List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Yifan Liu, Jie Liu, Yixuan Yuan
Abstract
Point-based 3D intracranial aneurysm segmentation is fundamental for automatic aneurysm diagnosis. Though impressive performances, existing point-based 3D segmentation frameworks still perform poorly around the edge between vessels and aneurysms, which is extremely harmful for the clipping surgery process. To address the issue, we propose an Edge-oriented Point-cloud Transformer Network (EPT-Net) to produce precise segmentation predictions. The framework consists of three paradigms, i.e., dual stream transformer (DST), outer-edge context dissimilation (OCD) and inner-edge hard-sample excavation (IHE). In DST, a dual stream transformer is proposed to jointly optimize the semantics stream and the edge stream, where the latter imposes more supervision around the edge and help the semantics stream produce sharper boundaries. In OCD, aiming to refine features outside the edge, an edge-separation graph is constructed where connections across the edge are prohibited, thereby dissimilating contexts of points belonging to different categories. Upon that, graph convolution is performed to refine the confusing features via information exchange with dissimilated contexts. In IHE, to further refine features inside the edge, triplets (i.e. anchor, positive and negative) are built up around the edge, and contrastive learning is employed. Differently from previous contrastive methods of point clouds, we only select points nearby the edge as hard-negatives, providing informative clues for discriminative feature learning. Extensive experiments on the 3D intracranial aneurysm dataset IntrA demonstrate the superiority of our EPT-Net compared with state-of-the-art methods.
Link to paper
DOI: https://link.springer.com/chapter/10.1007/978-3-031-16443-9_10
SharedIt: https://rdcu.be/cVRyo
Link to the code repository
https://github.com/CityU-AIM-Group/EPT
Link to the dataset(s)
https://github.com/intra3d2019/IntrA
Reviews
Review #1
- Please describe the contribution of the paper
The authors present a novel point-based 3D aneurysm segmentation using transformer. The proposed method consists of three major components: 1) dual stream transformer for both semantic segmentation and edge classification, 2) edge context dissimilation achieved by graph convolution and 3) hard sample mining of edge points by constructive learning. The proposed method is evaluated on a public dataset. Experiments show that the presented method provides promising results.
- 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 strengths of the paper:
1: The authors present a novel method for intracranial aneurysm segmentation by edge-oriented processes. These processes include edge point classification by point-based transformer, edge point refinement by graph convolution network and hard sample mining around target boundary.
2: The authors evaluate the method on a public dataset compared to previously reported results. Effectiveness of proposed components are also studied as well.
3: The paper is well organized and the presentation of the paper is great.
- 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.
Overall of presentation of the proposed method is great. I have a minor question about the paper.
The authors provide three edge-oriented processes to incorporate boundary information of the target. Although the overall results and some ablation studies are provided in the experiments. The details of effectiveness of each component is not given in the paper. It would be better to give more analysis about the presented components.
- 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 some details of the experimental protocols are missing in the paper, I would say that it is not that difficult to reproduce the experiments.
- 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
The overall presentation of the paper is great. I have some suggestions.
1: It would be great to provide the experimental results of the effectiveness of each proposed component.
- Some experimental details are missing in the paper. For example, the details parameters of construction of the edge graph.
- 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 authors present a novel point-based method for 3D intracranial aneurysm segmentation. The method aims to pay more attention to the target boundary. The edge oriented segmentation method can be applied to other segmentation tasks in the 3D medical image domain.
- Number of papers in your stack
4
- What is the ranking of this paper in your review stack?
1
- Reviewer confidence
Confident but not absolutely certain
- [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed
N/A
- [Post rebuttal] Please justify your decision
N/A
Review #2
- Please describe the contribution of the paper
This paper propose a new framework to segment intracranial aneurysm from point clouds containing both aneurysm and blood vessels, and emphasis is placed on how to segment the aneurysm edge accurately. To this end, the proposed framework consists of three parts, a dual transformer to segment the aneurysm and the edge separately, and a graph convolution part and a contrastive learning part to further enhance edge segmentation. Experiments are done on a public intracranial vessel and aneurysm dataset consisting of 116 annotated aneurysm. Better performance than baselines are achieved on aneurysm 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.
The proposed aneurysm segmentation framework is new and the motivation of each part is reasonable. Better performance on aneurysm segmentation is achieved than baselines. Evaluation is performed on public dataset. The paper is well organized and easy to follow.
- 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 clinical significance of this work is doubtful. If aneurysm segmentation is critical for clinical decision making, it is better to do the segmentation carefully by the doctors and it is not difficult to segment an aneurysm manually.
- The proposed model is rather complex while the dataset used for evaluation is rather small, in which there are only 116 aneurysm.
- The proposed method is based on point cloud, while generating the point clouds from original images may introduce some errors, which further increase the uncertainty of this method in clinical practice.
- Ablation study is not complete, since the results of applying any two of the three components are desirable.
- 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
Reproducibility is OK though parameter tuning is not provided. “An analysis of statistical significance of reported differences in performance between methods.” is not provided though the answer to this question is YES.
- 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
- Consider improving on the main weakness.
