List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Xiao Zhang, Jingyang Zhang, Lei Ma, Peng Xue, Yan Hu, Dijia Wu, Yiqiang Zhan, Jun Feng, Dinggang Shen
Abstract
Coronary artery segmentation is a critical yet challenging step in coronary artery stenosis diagnosis.
Most existing studies ignore important contextual anatomical information and vascular topologies, leading to limited performance.
To this end, this paper proposes a progressive deep-learning based framework for accurate coronary artery segmentation by leveraging contextual anatomical information and vascular topologies.
The proposed framework consists of a spatial anatomical dependency (SAD) module and a hierarchical topology learning (HTL) module.
Specifically, the SAD module coarsely segments heart chambers and coronary artery for region proposals, and captures spatial relationship between coronary artery and heart chambers.
Then, the HTL module adopts a multi-task learning mechanism to improve the coarse coronary artery segmentation by simultaneously predicting the hierarchical vascular topologies i.e., key points, centerlines, and neighboring cube-connectivity.
Extensive evaluations, ablation studies, and comparisons with existing methods show that our method achieves state-of-the-art segmentation performance.
Link to paper
DOI: https://link.springer.com/chapter/10.1007/978-3-031-16443-9_38
SharedIt: https://rdcu.be/cVRyS
Link to the code repository
N/A
Link to the dataset(s)
N/A
Reviews
Review #1
- Please describe the contribution of the paper
-Proposal of novel segmentation framework of coronary artery region. -SAD that consider relationships between large and small (artery) anatomical structures for segmentation was proposed. -HTL that effectively models topological characteristics of artery structure was proposed.
- 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.
-Anatomical structure relationships and topological features of segmentation target contribute to segment small targets. The proposed framework well combines this information with deep learning-based segmentation models. -The proposed method achieved better segmentation accuracy in the quantitative evaluation.
- 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.
-Adding evaluation results using public dataset is better.
- Please rate the clarity and organization of this paper
Very Good
- Please comment on the reproducibility of the paper. Note, that authors have filled out a reproducibility checklist upon submission. Please be aware that authors are not required to meet all criteria on the checklist - for instance, providing code and data is a plus, but not a requirement for acceptance
Reproducibility is OK. The authors will release codes and models.
- Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review: https://conferences.miccai.org/2022/en/REVIEWER-GUIDELINES.html
- Adding evaluation results using public dataset will clarify the effectiveness of the proposed method.
- Rate the paper on a scale of 1-8, 8 being the strongest (8-5: accept; 4-1: reject). Spreading the score helps create a distribution for decision-making
7
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
-Key factors to segment small segmentation target, including anatomical structure relationship and topological information, are integrated in the FCNs in the proposed framework. This is important work in segmentation of small anatomical structures. -Evaluation results prove effectiveness of the proposed method.
- Number of papers in your stack
4
- What is the ranking of this paper in your review stack?
1
- Reviewer confidence
Somewhat Confident
- [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed
Not Answered
- [Post rebuttal] Please justify your decision
Not Answered
Review #2
- Please describe the contribution of the paper
This paper proposed a progressive learning-based framework, a spatial anatomical dependency module and a hierarchical topology learning module to realize the accurate coronary artery segmentation.
- 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.
There are two main strengths of the paper: 1) This paper proposed a novel architecture that takes into account the dependency of vessel and chamber. 2) The HTL proposed in this paper improves the segmentation results.
- 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.
1) For the SAD module proposed in the method, if the coarse mask is severely fractured or has a large number of over segmentation, is the distance map accurate enough to be used for subsequent training? It is not mentioned in the discussion of the paper. Meanwhile, if the image is cropped, the chamber structure will be destroyed, can the distance map be used correctly? 2) HTL proposed in this paper is functionally more like an integrated learning or multi-task model, and there is no detailed description of how topological constraints are provided between tasks. 3) Inadequate experimental evaluation. This paper claimed to achieve the accurate segmentation, but the results did not show the segmentation performance for the regions with the stenosis. Meanwhile, this paper proposed the use of the topology, but the paper did not use OV, OF and other indicators to evaluate the continuity of main artery segmentation.
