been explored (e. g. , BMANet), reducing the originality of the proposed design.

    2.It is not clearly specified how feature maps (e. g. , F_i and F_(i+1)) are defined and selected. In particular, the indexing iis not justified, and the distinction between low-level and high-level features remains ambiguous.

    3.There is a task–evaluation mismatch. Although the modules (SAOP, HIAR, SFA) are claimed to be designed for carotid artery segmentation, it is unclear how well they generalize to other organs or lesions. Moreover, the evaluation on datasets such as MSD Lung, MSD Spleen, and BraTS is not fully aligned with the primary task of 3D carotid artery segmentation stated in the paper.

  • 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.

    (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?

    Several core components extend existing techniques, such as multi-view pooling and attention-based boundary refinement. Although the method is well-motivated and achieves good empirical results, concerns remain regarding the clarity of the architecture and the consistency between the targeted task and the chosen evaluation datasets.

  • Reviewer confidence

    Very confident (4)

  • [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.

    Reject

  • [Post rebuttal] Please justify your final decision from above.

    While the authors provide some clarifications, the rebuttal does not sufficiently address the main concerns. The claimed novelty remains limited, as the core ideas remain close to well-established strategies, and the rebuttal does not sufficiently demonstrate a new mechanism or substantial methodological advance over existing approaches. Moreover, the clarification that F_i and F_i+1 are adjacent encoder features does not fully resolve the methodological ambiguity, as the choice of feature levels and the low-/high-level distinction remain insufficiently justified. The paper should clearly justify why these specific adjacent feature levels are chosen, how the low-level/high-level distinction is operationally defined, and whether the indexing strategy is essential to the performance. The rebuttal indicates that this will be clarified in revision, but it does not provide enough evidence that the current formulation is technically well justified.



Review #2

  • Please describe the contribution of the paper

    The authors propose MVDA-Net, a novel U-shaped architecture designed to address the inherent challenges of 3D carotid artery segmentation in CTA imaging. The primary contribution of this work lies in the integration of three specialized modules: the Spatially-Aware Orthogonal Projection (SAOP) module, which enhances global context modeling and small vascular structure representation through multi-view anatomical plane projections ; the Hierarchical Inverted Attention Refinement (HIAR) module, which propagates boundary-enhancing inverted attention from deep to shallow layers to achieve progressive boundary refinement ; and the Synergistic Feature Alignment (SFA) module, designed to optimize encoder-decoder integration through difference-based spatial alignment and summation-based semantic balancing.

  • 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-organized and follows a logical structure that enhances technical readability. The authors provide a comprehensive overview of the MVDA-Net architecture, effectively utilizing clear diagrams to illustrate the system workflow and individual module configurations. Each primary component, including the SAOPFormer, HIAR, and SFA, is clearly defined with its own dedicated subsection and mathematical formulation, ensuring the proposed framework is easy to follow.
    • A notable strength of this work is the explicit alignment between identified challenges in 3D carotid arterial segmentation and the proposed solutions. The authors specifically design the Spatially-Aware Orthogonal Projection (SAOP) module to address discontinuous lesion distribution and extreme foreground scarcity. Simultaneously, the Hierarchical Inverted Attention Refinement (HIAR) and Synergistic Feature Alignment (SFA) modules are purposefully engineered to overcome low-contrast boundary delineation and insufficient spatial-semantic complementarity, respectively. -The model is evaluated by validating the MVDA-Net across a diverse range of datasets. Beyond testing on a specialized in-house carotid artery dataset, the authors perform extensive experiments on three distinct public benchmarks: MSD Lung, MSD Spleen, and BraTS. Achieving the highest performance across these various anatomical structures and imaging modalities (CT and MRI) provides evidence of the model’s robustness and generalization capabilities.
  • 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.
    • Although the paper emphasizes boundary refinement via the HIAR module, the evaluation relies solely on the Dice score. To adequately validate the claimed improvements in boundary precision, surface-based metrics such as Hausdorff Distance (HD95) should be included.
    • While 5-fold cross-validation is mentioned for the in-house dataset, the experimental setup for the three public benchmarks is unclear. The absence of standard deviations (std), confidence intervals, and p-values makes it difficult to assess whether the performance gains over baselines are statistically significant.
    • There is a lack of clarity in Equations (4) and (5) relative to the architecture shown in Figure 3b. The mathematical formulation of the SFA module is not sufficiently detailed to ensure reproducibility, which is particularly concerning given that the source code is not provided.
    • The reported input size of 192×192×32 for the MSD Lung dataset appears unusually shallow for the Z-axis. The authors should verify this configuration, as such a significant reduction in spatial resolution could lead to substantial information loss.
    • The assertion that this is the “first model specifically designed for 3D carotid arterial segmentation in CTA” is an overclaim. Previous studies have already addressed this specific clinical task.
    • Huang, X., Wang, J., & Li, Z. (2023). 3D carotid artery segmentation using shape-constrained active contours. Computers in Biology and Medicine, 153, 106530.
    • Xie, H., Gu, H., Li, M., Zhu, L., Wang, T., Li, Z., & Wu, H. (2025). Carotid artery segmentation in computed tomography angiography (CTA) using multi-scale deep supervision with Swin-UNet and advanced data augmentation. Quantitative imaging in medicine and surgery, 15(4), 3161–3175. Therefore, the authors should revise their positioning relative to these existing works.
  • 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.