- The last sentence of 3rd paragraph of Introduction says “3DMedPT can still not perform well around the edge between vessels and aneurysms due to the less supervision and ambiguous features, where is extremely harmful for the clipping surgery process.” However, visualization of 3DMedPT results are not provided and its performance in segmenting edges are not clear.
- The first sentence of section 2.2 “Due to the ambiguous features generated from similar contexts, points around the edge are easily misclassified, which is harmful for the surgery process.” It is not enough to only say “is harmful for the surgery process” and it should be explained how it harms the surgery process.
- 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
4
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Clinical relevance is doubtful and experimental dataset is small.
- Number of papers in your stack
6
- What is the ranking of this paper in your review stack?
2
- Reviewer confidence
Very confident
- [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed
N/A
- [Post rebuttal] Please justify your decision
N/A
Review #3
- Please describe the contribution of the paper
Point cloud processing and analysis have been a popular topic in the community of 3D computer vision, however, medical point cloud studies are still in demand towards clinical applications of practical significance, such as clipping surgery and Intra identification. This is an interesting topic, which has not been well-explored yet. It is thus encouraging to see studies proposed to tackle geometric processing problems for medical usage.
- 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 proposed a three-stage paradigm to achieve 3D Intra Segmentation with clearer edge boundary.
-
The proposed method outperforms current SOTA by a considerable margin.
-
The paper is well-written with meaningful figures.
-
- 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.
-
Qualitative comparisons with SOTAs on the segmentation near edge boundaries are necessary but missing in the paper. The paper claims that they proposed a three-stage paradigm to focus on distinguishing the edge boundary segmentation (which thus improves the overall segmentation performance), however, this claim is unjustified in the manuscript. It would be insightful if we could see these improvements near edge boundaries qualitatively.
-
The ablation discussion looks lack of insight. However, considering the page limitation, I may find it personally acceptable if the aforementioned additional qualitative comparisons can be added to reveal the insights.
-
The authors achieved better results, however, it’s probably due to their usage of edge labels, which are unavailable for other SOTAs. Given a standard setting (where edge labels are unavailable to model training), it’s unsure whether the proposed method can still outperform others by a considerable margin. It’s likely their overall proposed pipeline might not even work when these edge labels are unavailable, as the proposed modules look rely heavily on the edge annotations.
-
- Please rate the clarity and organization of this paper
Excellent
- 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 paper is well-written, with clear specification. However, the authors did not promise to release code, thus I might have some concerns in its reproducibility.
- 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
Failure to address my concerns listed above might reduce my impression and marks. I am giving this result, considering the limited works in this domain and based on the expectation if these concerns could be satisfactorily addressed. Open source is also preferred by the community.
- 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
5
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
See my comments above.
- Number of papers in your stack
5
- What is the ranking of this paper in your review stack?
2
- Reviewer confidence
Confident but not absolutely certain
- [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed
N/A
- [Post rebuttal] Please justify your decision
N/A
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.
Overall, the reviewers are positive and recognize the novelty of the methods, but there are some concerns on the clinical significance of the method, the roles of edge, etc.
- 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).
3
Author Feedback
We thank the meta-reviewer and reviewers for their valuable comments and positive feedback regarding the idea. Reviewer #1: Q1: The details of the effectiveness of each component are not given in the paper. Due to the limited space, we only provide experiments removing each component. To provide more evidence of the effectiveness of each component, we further conduct experiments by adding each component to the baseline under the 512 sampling scheme. The results show that the mIoU of the baseline is 88.61%, while only adding DST can achieve 90.46%, only adding ODC can achieve 90.83%, and only adding IHE can achieve 90.35%. The performance increase can further prove the effectiveness of each component. Reviewer #2: Q1: The clinical significance of this work is doubtful. If aneurysm segmentation is critical for clinical decision making, it is better to do the segmentation carefully by the doctors and it is not difficult to segment an aneurysm manually. Answer: We agree that manually segmenting aneurysms is feasible, but it may be time-consuming since the doctors need to find the probable positions of aneurysms and then carefully draw the boundary lines to segment the aneurysms. While our work can automatically segment the aneurysms, even if the result is not precise enough for direct practical use, it can act as a rough indicator for the doctors, which can greatly reduce the laborious effort. Reviewer #3: Q1: Qualitative comparisons with SOTAs on the segmentation near edge boundaries are necessary but missing in the paper. Answer: To evaluate the segmentation performance around boundaries, we define a simple edge IoU metric (EIoU). EIoU computes the mIoU inside the edge region, which is defined as the point set that contains 16-nearest points of the edge points. The baseline model achieves 43.0% EIoU while EPT-Net can achieve 53.5% EIoU. The result reveals that the design of this work can effectively improve the segmentation performance around the edge. Q2: It’s likely their overall proposed pipeline might not even work when these edge labels are unavailable, as the proposed modules look rely heavily on the edge annotations. Answer: In fact, edge labels can be easily extracted from ground truth segmentation labels under the manual definition. For example, we can define a point as an edge point if its 8-nearest neighbors have multiple categories. Using this kind of computed edge label, we can achieve 91.37% mIoU, which is slightly worse than the 91.52% mIoU using ground truth edge labels but still outperform previous 3D segmentation methods.