- Please rate the clarity and organization of this paper
Good
- Please comment on the reproducibility of the paper. Note, that authors have filled out a reproducibility checklist upon submission. Please be aware that authors are not required to meet all criteria on the checklist - for instance, providing code and data is a plus, but not a requirement for acceptance
The reproducibility of this paper is good, the framework described in this paper is very detailed and the algorithm is also very clear.
- Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review: https://conferences.miccai.org/2022/en/REVIEWER-GUIDELINES.html
1) The evaluation does not properly analyse a vessel-level view as adopted by the scoring methodology used in the 2008 and 2012 MICCAI challenges. This evaluation is especially crucial as the correctness of the stenoses segmentation is much more important than the correctness of the “healthy” part of the vessel. Hence, the authors should make an effort to provide results on cases with stenosis and justify objectively. 2) The author should pay attention to the expression in the figure. The distance map proposed in the method is dependent on the surface of the chamber, but the figure shows the center of mass. 3) The author should analyze the specific reasons why the method brings improvement, rather than engineering the method and boasting the advantages
- Rate the paper on a scale of 1-8, 8 being the strongest (8-5: accept; 4-1: reject). Spreading the score helps create a distribution for decision-making
3
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
- The two main reasons that influence my decision are: 1) the experiment in this paper is not sufficient to demonstrate the innovation points proposed by the author, and the experimental results did not show the segmentation quality for the lesion areas, which was insufficient to demonstrate the realization of accurate coronary segmentation. 2) the method proposed in this paper is not a very interesting innovation, and the use of the spatial dependency has been weakened.
- Number of papers in your stack
5
- What is the ranking of this paper in your review stack?
3
- Reviewer confidence
Very confident
- [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed
Not Answered
- [Post rebuttal] Please justify your decision
Not Answered
Review #3
- Please describe the contribution of the paper
This paper proposes a two-stage coronary artery segmentation task framework. The framework consists of a spatial anatomical dependency module and a hierarchical topology learning module. The former provides rough spatial localization and introduces cardiac anatomical information, while the latter emphasizes topological information by using multi-task CNN networks, which helps to maintain thin vessel continuity.
- 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.
- Clear structure: The workflow of proposed framework is clear and detailed, which visually illustrates the main method flow.
- Valuable motivation: The paper addresses the optimization of practical details in coronary artery segmentation problems.
- Detailed method: It provides a very detailed description of method details, including formulas and diagrams. The method is reasonable in design and specific in description.
- Intuitive experiment: Different evaluation indexes are compared objectively with those of advanced methods. The ablation experiment also demonstrates the necessity of each network module.
- 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.
There are some minor reference problems. Section 2.2 - Cube-Connectivity Prediction: <1> Qin, Y. , et al. “AirwayNet: A Voxel-Connectivity Aware Approach for Accurate Airway Segmentation Using Convolutional Neural Networks.” 2019.
- Please rate the clarity and organization of this paper
Very Good
- Please comment on the reproducibility of the paper. Note, that authors have filled out a reproducibility checklist upon submission. Please be aware that authors are not required to meet all criteria on the checklist - for instance, providing code and data is a plus, but not a requirement for acceptance
Although the code is not public, the authors list most of the details of the method in the paper. It should be straightforward to reproduce the results.
- Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review: https://conferences.miccai.org/2022/en/REVIEWER-GUIDELINES.html
-
References are incomplete. The key points and cube-connectivity branches in HTL module come from existing algorithms in the field of computer vision, which are applied to coronary artery segmentation in the paper. Reference <1> has proposed a similar approach to cube-connectivity branches. The paper adds a channel on the basis of 26 neighborhood, but the main idea is similar . <1> Qin, Y. , et al. “AirwayNet: A Voxel-Connectivity Aware Approach for Accurate Airway Segmentation Using Convolutional Neural Networks.” 2019.
