List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Ruifei Zhang, Peiwen Lai, Xiang Wan, De-Jun Fan, Feng Gao, Xiao-Jian Wu, Guanbin Li
Abstract
Automatic and accurate polyp segmentation plays an essential role in early colorectal cancer diagnosis. However, it has always been a challenging task due to 1) the diverse shape, size, brightness and other appearance characteristics of polyps, 2) the tiny contrast between concealed polyps and their surrounding regions. To address these problems, we propose a lesion-aware dynamic network (LDNet) for polyp segmentation, which is a traditional u-shape encoder-decoder structure incorporated with a dynamic kernel generation and updating scheme. Specifically, the designed segmentation head is conditioned on the global context features of the input image and iteratively updated by the extracted lesion features according to polyp segmentation predictions. This simple but effective scheme endows our model with powerful segmentation performance and generalization capability. Besides, we utilize the extracted lesion representation to enhance the feature contrast between the polyp and background regions by a tailored lesion-aware cross-attention module (LCA), and design an efficient self-attention module (ESA) to capture long-range context relations, further improving the segmentation accuracy. Extensive experiments on four public polyp benchmarks and our collected large-scale polyp dataset demonstrate the superior performance of our method compared with other state-of-the-art approaches. The source code is available at https://github.com/ReaFly/LDNet.
Link to paper
DOI: https://link.springer.com/chapter/10.1007/978-3-031-16437-8_10
SharedIt: https://rdcu.be/cVRsW
Link to the code repository
https://github.com/ReaFly/LDNet
Link to the dataset(s)
https://datasets.simula.no/kvasir-seg/
https://polyp.grand-challenge.org/CVCClinicDB/
Reviews
Review #1
- Please describe the contribution of the paper
The paper proposes a method for endoscopic polyp segmentation that uses an adaptive/dynamic kernel for optimizing features relevant to a polyp; the method also utilized two new attention modules ESA and LCA. Rigorous evaluation on public data provides experimental support.
- 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 main contribution is the technical addition of dynamic kernels and ESA LCA to the endoscopic polyp segmentation problem. While inspired by work in general vision/learning it is novel in this setting.
-
Thorough experiments and comparative analysis are reported using public endoscopic datasets. Experiments on a new dataset are also reported with an indication (in form, not in the paper) that it will be released but no link.
-
- 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 authors should try to position the clinical value of the work a little better. At times, it is mentioned that the new method would improve concealed lesion detection/segmentation, however this is neither shown in experiments nor made clear as to logically how.
-
Results are reported on a new internal dataset. These, however, use a slightly different splits to the other experiments and also it is entirely unknown if the data will be seen by anyone else. Hence somewhat limited value.
-
- 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 code will be made available. The paper is clear enough to reproduce the method.
The authors talk about a new dataset, and in the form indicate it will be released, however, details on this are very scant. Insufficient information on the data, labelling procedure, etc. etc. No link.
- 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
Overall, the paper is very well presented and easy to follow. Some points for possible improvement below.
It would be helpful to explain the clinical use case better. For example, is this an approach to better segment the same polyp once detected in subsequent frames by using a the dynamic kernel? Or do you think the kernel will adapt to a particular colonoscopy? Also what is the segmentation need for polyps in terms of clinical utility?
When making claims, it would be good to have experimental basis/support. For example, in the LCA description the section ends with “which significantly improves the feature contrast and benefits to detect conceal polyps.” But where is this demonstrated?
The new dataset should be described fully including size, data labelling procedure, etc.
The experiments on public data are commendable and rigorous. What limitations to testing the proposed method do these datasets present? E.g. do you need videos to demonstrate the full utility of the method?
Perhaps if the method is meant to apply/adapt better to temporal information (video) then it would be relevant to also ref works which incorporate the temporal features. Several of these in MICCAI, e.g. Puyal, et al. 2020 and Wu, et. al 2021.
- 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?
This is a technically good and novel paper. It does need some clarification and also additional clinical context for improvement with potential ref to use in video.
