List of Papers Browse by Subject Areas Author List
Abstract
The automatic and accurate segmentation of the carotid artery vessel wall can assist doctors in clinical diagnosis. Medical images often have complex and blurry features, which makes manual data annotation very difficult and time-consuming. 3D CNN can utilize three-dimensional spatial information to more accurately identify diseased tissues and organ structures, but its segmentation performance is limited due to the lack of global contextual information correlation. This paper proposes a network based on CNN and Transformer to segment the carotid artery vessel wall. By combining the effectiveness of CNN in dealing with 3D image segmentation problems and the global attention mechanism of Transformer, it is possible to better capture and process the features of this information. By designing Joint Attention Structure Block (JAS), semantic information in skip connections can be enhanced. The feature fusion block (FF) is used to associate input information with each layer of feature maps, enhancing the detailed information of the feature maps. The effectiveness of this method has been verified through a large number of comparative experiments.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2024/paper/1398_paper.pdf
SharedIt Link: https://rdcu.be/dVZeA
SpringerLink (DOI): https://doi.org/10.1007/978-3-031-72378-0_33
Supplementary Material: N/A
Link to the Code Repository
N/A
Link to the Dataset(s)
N/A
BibTex
@InProceedings{Lei_Weaksupervised_MICCAI2024,
author = { Lei, Haijun and Tong, Guanjiie and Su, Huaqiang and Lei, Baiying},
title = { { Weak-supervised Attention Fusion Network for Carotid Artery Vessel Wall Segmentation } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
year = {2024},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15001},
month = {October},
page = {349 -- 358}
}
Reviews
Review #1
- Please describe the contribution of the paper
This paper proposes a novel network that integrates CNN and Transformer for segmenting the carotid artery vessel wall. The design includes a Joint Attention Structure Block (JAS) and a Feature Fusion block (FF), which aim to enhance semantic information in skip connections and detail in feature maps. However, a critical review reveals a significant gap in the discussion on the weak-supervised aspect, which is central to the paper’s title and claimed contributions.
- Please list the main strengths of the paper; you should write about 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 innovative combination of CNN with Transformer to capture and process features for carotid artery vessel wall segmentation.
- Design of JAS and FF blocks for enhanced feature processing.
- Empirical validation through comparative experiments.
- Please list the main weaknesses of the paper. Please provide details, for instance, if you think a method is not novel, explain why and provide a reference to prior work.
- The paper significantly lacks in-depth discussion and demonstration of the weak-supervised learning aspect, which is crucial given its prominence in the title and abstract.
- Absence of comparison with recent related works, particularly those employing deep learning for vessel wall segmentation.
- Limited discussion on the method’s generalizability across different types and qualities of medical images.
- Experimental section lacks diversity in datasets and real clinical application scenarios.
- 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.
- Do you have any additional comments regarding the paper’s reproducibility?
It is suggested that the authors provide access to the source code and dataset upon acceptance, which would greatly enhance the transparency and verifiability of the work.
- Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html
The paper presents an approach combining CNN and Transformer for carotid artery vessel wall segmentation. While the integration of these technologies is innovative, the paper falls short in discussing and demonstrating the weak-supervised learning aspect, a critical component of the proposed method. The absence of this discussion undermines the paper’s contribution and relevance to its stated objectives. Further, the lack of comparison with current related works and discussion on generalizability and practical application limits the paper’s impact. It is recommended that the authors address these significant gaps and consider the method’s weak-supervised aspect more thoroughly in future work.
- 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
Reject — should be rejected, independent of rebuttal (2)
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The paper’s failure to adequately discuss and demonstrate the weak-supervised learning aspect, which is central to its premise, significantly diminishes its contribution to the field. Combined with the absence of comparisons with recent related works and a lack of discussion on generalizability and practical application, these gaps warrant a rejection of the paper. It is crucial for the authors to address these issues comprehensively in any future submissions.
- Reviewer confidence
Confident but not absolutely certain (3)
- [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed
Reject — should be rejected, independent of rebuttal (2)
- [Post rebuttal] Please justify your decision
The paper still does not explained the weak-supervised in the title clearly The method used in the article is more of a post-processing operation than the weak-supervised paradigm[1].
