List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Jun Wu, Bo Shen, Hanwen Zhang, Jianing Wang, Qi Pan, Jianfeng Huang, Lixin Guo, Jianchun Zhao, Gang Yang, Xirong Li, Dayong Ding
Abstract
Corneal nerve fiber medical indicators are promising metrics for diagnosis of diabetic peripheral neuropathy. However, automatic nerve segmentation still faces the issues of insufficient data and expensive annotations. We propose a semi-supervised learning framework for CCM image segmentation. It includes self-supervised pre-training, supervised fine-tuning and self-training. The contrastive learning for pre-training pays more attention to global features and ignores local semantics, which is not friendly to the downstream segmentation task. Consequently, we adopt pre-training using masked image modeling as a proxy task on unlabeled images. After supervised fine-tuning, self-training is employed to make full use of unlabeled data. Experimental results show that our proposed method is effective and better than the supervised learning using nerve annotations with three-pixel-width dilation.
Link to paper
DOI: https://link.springer.com/chapter/10.1007/978-3-031-16440-8_5
SharedIt: https://rdcu.be/cVRvp
Link to the code repository
N/A
Link to the dataset(s)
N/A
Reviews
Review #1
- Please describe the contribution of the paper
The manuscript presents a semi-supervised learning framework for corneal nerve fibers segmentation in confocal images. The method is well described and characterized by many steps. Some of these steps are innovative and the overall method is a novelty. Results are well evaluated and show an improvement with respect to other methods for corneal nerve 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 paper is well written, methods and results are clear. Some parts of the proposed method are a novelty, and consequently the overall method is a novelty applied to corneal nerves images. The proposed method could be applied to other types of images.
- 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.
Many papers in literature describe methods for corneal nerve segmentation from confocal images. The proposed method achieves a limited improvement in the segmentation performance, and the clinical relevance of this small improvement is not demonstrated.
- 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 method is made by many step. The overall method is clear but the reproducibility of each single step is not, probably due to the limited number of available pages.
- 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 proposed method is interesting and achieves an improvement in corneal nerve segmentation. However, this improvement is limited with respect to other methods for corneal nerve segmentation. I suggest to the authors to investigate about the clinical relevance of this small improvement and about the generalisation of the proposed method to other types of images.
- 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 paper is well written, methods and results are clear. The proposed methods is interesting. Some parts of the proposed method are a novelty, and consequently the overall method is a novelty applied to corneal nerves images. The improvement in the segmentation of corneal nerves is limited.
- Number of papers in your stack
2
- 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
The authors proposed a semi-supervised segmentation framework that could work with only a few labeled data samples. They used masked image modeling (MIM) to pretrain a U-Net with unlabeled data and then fine-tune the model with labeled data and combined unlabeled data with pseudo-labels and actual labeled data to retrain the model. The performance is good and convincing.
- 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 is a novel application for segmentation to Corneal Confocal Microscope images. The framework they proposed is solid and strong to deal with the small labelled data size problem and they tested the model both in public and private datasets to show it effectiveness. To overcome the loss of pixel details, they used Refine Reparing Network to help refine the learning results of Coarse Repring Network which is excellent.
- 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 whole framework is solid but widely used in self-training, unsupervised training computer vision papers, so the novelty is not strong for a methodology paper for example, Kaiming He etc, Masked Autoencoders Are Scalable Vision Learners, used similar approach to do self-learning problems.
- 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
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
The framework is a very complex framework e.g., three stages, pertaining with two networks type GAN, fine-tuning, retraining etc., but the Ablation Study is very short, and the table is not convincing enough. For example, what is the influence of plain model with using pseudo-labels, dropping any training stages, what about the other combinations. Based on my understanding, in this complex framework both model design and dataset mixing shall play a role and Ablation Study shall distinguish them clearly.
- 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?
The paper is good at implemenation but the novelty of the paper is not strong for MICCAI.
- Number of papers in your stack
5
- 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
The submission propose a semi-supvervised learninig framework for nerve segmentation in CCM, which contains three main components: pretraining on large unlabeled datasets and refining on small labelled datasets, generating pseudo labels, and retraining with the full datasets. The stage of pretraining utilize a masked image modeling based GAN structure, including coarse repring network, refine reparing network, and disciminator, among which the encoder of refine reparing network is used in the whole semi-supervised framework.
- 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 organization of the whole manuscript is quite good with clear illustrations of the implementing pipeline. (2) The whole framework is reasonable to deal with the annotation problem in microscope images.
- 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) Some statements are too strong to be demonstrated. (2) Computational efficiency is not analyzed although it’s mentioned in the last sentence of Section 1 that “our method is data efficient and more friendly to computing resource.”
- 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
No code sharing was mentioned in the submission. The datasets are publicly avaible.
- 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) It’s unclear why gated convolution is used in coarse repairing network in the perspective of its difference to refine repairing network. (2) It may be improper to use the term “supervised “ in Table 2, as the ground truth are simulated from existing centerline, which are not real careful annotations and may be even worse than pseudo labels. Therefore, this manner is not able to demonstrate the proposed method is better than supervised method. Please use a weaker statement or provide stronger evidence. (3) Please provide analysis of computational considerations such as data efficiency and time efficiency. (4) Please provide description about what knid of preprocessing steps are needed to deal with noise and difference in background brightness. (5) Is there any reason to propose this framework for nerve segmentation? Is there any special design for this narrow object? Is it also suitable to solid objects?
- 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?
Good work but some important descriptions need to clarify.
- Number of papers in your stack
4
- 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.
All reviewers agreed on acceptance, and an early accept is recommended. However, the AC has couple concerns. 1)The authors stated in the end of the Introduction:… How to solve the disconnection and improve the pixel-wise accuracy… But I’m not convinced by the experimental results. Based on my observations, all the results are centerline-based, and the sample images for illustration are not challenging. The authors better to illustrate couple representative sample images, such as images with intensity inhomogeneity, so as to reveal the superiority of the given method in dealing with the disconnection issue. 2) Last sentence of Results: Our method can also distinguish large and small nerve fibers, especially for some tiny unclear nerves. Based on the results reported on the given table nor Fig.3, it is hard to agree with your argument. If possible, setting a experiment to proof your network is able to deal with the tiny nerves and better than other SOTA methods. 3) Please cite the correct reference of [39]: Automated Tortuosity Analysis of Nerve Fibers in Corneal Confocal Microscopy, IEEE Transactions on Medical Imaging, 2020, 39(9): 2725-2737.
Hope raised concerns from R#3 and AC will help the authors to improve and direct their further research.
- 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
N/A