List of Papers Browse by Subject Areas Author List
Abstract
Accurate vessel segmentation is critical for diagnosis. However, the annotation of vascular images cost a lot, and due to their diverse modalities and complex foreground structures, it is hard for learning-based methods to reduce annotation cost by training models of high domain generalization (DG) on partial modalities. To address this, we propose the Image-Sparse Annotation Completion (ISAC) segmentation model, which reformulates vascular segmentation as a mask completion task based on sparse-annotated supports. ISAC treats the segmentation task as incomplete mask reconstruction guided by image features and structural properties of the foreground in the sparse mask. Unlike pixel-wise classification, ISAC detects vessels according to the mask context supported regions, in which way the anatomical continuity of vascular foreground is improved. Additionally, to further avoid the reliance on high-cost manually annotated supports, we propose the Uncertainty-guided Patch Selection (UPS) module to extract high-quality supports from coarse pseudo labels, which enables ISAC to perform segmentation in zero-shot scenarios. Experiments on 7 vascular datasets across 3 modalities demonstrate that ISAC outperforms state-of-the-art methods in DG ability. The code is publicly available at https://github.com/Architect15806/ISAC.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/3384_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/Architect15806/ISAC
Link to the Dataset(s)
N/A
BibTex
@InProceedings{ZhaTia_ISAC_MICCAI2025,
author = { Zhao, Tianyu and Huang, Zihang and Jiang, Xixi and Zhang, Liang and Ding, Xiaohuan and Yang, Xin},
title = { { ISAC: Redefining the Vascular Segmentation Paradigm through Mask Completion for Cross-Domain Generalization } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15966},
month = {September},
page = {306 -- 316}
}
Reviews
Review #1
- Please describe the contribution of the paper
The paper proposes ISAC, a novel segmentation model that reformulates domain generalization (DG) for vascular segmentation as a mask completion task guided by Image-Sparse Annotation Completion. It introduces two major components: (1) a dual-branch encoder and support-guided decoder that fuses full images and sparse annotations via cross-attention; and (2) a Uncertainty-guided Patch Selection (UPS) module that selects high-quality support regions from pseudo labels for zero-shot segmentation. Extensive experiments on 7 datasets across 3 imaging modalities (Fundus, OCTA, and X-ray) demonstrate strong DG ability and state-of-the-art performance under zero-shot settings.
- 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.
Novel task formulation: Reformulating segmentation under sparse support as a mask completion problem guided by sparse patches is a refreshing alternative to conventional pixel-wise classification. Domain generalization capability with practical utility: The method enables segmentation on unseen domains without requiring too much manual annotations, a highly valuable property for real-world applications. UPS module: The patch selection mechanism using pseudo-label uncertainty and prototype similarity is novel and empirically effective. Strong experimental coverage: Evaluation spans three modalities and seven datasets, with performance surpassing existing DG methods.
- 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.
While the paper proposes a novel task formulation by treating vascular segmentation as a sparse mask completion problem, the methodological novelty remains limited. The overall framework primarily integrates existing components—such as CroCo for cross-view completion, DPT for dense prediction, and prototype-based uncertainty filtering—for a new task setting. These modules are adapted rather than fundamentally redesigned. Moreover, the motivation for the proposed formulation is not sufficiently grounded in the paper. It lacks a clear explanation of why existing domain generalization methods are inadequate and how the proposed mask completion strategy fundamentally addresses their limitations. The UPS module, though empirically useful, is designed heuristically, and key design choices (e.g., patch selection thresholds, cosine similarity) are not theoretically justified. Overall, the paper would benefit from a deeper methodological innovation and stronger conceptual motivation for its core contributions.
- 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.
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2025/en/REVIEWER-GUIDELINES.html
N/A
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(3) Weak Reject — could be rejected, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The paper proposes ISAC, a segmentation framework that reformulates vascular segmentation as a sparse-mask completion task to enhance domain generalization (DG). The idea of using sparse annotation and uncertainty-based patch selection is interesting and practically useful, and the method shows solid performance across multiple datasets. However, I have concerns about the level of methodological novelty. The overall system largely repurposes existing architectures (CroCo, DPT) and standard uncertainty-guided pseudo-labeling strategies, without introducing fundamentally new algorithmic components. Moreover, the motivation for the problem reformulation is not sufficiently discussed, and several design choices are heuristic and lack theoretical justification. While the results are encouraging, I do not believe the contribution is substantial enough for MICCAI acceptance in its current form. I recommend rejection, albeit weakly, as the paper has some potential if the novelty and motivation are further developed.