-
The experimental results in this paper are very intuitive, and it will be better if the following supplements can be made in the future. It is suggested to add some intermediate result diagrams in the supplementary material in the future to show the interpretability of each branch, such as distance factory diagram and key point diagram. According to the network design in this paper, it may be helpful for some clinical research fields that want to directly output the centerline or key points of coronary artery.
-
- Rate the paper on a scale of 1-8, 8 being the strongest (8-5: accept; 4-1: reject). Spreading the score helps create a distribution for decision-making
7
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The methodology is in general well described and evaluation has been done comprehensively with very good results.
- Number of papers in your stack
4
- What is the ranking of this paper in your review stack?
1
- Reviewer confidence
Very confident
- [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed
Not Answered
- [Post rebuttal] Please justify your decision
Not Answered
Primary Meta-Review
- Please provide your assessment of this work, taking into account all reviews. Summarize the key strengths and weaknesses of the paper and justify your recommendation. 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 presents a coronary artery segmentation method based on the hierarchical topology learning.
Strength: 1) sound method 2) better than the comparative methods 3) well-written
Weakness: 1) The claimed contribution, and the novelty of the network modules should be further explained.
Three reviewers have given the following comments: 1) R1 mainly considers the better performance, but lack of the validation on the public dataset. 2) R2 mainly considers the distance map in SAD module is not robust in the diverse conditions, no detailed presentation to explain the topological constraints 3) R3 mainly considers the valuable motivation, clear structure, detailed method and intuitive experiments.
- What is the ranking of this paper in your stack? Use a number between 1 (best paper in your stack) and 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).
2
Author Feedback
We thank AE and all the reviewers for their efforts and insightful comments. We address their major concerns below.
Q1: Clarification on claimed contribution and technical novelty of our method (Meta, R2) Our goal is to develop a novel method for coronary artery segmentation, by exploiting the artery-to-heart anatomical dependency, combined with a thorough intra-artery hierarchical topology (i.e., key points, centerline, and cube-connectivity). Our method can achieve promising results, even for challenging cases such as thin vessels. Besides, our method can also facilitate clinical applications such as surgical planning where surgeons need to avoid hurting the vessels. Also, coronary segmentation is important for coronary artery stenosis (CAS) grading quantification (as mentioned in the first few sentences of our paper, which may be not clear in previous version). And now we are working on CAS diagnosis as our long-term goal and conducting further research.
Q2: Availability of public dataset (Meta, R1) The lack of public coronary datasets has always been a problem in the field. To our knowledge, there are only two large public datasets, i.e., “ASOCA” and “orCaScore”. However, they are not available at present. Therefore, we had to collect a large private dataset and annotate them by experienced doctors (Section 3.1).
Q3: Robustness of distance map calculation in the SAD module (Meta, R2) The distance map is calculated from full-size image and then cropped into patches (Fig 2). Hence, the heart chamber will always be intact when distance map is calculated. In particular, the distance map is calculated utilizing the coarsely-segmented coronary artery, to derive a soft mask where the coronary artery is denoted by the distance with normalized value from 0 to 1 (Eq.2). This soft mask provides spatial localization of vessels. We then concatenate the distance map with the original image, and feed them together into the second stage network to learn hierarchical topological representations and refine coarse segmentation results by fine-grained information, e.g., key points, centerlines and cube-connectivity. The inaccuracy of coarse segmentation can be further refined by the HTL module, as confirmed by results in the ablation study (Table 2).