- 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 #2
- Please describe the contribution of the paper
This manuscript proposes the lesion-aware dynamic kernel (LDNet) for polyp segmentation, which is generated conditioned on the global information and updated by the multi-level lesion features. The dynamic kernel endows LDNet with more flexibility to attend to diverse polyps’ regions. Besides, LDNet use two tailored attention modules (ESA and LCA) to improve the feature representation and enhance the context contrast. Extensive experiments and ablation studies demonstrate the effectiveness of LDNet.
- 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 paper is well-written and -organized. The novelty is interesting in this field. The performance is competitive when compared with other approaches.
- 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.
- (optional) the authors could add more efficiency comparison in the benchmark table.
- There are several recent works [1,2] may helpful to the section of related works.
[1] Progressively Normalized Self-Attention Network for Video Polyp Segmentation [2] Video Polyp Segmentation: A Deep Learning Perspective
- 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 authors say that “The code and collected dataset will be made available.”. Also, the reviewer think it is not difficult to reproduce LDNet according to the details provided by authors.
- 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
Please see Section 3 &4
- 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?
This manuscript is ready for the acceptance of MICCAI. It has detailed methodology, enough experiments and competitive performance. I vote for ‘strong accept’ in first round.
- Number of papers in your stack
5
- 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
N/A
- [Post rebuttal] Please justify your decision
N/A
Review #3
- Please describe the contribution of the paper
This paper describes a method for polyp segmentation that combines a U-net segmentation architecture with dynamic kernel update as well as self-attention and cross-attention modules. These components are brought together to produce accurate lesion segmentations validated across various datasets.
- 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 bring together various elements including a U-net segmentation architecture, dynamic kernel generation and update, and self-attention and cross-attention modules.
Evaluations show high accuracy in lesion segmentation across various datasets.
- 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.
Unclear what is meant by ‘conceal’ polyps. A quick literature search did not reveal a definition. Authors should define any terms they are introducing in the paper.
While the paper shows improvement, it is hard to assess the significance of the improvements. Authors mention that their ablation studies revealed significant improvement upon adding dynamic kernels, for instance, however how the significance is calculated is not explained. How significant is an improvement of ~1% over baseline results in the 90% range? Is this improvement a result of the additional parameters introduced by this method and how many more parameters are needed to generate a ~1% improvement?
- Please rate the clarity and organization of this paper
Poor
- 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 could be improved (see detailed comments below)
- 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
One aspect the authors should try to highlight is how their methods differ from prior work that introduced the various components brought together in this work. While putting various components together is commendable, it would be nice to see where authors have introduced changes to bring these components together. If authors can highlight these main modifications (e.g., after contributions in the introduction section), it could help bring out the impact of this paper.
While authors explain various components of their method is detail, some further details critical for reproducibility are required. Some aspects that were unclear are below:
-
Were all the other architectures that authors compare against also trained in the same way? I.e., using the same datasets and splits?
-
Why is K=1? Were other Ks tried? Was this K chosen via experiments with the validation dataset?
-
Can authors elaborate on how the binary cross entropy and dice loss were combined? I.e, are these weighed equally throughout training?
-
How are recall, specificity, precision and accuracy defined? As in, how do authors decide that a particular lesion is correctly identified? Is this done via some threshold on the Dice score or IoU? If so, please describe. It seems that the Dice scores for held out datasets are quite a bit lower than datasets used for training while accuracy remains fairly unchanged. So understanding how accuracy, etc. are computed might help understand this disparity.
-
Finally, while all the metrics shown in evaluation are useful to see, it may also be nice to include how many parameters each of the architectures are using in order to solve the segmentation problem.
-
- 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?
There are several details missing and the manuscripts needs to be polished further.
- Number of papers in your stack
5
- What is the ranking of this paper in your review stack?
5
- 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.
The reviewers all commend the relevance of the proposed manuscript and the performance demonstrated by the proposed method which makes it wholly suitable for publication at MICCAI. It is however emphasised that there are points of improvement possible regarding the clarity of the manuscript. An increased level of details would be needed for instance for the experimental setting. Further, discussion on the clinical relevance of the observed improvement along with details on possible practical implementation would be important to add to the final version.
- 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).
4
Author Feedback
N/A