[1] https://en.wikipedia.org/wiki/Weak_supervision
Review #2
- Please describe the contribution of the paper
This manuscript introduces a network that combines CNN and Transformer architectures for the segmentation of the carotid artery vessel wall. This innovative approach enables the segmentation of relatively complete carotid artery vessel walls from coarse labels.
- Please list the main strengths of the paper; you should write about a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
This manuscript presents a novel formulation for effectively implementing a segmentation architecture for the carotid artery vessel. It combines CNN and Transformer technologies and introduces a Joint Attention Structure Block and a Feature Fusion Block to enhance the network’s capabilities. The effectiveness of this approach is validated through detailed experimental verification.
- Please list the main weaknesses of the paper. Please provide details, for instance, if you think a method is not novel, explain why and provide a reference to prior work.
While the manuscript integrates CNN and Transformer for segmenting the carotid artery vessel wall, this combination does not present a distinct novelty. Similar approaches have already been explored in the field of medical image segmentation. The descriptions and diagrams pertaining to the Joint Attention Structure Block and Feature Fusion Block lack clarity and detail, impeding a full understanding of their operational mechanics and theoretical underpinnings. Furthermore, although the paper claims extensive experimental validation of its approach, it fails to provide a rigorous comparative analysis with state-of-the-art methods. Such a comparison is essential to clearly demonstrate any advantages or improvements. Specifically, the paper does not compare its results with those of other advanced Transformer-based segmentation methods, which could help establish the relative merit of the proposed approach.
- 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.
- Do you have any additional comments regarding the paper’s reproducibility?
N/A
- Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html
I recommend that the authors enhance the descriptions of the Joint Attention Structure Block and Feature Fusion Block by providing more detailed explanations, including specific mathematical formulations, and offering more detailed illustrations in the diagrams. Additionally, it is advisable to include comparative experiments involving Transformer-related methods to more effectively highlight the advantages or distinctiveness of the proposed approach compared to current state-of-the-art techniques.
- 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
Weak Accept — could be accepted, dependent on rebuttal (4)
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The method proposed in the manuscript facilitates the segmentation of relatively complete carotid artery vessel walls from coarse labels. However, the implementation lacks sufficient novelty. The authors should enhance the paper’s clarity and technical rigor by providing more detailed diagrams and step-by-step explanations of the key components. While the experimental results, including ablation and comparative studies, demonstrate positive outcomes and visualizations, the design of the comparative experiments remains incomplete. The authors are encouraged to conduct further comparisons with more advanced and recent algorithms to clearly delineate the advantages and innovations of their approach over existing methods.
- Reviewer confidence
Confident but not absolutely certain (3)
- [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed
Weak Accept — could be accepted, dependent on rebuttal (4)
- [Post rebuttal] Please justify your decision
The author’s rebuttal has addressed some of my questions; however, the organization and structure of the paper still lack clarity.
Review #3
- Please describe the contribution of the paper
This paper introduces a weak-supervised attention fusion network for carotid artery vessel wall segmentation. It highlights the importance of accurate segmentation in assisting clinical diagnosis and proposes a network based on CNN and Transformer to address the limitations of 3D CNN segmentation performance. The study focuses on the significance of carotid artery imaging in diagnosing atherosclerosis and stroke and explores the use of medical imaging methods such as MRI, CT, X-ray, and ultrasound. The paper presents the network architecture, including the Joint Attention Structure Block and Feature Fusion Block, and details the experimental validation of the proposed method, demonstrating its effectiveness in segmenting carotid artery vessel walls from rough labels.
- Please list the main strengths of the paper; you should write about 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 presents a weak-supervised attention fusion network designed for precise carotid artery vessel wall segmentation . This novel network is notable for its utilization of a combination of convolutional neural networks (CNN) and Transformer architectures, offering a more effective approach to segmenting arterial blood vessels. Notably, the study integrates the Vision Transformer module with the VNet model, enhancing the ability to process 3D image segmentation challenges.
- Please list the main weaknesses of the paper. Please provide details, for instance, if you think a method is not novel, explain why and provide a reference to prior work.
The paper does not provide a detailed comparative analysis with existing weakly-supervised methods for carotid artery vessel wall segmentation.