- 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 #2
- Please describe the contribution of the paper
This paper presents a Image-Sparse Annotation Completion (ISAC) segmentation model for domain generalized vascular segmentation. Specifically, ISAC treats the segmentation task as incomplete mask reconstruction guided by image features and structural properties of the foreground in the sparse mask. The sparse foreground mask of the new domain image can be provided manually or selected from pseudo labels via the proposed Uncertainty-guided Patch Selection (UPS) module. The proposed method has been evaluated on 7 vascular datasets across 3 modalities against other 4 domain generalization methods.
- 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 idea of formulating domain generalized vascular segmentation as a mask completion task based on target domain sparse support is interesting.
- The proposed UPS module allows the model to perform segmentation in the target domain without the need for manually annotated support.
- Experimental results demonstrate the superiority of the proposed method against other four competing methods.
- 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 concept of utilizing target domain information to enhance domain-generalized segmentation has been introduced in [1], but this paper is not properly cited. [1] Hu S, Liao Z, Zhang J, et al. Domain and content adaptive convolution based multi-source domain generalization for medical image segmentation[J]. IEEE Transactions on Medical Imaging, 2022, 42(1): 233-244.
- The proposed ISAC depends on an additional segmentation model (SEG in Fig. 1) to provide sparse annotations for the test image, which may limit its applicability.
- In Table 3, why is the performance before and after using MCA (without UPS) similar? This is confusing.
- In Fig. 3, why does the performance of the blue lines slightly decrease as \sigma increases?
- 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 authors claimed to release the source code and/or dataset upon acceptance of the submission.
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2025/en/REVIEWER-GUIDELINES.html
The paper proposes a new perspective by treating domain-generalized vascular segmentation as a sparse mask completion task. While the idea is promising and results are strong, the dependency on an auxiliary model and lack of clarity in some experimental results (e.g., Table 3, Fig. 3) need further discussion. Citation to relevant prior work should also be added.
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(4) Weak Accept — could be accepted, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The paper proposes a novel formulation of domain generalized vascular segmentation as sparse mask completion, supported by a UPS module that reduces manual effort. The method shows strong performance across diverse datasets. However, key prior work is not cited, the dependency on an external segmentation model limits practicality, and some results lack clear explanation. These points led to a weak accept.
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
N/A
- [Post rebuttal] Please justify your final decision from above.
N/A
Review #3
- Please describe the contribution of the paper
The paper addresses the challenge of segmenting vascular objects where existing segmentation methods’s segmentation results have gaps along the objects. The paper proposed to reformulate the problem as a mask completion problem. It uses uncertainty score to locate the low confidence areas and Integrating of CroCo-style masked cross-attention and dual-branch ViT encoders to help preserve vascular continuity. The experiments are comprehensive and outperform existing methods.
- 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.
- Reformulating vascular segmentation as mask completion based on sparse support is a creative and well-motivated departure from conventional ISC (image-semantic correspondence) approaches.
- Integration of CroCo-style masked cross-attention and dual-branch ViT encoders is well-justified. The masked pathway design addresses sparsity in support encoding.
- The uncertainty-aware selection of pseudo-support patches is a strong addition. It allows the model to work without annotations at test time, alleviating annotation costs.
- 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.
Complexity and Efficiency: the model architecture is very complex with numerous different modules and steps. It would be important to know the inference time compared to existing 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 authors claimed to release the source code and/or dataset upon acceptance of the submission.
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2025/en/REVIEWER-GUIDELINES.html
N/A
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(5) Accept — should be accepted, independent of rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Although the overall method is very complex, the idea of formulating segmentation problem as a mask completion problem is novel and inspiring. The author also conducted numerous experiments to show the effectiveness of the methods. However, computation time at inference time would be a major concern.
- 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
- If your recommendation is “Provisional Reject”, then summarize the factors that went into this decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. You do not need to provide a justification for a recommendation of “Provisional Accept” or “Invite for Rebuttal”.
This work received mixed reviews, with the key idea of reformulating vascular segmentation as a mask completion task. It is considered novel and valuable by all reviewers and the proposed ISAC framework demonstrates strong generalization performance across multiple domains and modalities, which is potentially impactful for the community. Therefore, a Provisional Accept decision is given.