List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Heng Li, Haofeng Liu, Huazhu Fu, Hai Shu, Yitian Zhao, Xiaoling Luo, Yan Hu, Jiang Liu
Abstract
Fundus photography is a routine examination in clinics to diagnose and monitor ocular diseases. However, for cataract patients, the fundus image always suffers quality degradation caused by the clouding lens. The degradation prevents reliable diagnosis by ophthalmologists or computer-aided systems. To improve the certainty in clinical diagnosis, restoration algorithms have been proposed to enhance the quality of fundus images. Unfortunately, challenges remain in the deployment of these algorithms, such as collecting sufficient training data and preserving retinal structures. In this paper, to circumvent the strict deployment requirement, a structure-consistent restoration network (SCR-Net) for cataract fundus images is developed from synthesized data that shares an identical structure. A synthesized cataract set (SCS) is first simulated to collect cataract fundus images sharing identical structures. Then high-frequency components (HFCs) are extracted from the SCS to constrain structure consistency such that the structure preservation in SCR-Net is enforced. The experiments demonstrate the effectiveness of SCR-Net in the comparison with state-of-the-art methods and the follow-up clinical applications.
Link to paper
DOI: https://link.springer.com/chapter/10.1007/978-3-031-16434-7_47
SharedIt: https://rdcu.be/cVRsf
Link to the code repository
https://github.com/liamheng/ArcNet-Medical-Image-Enhancement
Link to the dataset(s)
N/A
Reviews
Review #1
- Please describe the contribution of the paper
In this paper, the authors proposed a restoration network (SCR-Net) for cataract fundus images enhancement. They designed a synthesis model of cataract images to generate synthesized training data (SCS), a restoration model to get cataract fundus images from synthesized data, and a module for HFC alignment. These three parts were integrated into the backbone of SCR-Net. Their approach was evaluated on two public datasets and two private datasets respectively and test results showed 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.
(1) In order to overcome the difficulty of collecting paired cataract images, the authors designed a cataract simulation model to generate synthesized cataract sets as training data (2) To enforce the structure preservation in the restoration, they designed a model to extract the HFCs from the SCS to boost the model training.
- 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) Compared with other studies, the model proposed by the author has a certain improvement in evaluation metrics. But the samples of public data sets used for evaluation and comparison are few, public data sets are also few, so the model’s validity is limited. (2) IoU was not validated in the Kaggle dataset. (3) The authors said intuitively SCS generated by cataract simulation models with Eq. 2 share the same identical structures. They should use metrics for quantitative and qualitative evaluation. (4) In abstract part, the authors said the SCS were generated from cataract fundus images. However, in other parts, such as in Fig.1, SCS were generated from a clear image. They should be unified. (5) Were the clear images from post-operation eyes or normal eyes？If the clear images are from post-operation eyes, would they also suffer from the short of data. (6) The authors should not augment their conclusions. E.g., they demonstrate that their approach was effectiveness in the follow-up clinical applications and the existing algorithms ignored the performance improvement of clinical applications from the enhancement. But there was no corresponding evidence. (7) In conclusion part， the authors said “Thanks to its independence from annotations and test data, the proposed algorithm is convenient to deploy in clinics”. Please explain “independence” and the convenience on clinical 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
Upon release of the code, the reproducibility can be verified.
- 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 text lacks clarity and needs important English editing. The authors also need to be more precise, to do not overinterpret their results and to dampen their conclusions. E.g., “intuitively” lacks clarity and not scientific. (2) Other comments mentioned clearly in the strength and the weakness section of the papers.
- 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 study did provide several interesting results, the model proposed by the author has a certain improvement in evaluation metrics. The study overcomes the difficulty of collecting paired cataract images and enforces the structure preservation. But the samples of public data sets are few, so as to the public data sets. Besides IoU was not validated in the Kaggle dataset. the text also lacks clarity and not scientific, needs important English editing. Although the relevant weakness and comments are mentioned before, we still think the paper can be accepted weakly given that the comments given in the weakness are addressed properly.
- 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
Review #2
- Please describe the contribution of the paper
This paper aims at enhancing the quality of fundus images to improve the certainty of fundus examination, proposes a structure-consistent restoration network (SCR-Net) to enhance cataract fundus images in the absence of supervised data. The authors generate synthesized cataract set (SAS) by synthesizing cataract fundus images sharing identical structures according to the imaging principle. To boost the training stage and structure preservation in SCR-Net, the high-frequency components (HFCs) are extracted from network to constrain structure consistency. Following comparison experiments and ablation studies prove that the proposed algorithm has achieved state-of-the-art.
- 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 article focuses on capturing cataract-invariant features of retinal structures in image restoration, generates a synthesized cataract set by a simulation model with several parameters. Moreover, a low-pass Gaussian filter is adopted to extract the low -frequency components and maintain structure consistency. The authors expound the design of the algorithm clearly with detailed formulas and experiments. The comparison of several algorithms is also intuitively showed and demonstrated with restoration image 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.
- The value settings of specific parameters in most formulas are not clear, the authors need to show the actual parameter values in paper and explain the reasons for choosing these values. Moreover, some revelant experimental data in parameter settings need to be introduced, which means it would be better if following ablation studies of parameters are shown in the experiment section.
- In introduction section, the article summarizes the challenges and shortcomings of cataract restoration algorithms, however, “To address these problems” in page 2 is not suitable. The point four, “ignoring the performance improvement of clinical applications from the enhancement”, seems not clearly illustrated how it was solved. Besides, several experiments to evaluate clinical applications need to be done.
