List of Papers Browse by Subject Areas Author List
Abstract
In recent years, deep learning-based vessel segmentation methods have made significant progress. However, the diversity of image modalities and the high-cost of acquiring sufficient annotated data constrain the performance of existing approaches. Given that the primary objective of segmenting various types of vessels is to extract high-frequency tubular structures, leveraging existing annotated datasets for training and fast generalizing to novel vessel segmentation tasks is an ideal solution to the above challenges, which can be achieved by the few-shot segmentation (FSS) paradigm. Unfortunately, the significant differences in texture and thickness among different types of vessels leave unsolved challenges. To address this issue, we propose a novel framework that incorporates FSS into cross-domain vessel segmentation. In particular, we construct high-frequency auxiliary modalities to guide the model in focusing on high-frequency features, which are highly correlated with vessel regions, thereby bridging the texture gap between images of various vessel types. Furthermore, we design a Dual-Modal Feature Extraction and Fusion (DM-FEF) module to extract modality-specific features. Finally, addressing the thickness variations between different vessels, we designed a Multi-Branch Feature Extractor (MBFE) module to capture the diverse characteristics of vessels with different thickness, enabling the model to perceive the thickness differences between distinct vessels. Experimental results on six public datasets demonstrate the effectiveness of our method. Source code: https://github.com/ZiH-Huang/FSS_Cross.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/0782_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/ZiH-Huang/FSS_Cross.
Link to the Dataset(s)
N/A
BibTex
@InProceedings{HuaZih_AMultiBranch_MICCAI2025,
author = { Huang, Zihang and Zhao, Tianyu and Zhang, Liang and Yang, Xin},
title = { { A Multi-Branch Framework for Cross-Domain Vessel Segmentation via the Few-Shot Paradigm } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15964},
month = {September},
page = {12 -- 22}
}
Reviews
Review #1
- Please describe the contribution of the paper
This paper proposes a novel framework that integrates few-shot segmentation (FSS) with cross-domain vessel segmentation.. The authors introduce three core components:
- High-Frequency Auxiliary Modality: Enhances high-frequency, vessel-relevant features through a Mixup-based fusion of saliency and Laplacian transformations.
- Dual-Modal Feature Extraction and Fusion: A dual-branch module with cross-attention to extract and combine features from original and auxiliary images.
- Multi-Branch Feature Extractor: Separates vessels by thickness (thin, thick, and all) with dedicated branches, improving performance on structures of varying sizes.
The proposed method outperforms 10 state-of-the-art baselines across six public datasets in three challenging cross-domain settings. Extensive ablation studies support the contribution of each module.
- 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.
Experiments: Evaluated on six diverse datasets, the framework demonstrates robust performance across domains and modalities. The comparison includes both classical, generalization, and few-shot baselines, providing a fair and comprehensive benchmark. Also the detailed ablations quantify the contribution of each architectural component.
- 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 proposed framework is overly complex, which may hinder its ease of training and limit its practical applicability.
- The model produces three separate outputs, but the authors do not clearly specify how these outputs are used during inference, or how the thin and thick vessel predictions are combined during testing.
- 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.
- 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.
(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?
Although the complexity of the method may affect both readability and practical applicability, the authors present comprehensive experiments that demonstrate its effectiveness. Overall, I would suggest a weak acceptance.
- Reviewer confidence
Somewhat confident (2)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
N/A
- [Post rebuttal] Please justify your final decision from above.
N/A
Review #2
- Please describe the contribution of the paper
- A cross-domain vascular segmentation approach is proposed by integrating the few-shot segmentation (FSS) paradigm to address the variations across imaging modalities and vessel types.
- A high-frequency auxiliary modality is introduced to guide the model in focusing on vascular high-frequency structures, improving adaptability across different domains.
- A Dual-Modality Feature Extraction and Fusion (DM-FEF) module is designed to effectively combine information from both the original and high-frequency modalities for better feature representation.