- 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 provide sufficient information for reproducibility.
- Do you have any additional comments regarding the paper’s reproducibility?
N/A
- Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html
1.Strengthen the paper’s originality by including a more comprehensive comparative analysis with existing weakly-supervised methods for carotid artery vessel wall segmentation. This would demonstrate the unique advantages of the proposed method and position it clearly within the existing landscape of segmentation techniques. 2.The best result in Table 2 is not highlighted. 3.To implement rough annotation of labels in the data section, a detailed explanation should be provided. 4.Conducting additional comparative experiments to evaluate the performance of networks specifically designed for carotid artery segmentation.
- 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
Weak Accept — could be accepted, dependent on rebuttal (4)
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The absence of a detailed comparative analysis with existing weakly-supervised methods limits the comprehensive understanding of the method’s advantages over alternative approaches and weakens the paper’s scientific rigor.
- Reviewer confidence
Confident but not absolutely certain (3)
- [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed
Weak Accept — could be accepted, dependent on rebuttal (4)
- [Post rebuttal] Please justify your decision
The author’s rebuttal addressed my doubts and I agree to accept it.
Author Feedback
We sincerely thank all reviewers for their valuable comments. We have made revisions to the grammar and content issues raised one by one. There are our feedbacks for the major weaknesses. Method details(R1, R3, R4): This study proposes an attention fusion-based segmentation network for precise segmentation of 3D carotid artery vessel walls. The network integrates ViT’s global attention with VNet’s local feature extraction, overcoming VNet’s limitation in capturing long-range dependencies. A joint attention module and feature fusion module strengthen semantic information transfer and detail capture, enabling more precise vessel contour depiction. Comparison methods (R1, R3, R4): In Table of the comparison methods we have chosen relevant works for comparison. Due to different data types and the lack of open-source code, we did not replicate these methods on our dataset for experimentation. JAS(R1, R3): This study proposes a joint attention-based feature fusion module for medical image segmentation, integrating multi-scale dilated convolutions and attention. It enhances generalization by fusing features from parallel dilated convolutions. A self-attention fusion and SimAM module boost key feature representation, effectively combining multi-scale features and attention. This improves segmentation performance by emphasizing critical features. Weak-supervied(R1, R4): This study employed a segmentation method based on coarse labels, complemented by specific post-processing steps to enhance segmentation accuracy. Initially, morphological erosion was applied to remove small details and isolated pixels, thus reducing mis-segmentation. Subsequently, dilation was utilized to fill in holes, connect fragmented regions, and restore object shapes. In the post-processing phase, the largest connected region was constructed, and by comparing the sizes of different regions, smaller, potentially mis-segmented areas were identified and eliminated. This approach effectively integrates initial segmentation with post-processing steps, significantly improving the accuracy and reliability of the segmentation results. Particularly in complex medical images, such precise segmentation is crucial for diagnosis and treatment, ultimately yielding refined segmentation outcomes. Rough annotation(R4): In the backbone network presented in the document, you may observe the utilization of coarse-grained labels. However, taking into account the consideration of the layout and formatting, the paper has refrained from separately listing the coarse-grained labels prior to processing. Data section(R4):During the label processing, we gracefully resize the initial labels from 432x432x432 to 412x512x512, then refine them with a 3D median filter to smoothen jagged edges from interpolation. Finally, we employ graphics-based opening and closing operations to fill in any voids on the vessel surface, producing refined training labels. We appreciate your attention and consideration in this matter.
Meta-Review
Meta-review #1
- 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. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
The authors are advised to change the title of their manuscript, as highlighted by #R1 (less emphasis on weak-supervision)
- What is the rank of this paper among all your rebuttal papers? Use a number between 1/n (best paper in your stack) and n/n (worst paper in your stack of n papers). If this paper is among the bottom 30% of your stack, feel free to use NR (not ranked).
The authors are advised to change the title of their manuscript, as highlighted by #R1 (less emphasis on weak-supervision)
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. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
None
- What is the rank of this paper among all your rebuttal papers? Use a number between 1/n (best paper in your stack) and n/n (worst paper in your stack of n papers). If this paper is among the bottom 30% of your stack, feel free to use NR (not ranked).
None