    (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?

    The paper is well-structured and tackles a clinically relevant problem, with a clear link between challenges and the proposed modules, and consistent improvements across datasets. However, the novelty is limited and mostly incremental. The evaluation is not fully convincing due to missing boundary metrics and statistical analysis. There are also reproducibility concerns, such as unclear formulations and no code, along with some overstated claims about being the first work in this area.

  • 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 cleared up their design choices, fixed the overclaim about being the first, and added all the missing details on evaluation and reproducibility. More importantly, the method delivers solid gains across their private dataset and three public benchmarks, and the ablation study proves all three modules actually work.



Review #3

  • Please describe the contribution of the paper

    The paper introduces the Multi-View Dual-Alignment Network (MVDA-Net), a novel architecture designed specifically for 3D carotid artery segmentation in CTA imaging. MVDA-Net addresses critical challenges in vascular segmentation, including the discontinuous distribution of lesions, foreground scarcity, and boundary refinement. The model incorporates three key modules: the Spatially-Aware Orthogonal Projection (SAOP) module, the Hierarchical Inverted Attention Refinement (HIAR) module, and the Synergistic Feature Alignment (SFA) module. The SAOP module improves small target representation and spatial context modeling by leveraging multi-view projections across anatomical planes, enhancing feature propagation across discontinuous regions. The HIAR module focuses on boundary delineation by refining low-contrast vessel walls and calcification contours through inverted attention mechanisms. The SFA module enhances encoder-decoder feature alignment by capturing both spatial and semantic complementarity. The paper demonstrates the superiority of MVDA-Net through extensive experimental validation on an in-house dataset and public benchmarks, showing its superior performance in segmentation accuracy and boundary delineation.

  • 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.

    A key strength of this paper is the introduction of the MVDA-Net architecture, which addresses specific challenges in carotid artery segmentation using three innovative modules: SAOP, HIAR, and SFA. The SAOP module enhances small target representation and addresses foreground scarcity by incorporating multi-view projections across anatomical planes. The HIAR module refines boundary delineation by propagating attention from deep to shallow features, improving the delineation of vessel walls and calcifications. The SFA module efficiently aligns encoder-decoder features, improving spatial-semantic complementarity, which is crucial for accurate segmentation of complex vascular structures.

    The method is thoroughly validated using both an in-house carotid artery dataset and public benchmark datasets, demonstrating its state-of-the-art performance. The experiments highlight MVDA-Net’s superiority in segmentation accuracy, particularly for challenging structures like vessel walls and calcifications. Carotid artery segmentation is critical for the diagnosis and treatment of atherosclerotic vascular diseases, and the model’s potential to improve segmentation accuracy could have significant implications in clinical decision-making.

  • 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 presents a strong methodology for carotid artery segmentation, but there are several areas that could be further strengthened or clarified.

    1.While the method is validated on three public datasets (Lung, Spleen, and BraTS), the segmentation tasks on these datasets differ significantly from the task of carotid artery segmentation in terms of morphology, texture, and pathological features. The excellent performance on non-vascular organs and tumors demonstrates the model’s strong feature learning ability, but it does not directly prove the model’s effectiveness in addressing carotid artery-specific challenges, such as the subtle double-edge boundaries of the vessel wall or irregular calcification contours. The results on these non-vascular datasets do not fully validate the model’s ability to handle the unique complexities associated with carotid artery segmentation.

    2.The paper primarily uses Dice as the performance evaluation metric, which, while commonly used in segmentation tasks, is not comprehensive enough to assess boundary precision. For a task like carotid artery segmentation, where boundary accuracy is crucial, additional metrics such as Hausdorff distance, average surface distance, or other boundary-specific metrics would provide a more thorough evaluation of the model’s performance. These metrics would particularly help validate the effectiveness of the HIAR module in boundary refinement.