- A Multi-Branch Feature Extractor (MBFE) module is proposed to distinguish vascular structures of varying thickness, enhancing the model’s generalization capability.
- 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.
- It is possibly the first to incorporate FSS into the cross-domain vessel segmentation task, showing a certain degree of novelty. 2.The paper is well-structured and has good readability. 3.Through comprehensive experiments on six public vessel datasets, the paper effectively verifies the advantage of its method in addressing cross-domain few-shot segmentation.” 4.The ablation study also demonstrates the effectiveness of each module and design.
- 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 structure of the MBFE module is generally well defined, the descriptions of its branches are somewhat brief. For example, the statement “Ball provides global cues” could be further elaborated—specifically, how Ball integrates or interacts with Bthick and Bthin should be clarified. Additionally, the prototype sampling strategy for support samples lacks justification for setting N = 400. It would be helpful to explain whether this value was empirically chosen and whether the method is sensitive to this parameter.
-
The structure of the decoders used in Equations (5) and (6) is not clearly described. It is unclear whether they are implemented using simple convolutional layers, MLPs, or another architecture. It should also be stated whether the decoders across different branches share parameters.
-
In Equation (8), the loss term appears as L_thick_dice instead of L_dice_thick, which may be a formatting issue. For clarity and consistency, it is recommended to use uniform notation throughout the paper.
-
- Please rate the clarity and organization of this paper
Good
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The 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
see weaknesses
- 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?
Despite some minor weaknesses (see comments above), the paper is overall solid and well-executed.
- 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 #3
- Please describe the contribution of the paper
This paper presents a method for vessel segmentation, intended to improve the generalizability of deep learning-based methods with limited training data, considering the various characteristics of images from different domains. Using a few-shot segmentation approach, the method only relies on 1 pair of support image and its label for segmenting a new input image. Using targeted designs for input (image together with basic pre-processed contrast enhanced features), network (multiple branches comprising simple residual blocks, inception residual blocks, and cross-attention layers), and loss (separate loss functions for thin and thick vessels, similarity comparison with prototype features of support image), the method demonstrates considerably improved accuracy compared to existing state-of-the-art.
- 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.
S1. Achieves strong segmentation performance, with key metrics such as DICE score showing significant improvement over existing state-of-the-art methods. S2. Comprehensive ablation study, clearly demonstrating the contribution of each component to the overall performance improvement. S3. Technical contribution in the overall framework, combining the targeted designs for input (image together with basic pre-processed contrast enhanced features), network (multiple branches comprising simple residual blocks, inception residual blocks, and cross-attention layers), and loss (separate loss functions for thin and thick vessels, similarity comparison with prototype features of support image).
- 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.
W1. The presentation of the technical contribution could be improved. While individual components—such as saliency detection, the network architecture, separation of thin and thick vessels, and the use of prototype features—are not novel, their combination appears to be. Rather than presenting each element as a standalone contribution, the authors should emphasize that the novelty lies in the integration of these components within a unified framework for few-shot segmentation, and then describe the subcomponents accordingly. W2. The presentation of the experimental evaluations could be strengthened. Performance may vary depending on the choice of support images, which is only briefly addressed by mentioning an average over three tests. The authors are encouraged to increase the number of runs and report standard deviations to provide a clearer picture of variability. Additionally, only a single qualitative example is shown per evaluation dataset; including more examples—especially failure cases—would offer valuable insight. Lastly, instead of using vague labels like ‘Setting 1’, more intuitive notations such as XCAD/XCA → OCTA_X would enhance readability and clarity. W3. The saliency detection component of HFAM could be enhanced, considering the extensive body of research on vascular structure enhancement using various filtering techniques. Methods such as Hessian-based filtering, while mentioned only as a baseline, could be more deeply explored or integrated to strengthen the saliency module.
- 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?
The strong segmentation performance and the comprehensive method framework lead me to my recommendation, despite the minor shortcomings of the evaluation.
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
N/A
- [Post rebuttal] Please justify your final decision from above.
N/A
Author Feedback
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”.
N/A