List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Jia Guo, Xinglong Liu, Yinan Chen, Shaoting Zhang, Guangyu Tao, Hong Yu, Huiyuan Zhu, Wenhui Lei, Huiqi Li, Na Wang
Abstract
Pulmonary embolism (PE) is life-threatening and computed tomography pulmonary angiography (CTPA) is the best diagnostic techniques in clinics. However, PEs usually appear as dark spots among the bright regions of blood arteries in CTPA images, which can be very similar with veins that are less bright and soft tissues. Even for experienced radiologists, the evaluation of PEs in CTPA is a time-consuming and nontrivial task. In this paper, we propose an artery-aware 3D fully convolutional network (AANet) that en-codes artery information as the prior knowledge to detect arteries and PEs at the same time. In our approach, the artery context fusion block (ACF) is proposed to combine the multi-scale feature maps and generate both local and global contexts of vessels as soft attentions to precisely recognize PEs from soft tissues or veins. We evaluate our methods on the CAD-PE dataset with the artery and vein vessel labels. The experimental results with the sen-sitivity of 78.1%, 84.2%, and 85.1% at one, two, and four false positives per scan have been achieved, which shows that our method achieves state-of-the-art performance and demonstrate promising assistance for diagnosis in clinical practice.
Link to paper
DOI: https://link.springer.com/chapter/10.1007/978-3-031-16431-6_45
SharedIt: https://rdcu.be/cVD61
Link to the code repository
https://github.com/guojiajeremy/AANet
Link to the dataset(s)
https://ieee-dataport.org/open-access/cad-pe
https://www.kaggle.com/datasets/andrewmvd/pulmonary-embolism-in-ct-images
Reviews
Review #1
- Please describe the contribution of the paper
The authors propose a novel 3D fully convolutional network for pulmonary embolism (PE) detection with artery-aware. An artery context fusion block is embedded in the network to generate artery context to guide PE detection. Even-dice loss is introduced to avoid gradient exploding. The method was evaluated with a public dataset (CAD-PE), and achieved state-of-the-art performance.
- 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.
A novel artery-aware framework for pulmonary embolism detection is proposed. A loss function even-dice loss is introduced and balances the artery / PE loss. The experiment with public data and various settings is persuasive.
- 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) The network was pretrained on a large-scale dataset LUNA16. As the LUNA16 dataset is for lung nodule detection, the pre-training processing is not straightforward to understand. 2) A challenge dataset CAD-PE was used. The authors made extra annotations (pulmoanry artery) on the public data, which was used as auxilary class for PE detection. The improvement is noticiable, but it might be unfair to compare with others participants of the challenge 3) Based on the network prediction, post-process (morghology closing) is used after threshold. How much does post-processing improve the performance.
- 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
The reproducibility of the paper is OK.
- 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
None
- 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
6
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The novel methodology and experiments
- 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
The authers proposed an artery-aware 3D fully convolutional network (AANet) that encodes artery information as the prior knowledge to detect arteries and PEs. They developed An artery context fusion block (ACF) generating the context of artery as in-network prior knowledge to guide PE prediction. The methods are evaluated on the CAD-PE dataset with the artery and vein vessel labels which shows good performance.
- 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 experiment is of great significance for the PE diseases, and the technical method is reliable. In this paper, an artery-aware network (AANet) to segment PE in CTPA image that fully utilizes artery context is proposed. The method has certain novelty in application.
- 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.
- At the end of the first paragraph of the introduction section, the author only stated that the high false positive PE diagnosis was partly caused by doctors, and did not discuss other reasons in detail, such as the impact of the gray, size and other characteristics on the CTPA images.
- At the end of the third paragraph of the introduction section, the data set mentioned by the authors should be cited.
- In the AANet of the Methods section, the authors did not explain the role of using different numbers (two or three) residual blocks, the difference between stdConv and Conv, and the role of fusion multiscale modules for PE segmentation.
- There are too few method measures to confirm the accuracy of the method.
- In the results section, the PE 3D or 2D segmentation results should be compared with the ground truth.This method should be compared with other methods.
- The discussion and conclusions are insufficient, and more should be discussed about its advantages over other methods and its possible future applications.
- Please rate the clarity and organization of this paper
Satisfactory
- 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 article is possible.The network structure parameters and loss function are described in the article.
- 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
- At the end of the first paragraph of the introduction section, the author only stated that the high false positive PE diagnosis was partly caused by doctors, and did not discuss other reasons in detail, such as the impact of the gray, size and other characteristics on the CTPA images.
- At the end of the third paragraph of the introduction section, the data set mentioned by the authors should be cited.
- In the AANet of the Methods section, the authors did not explain the role of using different numbers (two or three) residual blocks, the difference between stdConv and Conv, and the role of fusion multiscale modules for PE segmentation.
- There are too few method measures to confirm the accuracy of the method.
- In the results section, the PE 3D or 2D segmentation results should be compared with the ground truth.This method should be compared with other methods.
- The discussion and conclusions are insufficient, and more should be discussed about its advantages over other methods and its possible future applications.
- 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?
This paper describes the network structure and the experimental process completely, but the experimental results and discussion part are insufficient.The authors added artey-vein prior knowledge as a novel weight for PE segmentation, but did not compare with other methods.
- 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
N/A
- [Post rebuttal] Please justify your decision
N/A
Review #3
- Please describe the contribution of the paper
Clinically, this paper aims at pulmonary embolism detection. Technically, this paper puts forward a way to introduce attention to arteries in the segmentation networks.
- 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.
Authors put forward a way to introduce attention in the segmentation networks, together with a loss function (Even-Dice Loss) which could work even if there is no foreground in the sampled batch.
- 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 method itself requires more annotation than traditional methods, and the annotation (as shown in Fig 2) requires lots of work. Although authors mention that this network and weights could be used as a pretrained model for other tasks, it is still questionable whether this method could be applied to other similar diseases.
The study/experiment design could be modified to make it clear to readers. To make a meaningful comparison with all methods, it would be better if authors could mention whether all methods were pretrained with LUNA16. And although mentioned in the paper that citation [13] missed some small cases, I would still want to see how AANnet performs on the 80+ cases subsets (as mentioned in [13]).
- 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
Authors claim that the dataset and the network/weights will be made public. I believe that the reproducibility is good.
- 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
in Table 1, I believe “SDiceLoss” should be “PSDiceLoss” based on this paper’s context.
Mention if all methods were pretrained with LUNA16.
Show AANnet’s performance on the 80+ cases subsets (as mentioned in [13]).
- 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
6
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The performance of the method is good, which outperforms existing methods, and the loss function mentioned could be generalized to other tasks. However, the study design could be improved (at least with better description in the paper).
- Number of papers in your stack
3
- 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
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.
The authors propose a 3D CNN for pulmonary embolism (PE) detection. The novelty lies in their attention-based context fusion block, that guides the PE-detection. Additionally, they introduce an Even-Dice Loss, carry out an ablation study and show interesting visuals of their segmentations.
The main criticism is that additional annotations are necessary for proposed method to work and, thus, comparison to scores reported by the benchmark of the data used may be unfair. While it this is true, it is noteworthy that proposed method outperforms all SOTA-methods. Lastly, it would have been nice to see the effect of post-processing on the influence of performance, but in the light of novelty of the paper, I see this as an accept.
- 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 all the reviewers for their thorough reviews. According changes will be made in the camera-ready submission.