    3.The ablation study in Table 2 shows the performance improvements from combining different modules. However, the paper does not delve deeply into the impact of specific design choices within each module.

    4.The paper does not mention important efficiency metrics such as model parameter count, FLOPs, training/inference time, or memory usage. These metrics are critical for assessing the feasibility of deploying MVDA-Net in clinical settings, where resource constraints may play a significant role. A discussion on these aspects would help determine whether the model is suitable for real-time clinical applications or large-scale clinical deployment.

  • 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.

    (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?

    Based on the paper’s contributions and thorough evaluation, I would rate it a 5 (Accept), as the proposed model presents solid innovation and strong validation results.

  • 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.

    After reading the authors’ rebuttal, I lean toward acceptance.

    The rebuttal addressed most of the concerns raised in my original review, including clarification of the module motivations, additional discussion on the evaluation strategy, and supplementary efficiency-related information. The authors also acknowledged several overstated claims and agreed to revise them in the final version.

    Although some concerns remain, particularly regarding the lack of boundary-specific metrics and more comprehensive statistical analysis in the main manuscript, I believe these issues are relatively minor and do not fundamentally affect the overall technical soundness of the work.

    Overall, the paper is borderline but slightly above the acceptance threshold due to its solid engineering design, clinically relevant application, and consistent empirical improvements across datasets.



Author Feedback

We appreciate the comments and respond below. Reviewer#1 Q1: Originality of the proposed design. Our method specifically targets key challenges in carotid artery CTA segmentation: discontinuous lesion distribution, extreme foreground scarcity, inadequate capture of low-contrast subtle double-edge boundaries of the vessel wall, and irregular calcification contours. Different from APAUNet and LW-CTrans, our SAOP addresses the challenge of discontinuous lesions through spatial-aware feature recalibration, improving the spatial distinguishability of multiple lesion regions. Different from BMANet, our HIAR applies an inverted attention mechanism to deep encoder features to suppress high-confidence region responses, guiding shallow features to focus on uncertain regions such as boundaries. To further enhance shallow feature representation, we design 3D SimAM to selectively enhance fine-grained details while suppressing redundancy. Q2: Unclear definition of feature maps. In HIAR, F_i and F_(i+1) (1≤i≤4) denote feature maps from adjacent encoder stages. F_i from a shallower stage provides richer spatial details (low-level), while F_(i+1) from a deeper stage provides stronger semantic information (high-level). These definitions will be clarified in the revised manuscript. Q3: Task-evaluation mismatch and cross-dataset validation. Except for our self-constructed dataset, no publicly available carotid artery test set provides dense voxel-wise annotations. We further tested on the MSD Lung, MSD Spleen, and BraTS datasets due to their shared challenges with carotid artery segmentation, including extremely low foreground proportions (0.04%, 0.47%, and 1.16%, respectively) and irregular boundaries. Evaluation on these datasets validates the generalization of our method under similarly challenging segmentation scenarios. Reviewer#2 Q1: Lack of boundary-specific evaluation metrics. In the HIAR ablation study on CA, the model with HIAR yielded a better HD95 compared to the baseline. This confirms the effectiveness of HIAR in boundary refinement. Due to page limits, these results will be released with the code upon acceptance. Q2: Experimental setup and statistical significance analysis. For MSD Lung and BraTS, we followed the official splits of UNETR++ and nnFormer, respectively. These details will be added in the revised manuscript. Statistical significance analysis will be released with the code upon acceptance. Q3: Unclear descriptions in Eq.(4)-(5) and SFA. We will clarify Eq.(4)-(5) and SFA with detailed symbols and interactions in the revised manuscript and release the code upon acceptance. Q4: Input size for MSD Lung dataset. For MSD Lung, we adopt the same nnUNet-based preprocessing and patch planning strategy as UNETR++. The 192×192×32 is the training patch size; nnUNet’s sampling strategy (33% foreground-centered, 67% random) is applied, ensuring no information loss. Q5: Overclaim regarding first 3D carotid artery model. We acknowledge the overclaim and will remove the “first” claim and clarify our positioning relative to prior works. Existing studies only focus on anatomical carotid segmentation, whereas our work targets multi-structure segmentation of vessel wall, lumen, and calcification in carotid CTA. Reviewer#3 Q1: Please refer to our response to R1Q3.Q2: Please refer to our response to R2Q1.Q3: Ablation analysis. As shown in Tab.3, ablation on SAOPBlock verifies the effectiveness of adaptive fusion weights. For SFA, the difference-based branch models spatial complementarity, while the summation-based branch balances semantic information. Ablating them individually and jointly confirms that their combination is superior, validating the necessity of both. Due to page limits, detailed results will be released with the code upon acceptance. Q4: Computational efficiency. On CA, our method achieves 50.02M parameters, 357.69 GFLOPs, 16.82 GB memory, and 0.068 s inference time, with comparable computational efficiency to baselines.