Q4: Detailed description of HTL module, e.g., implementation of topological constraints (Meta, R2) Our HTL module imposes topological constraints by performing multiple topology-related tasks in the following 3 branches: 1) detection of key points for tree-topology delineation; 2) regression of centerline heatmap for skeleton tracking; and 3) prediction of cube-connectivity for spatial connectivity. Such three topology-related tasks would make the network aware of the vessel topology at 3 levels, which provides topological constraints to regularize the segmentation results. We also explore integration of proposed topological constraints by using a shared decoder. However, representations of key points detection and centerline regression are sparser than the cube-connectivity prediction, causing bias in integration. The results show that separately processing topological information of each level can achieve the highest Dice gain of 7%. By contrast, a shared decoder only achieves a Dice gain of 2.6%.
Q5: Justification on the choice of evaluation metrics (R2) We evaluate all methods by employing all metrics of “2020 MICCAI Challenge-ASOCA”. Moreover, we further adopt common metrics, i.e., TPR, mIoU, and ASD in coronary artery segmentation [1], to highlight the effectiveness of the proposed method. To our understanding, the OV is similar to the Dice score [2], while metrics such as OF are often used to evaluate the accuracy of centerline extraction instead of vessel segmentation [2-3]. [1] j.cmpb.2018.02.001 [2] j.media.2009.06.003 [3] j.media.2018.10.005
Post-rebuttal Meta-Reviews
Meta-review # 1 (Primary)
- Please provide your assessment of the paper taking all information into account, including rebuttal. Highlight the key strengths and weaknesses of the paper, clarify how you reconciled contrasting review comments and scores, indicate if concerns were successfully addressed in the rebuttal, and provide a clear justification of your decision. If you disagree with some of the (meta)reviewer statements, you can indicate so in your meta-review. Please make sure that the authors, program chairs, and the public can understand the reason for your decision.
The rebuttal addresses my concerns.The author explains in detail the motivation and the innovation. At the same time, the author also explained the role of each module and the relationship.
- After you have reviewed the rebuttal, please provide your final rating based on all reviews and the authors’ rebuttal.
Accept
- 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).
2
Meta-review #2
- Please provide your assessment of the paper taking all information into account, including rebuttal. Highlight the key strengths and weaknesses of the paper, clarify how you reconciled contrasting review comments and scores, indicate if concerns were successfully addressed in the rebuttal, and provide a clear justification of your decision. If you disagree with some of the (meta)reviewer statements, you can indicate so in your meta-review. Please make sure that the authors, program chairs, and the public can understand the reason for your decision.
The main strength of this work is to present an - up to my knowledge - original contribution to the problem of segmenting coronary arteries, which may be relatively small vessels. While to some degree engineered, the methodology still seems to be a valuable contribution in this area, which has high clinical relevance.
While the overall evaluation and comparison to state of the art seems convincing to me, the main weakness is the lack of a detailed evaluation of stenosis sections of coronary arteries, as it was indicated by R2. Unfortunately, the rebuttal does not address this concern.
Overall, I am still voting for acceptance, since I think the strength slightly outweighs this weakness, and I would see the work as a starting point to perform such more detailed, practically relevant evaluations.
- After you have reviewed the rebuttal, please provide your final rating based on all reviews and the authors’ rebuttal.
Accept
- 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).
1
Meta-review #3
- Please provide your assessment of the paper taking all information into account, including rebuttal. Highlight the key strengths and weaknesses of the paper, clarify how you reconciled contrasting review comments and scores, indicate if concerns were successfully addressed in the rebuttal, and provide a clear justification of your decision. If you disagree with some of the (meta)reviewer statements, you can indicate so in your meta-review. Please make sure that the authors, program chairs, and the public can understand the reason for your decision.
The rebuttal addressed major concerns raised by reviewers and AC. The designed method is well-motivated and presented. The results show obvious improvements. Although there lacks a detailed evaluation of stenosis (which can be important in real-world application), the current paper is in good shape for MICCAI.
- After you have reviewed the rebuttal, please provide your final rating based on all reviews and the authors’ rebuttal.
Accept
- 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).
1