- Though the authors have done a solid work, the novelty of the work is limited. The designment of the whole restoration process is too complicated, it’s hard to say whether the final results are benefit from the structure consistency idea or just many additional detailed local designments.
- 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
Two private datasets are adopted in the training and evaluation stage. Besides, several value settings of parameters are lost in the article. Adding the instructions of these parts in the paper would contribute to the improvement of reproducibilily.
- 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 parameters showed in the figures should be explained. In Fig 1, three kinds of loss fuctions, LR、LH、Lcyc, their meanings of symbol are not found in the instruction.
- A detailed description of quantitative metrics of restoration and segmentation in Table 2 is required. It’s better to clarify the reasons and advantages of choosing those metrics.
- 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 novelty of the paper is limited, but the contribution is solid.
- 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
N/A
- [Post rebuttal] Please justify your decision
N/A
Review #3
- Please describe the contribution of the paper
The authors proposed a SCR-Net, which utlizes the high frequency structure consistency of fundus image, to enhance degraded fundus image. Moreover, a synthesis model for generating cataract images is proposed following the principle of fundus image.
- 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 way of generating SCS is well explained and easy to follow. Additionaly, the SCR-Net is technically sound.The evaluation is not limited to the restoration but to various clinical applications such as segmentation and classification.
- 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 discussion of three loss function in Eq.7 is not sufficent, conducting ablation studies on them would make the proposed method more convincing. 2.It would be better to present more qualitative results of cataract restoration results on appendix, considering Fig.3 only contains one test sample.
- 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
It is likely to be reproducible.
- 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 author may consider discuss more about the loss function, as mentioned above.
- 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?
This paper is technical sound, easy to understand, and satisfying in its novelty. Specifically, the authors analyzed the characteristics of fundus imaging and designed a cataract simulation model. In addition, the authors designed a feature alignment module for learning the correspondence between high frequency features and origin features, which is reasonable. However, the experiments are not sufficient, for example, some ablation studies are missing. Considering the above issues, I recommend to accept this paper.
- Number of papers in your stack
4
- 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.
this paper proposed a restoration model for cataract fundus image enhancement by enforcing the structure preservation, the model is trained with the synthesiezed data to overcome the difficulty of collecting paired catatact image. Given three consistent positive reviews, I recommend accepting this submission. However, there are many concerns arised in the reviews, such as limited experiemts, English presentation, settings of hyper-parameters, the authors should address the detailed comments from the reviewers in the camera-ready manuscript.
- 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).
1
Author Feedback
Thanks very much for taking the time to review this manuscript. We appreciate all your comments and suggestions! Please find our itemized responses below. Q1: The evidence to demonstrate the proposed approach is effective in the follow-up clinical applications. (Reviewer #1-5-(6), Reviewer #2-5-(2)) A: We are grateful for the reviewer’s concern. The major follow-up clinical applications for fundus image enhancement include vessel segmentation and disease diagnosis, which have been presented in our experiments to demonstrate the advantage of our algorithm.
Q2: Verification of the effectiveness of the network designments and losses. (Reviewer #2-5-(3), Reviewer #3-5-(1)) A: We appreciate the reviewer’s recommendation. An ablation study has been provided in Table 2 to demonstrate the effectiveness of the design. As a result of the length limitation, only the major modules are verified in the submitted manuscript version. More detailed verification will be provided in our following study.
Q3: The value settings of specific parameters need to show the actual parameter values in the paper and provide relevant ablation studies. (Reviewer #2-5-(1)) A: We thank the reviewer for this reminder. The parameters in Eq.2, 3, and the number of K have been added to the manuscript. In addition, the code of our study will be publicly available. Due to the length limitation, ablation studies of parameters will be provided in our following study.
Q4: The samples of public data sets used for evaluation and comparison are few, and public data sets are also few. (Reviewer #1-5-(1)) A: There is gratitude for this comment. Actually, limited data volume is an important challenge in this task, and the motivation of our study.
Q5: IoU was not validated in the Kaggle dataset. (Reviewer #1-5-(2)) A: Many thanks for the comment. As high-quality reference is not available in the Kaggle dataset, it is unable to obtain the metric of IoU.
Q6: Quantitatively and qualitatively evaluate the SCS sharing the same identical structures. (Reviewer #1-5-(3)) A: We appreciate the reviewer’s recommendation. As described in the section, the SCS is obtained from identical clear images with varying cataract blur. Thus identical structures are carried by the SCS.
Q7: Different descriptions of the SCS in the abstract and other parts. (Reviewer #1-5-(4)) A: We would like to thank the reviewer for the suggestion. The description in the abstract has been revised.
Q8: Were the clear images from post-operation eyes or normal eyes？(Reviewer #1-5-(5)) A: There is gratitude for this comment. Normal fundus images are used as the clear ones in the training phase, while the post-operation ones are only used for tests.
Q9: Please explain “independence” and the convenience of clinical applications. (Reviewer #1-5-(7)) A: Many thanks for the comment. Because the proposed algorithm requires no segmentation annotation or paired high-quality reference, it is independent of annotations and is convenient to implement.
Q10: It would be better to present more qualitative results of cataract restoration results in the appendix. (Reviewer #3-5-(2)) A: We appreciate the reviewer’s recommendation. More results have been exhibited in the appendix.