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.

    This paper proposes MVDA-Net for 3D carotid artery segmentation in CTA and introduces three dedicated components, SAOP, HIAR, and SFA, to address foreground scarcity, discontinuous lesions, and boundary refinement. Reviewers appreciated the clear organization of the paper, the generally well-structured method presentation, and the extensive empirical evaluation on both an in-house carotid dataset and multiple public benchmarks. The clinical relevance of carotid artery segmentation was also recognized, and the reported performance is promising.

    At the same time, the reviewers raised several important concerns that require clarification. A central issue is the degree of methodological novelty, as some of the core design elements appear related to existing multi-view context modeling, attention-based refinement, and feature alignment approaches. The authors should better position the contribution relative to prior work and moderate claims such as being the first model specifically designed for 3D carotid artery segmentation. Reviewers also noted that some technical details remain unclear, particularly the definition of intermediate feature maps and the mathematical formulation of the SFA module, which affects reproducibility.

    On the experimental side, the current evaluation does not fully support all claims. In particular, boundary refinement is emphasized, but only Dice is reported; additional boundary-sensitive metrics such as HD95 or ASD would be important. Reviewers also asked for clearer reporting of the experimental setup on the public datasets, inclusion of uncertainty or significance analysis, and discussion of efficiency metrics. Finally, while the public benchmark results are useful, their connection to the carotid-specific challenges of the target task should be more carefully justified.

    Overall, the paper addresses an important application and shows promising results, but the current reviews indicate that several issues regarding novelty, reproducibility, and evaluation remain unresolved. I encourage the authors to use the rebuttal to clarify these points.

  • 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 presents MVDA-Net for 3D carotid artery segmentation in CTA, combining slice-aware, hierarchical, and boundary-oriented modules. The application is clinically relevant, the technical design is clearly motivated, and the empirical results are consistently strong across multiple datasets.

    After rebuttal, two reviewers (R2 and R3) are clearly positive, noting that the authors addressed the main concerns on overclaims, reproducibility, and evaluation details. R1 remains cautious about novelty and the justification of feature-level choices, but these reservations do not override the overall technical soundness and demonstrated practical value. The work is recognized as incremental rather than breakthrough, yet the integrated design delivers solid gains validated by thorough ablation and cross-dataset experiments.

    I recommend accept. For the camera-ready version, the authors must: (1) include boundary-sensitive metrics (HD95 or ASD) to substantiate the boundary refinement claims; (2) provide a clear justification for the selection of adjacent encoder feature levels and the operational definition of low-level vs. high-level features; and (3) moderate all remaining overclaims, particularly regarding novelty and task-specificity, in line with the rebuttal commitments.



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.

    This paper proposes MVDA-Net, a U-shaped architecture for 3D carotid artery segmentation in CTA imaging, built from three modules (SAOP, HIAR, and SFA) targeting global context, boundary refinement, and encoder-decoder feature alignment. The majority opinion is positive, though all three reviewers flag limited methodological novelty: the core ideas stay close to well-established strategies, and the contribution is largely one of combining existing components. R1 remains the most critical, noting that the rebuttal did not adequately justify the design choices or clarify operational definitions. R2 and R3 lean toward acceptance, with R2 reversing an initial negative assessment based on the rebuttal. All acknowledge that the method is technically sound and shows consistent gains across a private dataset and three public benchmarks, with an ablation supporting each module. On balance, the concerns over limited methodological novelty and the somewhat composite nature of the architecture do not outweigh the consistent results and the clinical value of the application. Assuming the authors sufficiently address the requested revisions—R1’s concerns on methodological ambiguity, removal of overstated claims, and stronger statistical and evaluation rigor—I lean toward acceptance.



Meta-review #3

  • After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.

    Reject

  • Please justify your recommendation.

    The reviewers have a consensus on the potential merit and good quality. But in terms of a large volume of medical image segmentation model papers, this paper only justified one task with an existing method with modification. The good quality of this paper still maintains its competitiveness.



back to top