Back to top List of papers List of papers - by topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Sharif Amit Kamran, Khondker Fariha Hossain, Alireza Tavakkoli, Stewart Lee Zuckerbrod, Kenton M. Sanders, Salah A. Baker
Abstract
High fidelity segmentation of both macro and microvascular structure of the retina plays a pivotal role in determining degenerative retinal diseases, yet it is a difficult problem. Due to successive resolution loss in the encoding phase combined with the inability to recover this lost information in the decoding phase, autoencoding based segmentation approaches are limited in their ability to extract retinal microvascular structure. We propose RV-GAN, a new multi-scale generative architecture for accurate retinal vessel segmentation to alleviate this. The proposed architecture uses two generators and two multi-scale autoencoding discriminators for better microvessel localization and segmentation. In order to avoid the loss of fidelity suffered by traditional GAN-based segmentation systems, we introduce a novel weighted feature matching loss. This new loss incorporates and prioritizes features from the discriminator’s decoder over the encoder. Doing so combined with the fact that the discriminator’s decoder attempts to determine real or fake images at the pixel level better preserves macro and microvascular structure. By combining reconstruction and weighted feature matching loss, the proposed architecture achieves an area under the curve (AUC) of 0.9887, 0.9914, and 0.9887 in pixel-wise segmentation of retinal vasculature from three publicly available datasets, namely DRIVE, CHASE-DB1, and STARE, respectively. Additionally, RV-GAN outperforms other architectures in two additional relevant metrics, mean intersection-over-union (Mean-IOU) and structural similarity measure (SSIM).
Link to paper
DOI: https://doi.org/10.1007/978-3-030-87237-3_4
SharedIt: https://rdcu.be/cyl9A
Link to the code repository
https://github.com/SharifAmit/RVGAN
Link to the dataset(s)
N/A
Reviews
Review #1
- Please describe the contribution of the paper
This paper describes a novel dual-GAN network (one gan extract coarse structures and the other one finds fine details) to segmentation retinal vessels on funds images. Technical contribution is solid and experimental results are 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.
1) The GAN-network is novel with two sets of Generator and discriminator networks. One is used to extract coarse structures, and the other one to identify fine details. 2) The encoder and decoder paths are redesigned for vessel segmentation in the generator networks, including distinct identity blocks and spatial feature aggregation 3) The loss of the entire network is novel, as it combines adversarial loss and weighted feature matching loss. 4) Experimental results are convincing as it shows the improvement of microvascular segmentation (Figure 1) and accuracy improvement over the existing approaches (Table 1).
- 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 main weakness is the minor improvement (<1%) over existing approaches (Table 1), and some accuracy measurements are even worse. It is questionable for this method to be used in the real clinical circumstances. In the real practice, simple and stable network is preferable while the proposed method is very complicated (contains two GANs).
- 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
It is doable, but the proposed method is very complicated.
- Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review: https://miccai2021.org/en/REVIEWER-GUIDELINES.html
1) It is not clear how two generator networks (extracting coarse and fine details) are combined to achieve the final vessel segmentation. 2) The number of training, validation, and testing datasets in three different public datasets, DRIVE, CHASE-DBI, and STARE are unclear. 3) Although the proposed method seems to outperform existing approaches, the improvement is minor (less that 1%) and some accuracy measurements are even worse than existing approaches. I suspect the clinical value of the proposed method as it is a very complicated approach. It would be more stable to use the simple network structure, such as baseline U-Net. 4) It is well known that GAN network is not stable to train, while the proposed method contains two GANs, I would strongly encourage the authors to perform stability analysis as well as report computational time.
- Please state your overall opinion of the paper
borderline accept (6)
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Minor accuracy improvements with the usage of a very sophisticated network structure.
- What is the ranking of this paper in your review stack?
1
- Number of papers in your stack
5
- Reviewer confidence
Very confident
Review #2
- Please describe the contribution of the paper
The authors propose a new architecture for a better segmentation of the retinal vascular tree. I found the method interesting, specially with respect to the microvessels, the real challenge of this issue in retinographies.
The authors tested the method with public datasets of reference and compared with other reference architectures.
- 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 analysis is of significant interest.
The extraction of the retinal vascular tree presents a significant potential in the analysis of many eye and systemic diseases as hypertension or diabetes, motivated over the years.
The performance of the designed architecture is adequate.
The method is evaluated adequate and with several public datasets of interest.
The method is compared with other architectures of contrasted performance in related issues.
The limitations and potential improvements of the methods are not discussed.
- 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 English can be improved, please revise the redaction for a better clarity of the reader. Details as, for reference, space before [cites] or without space must be uniformized.
The are some loss decisions, as Hinge-Loss that must be contextualized and motivated the inclusion in the issue.
There are some design characteristics that must be better motivated in the context of the performance of the microvessels. The slight tubular representation of the small vessels is exploited by this architecture to obtain a better performance, but in this line there are many architectural design that is not clear the reason of the inclusion and the real impact in the results.
In that line, I found missing some ablation studies to motivate them.
The authors indicate the potencial of this analysis in degenerative diseases, without not further reference or contextualization.
- 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 authors describe the method to be reproduced sufficiently. No further information or code is reported about the work. In that line the authors just describe the details of the architecture to be reproduced and the experiments that were conducted in public datasets.
- Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review: https://miccai2021.org/en/REVIEWER-GUIDELINES.html
In line with the weaknesses that were described, I think the manuscript and work can be improved in several lines:
In terms of redaction, English and slight details as the space or not space before the cites.
Ablation studies are desired, specially in this kind of works with many architectural design decisions and inclusions.
In line with the architecture, also the motivation of this design in relation with the characteristics of the problem, specially the microvessels that is the real actual problem in retinal vascularity can be better contextualized.
Limitations and potential improvements of the work, despite the accurate results, would enrich the manuscript.
Also, the contextualization of sentences as those related with degenerative diseases have to be better motivated to see the real link and potential of this analysis.
- Please state your overall opinion of the paper
accept (8)
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Despite the limitations and potential improvements that can be made in the manuscript, my decision is based on the issue, the context, the architectural proposal that seems to be interesting, the validation in several public datasets and the comparison with other architectures with accurate performance in the problem.
- What is the ranking of this paper in your review stack?
2
- Number of papers in your stack
4
- Reviewer confidence
Very confident
Review #3
- Please describe the contribution of the paper
This study introduces a multi-scale GAN for retinal vessel segmentation. The network uses a novel feature matching loss. The network is evaluated on 3 standard retinal segmentation datasets and shows SOTA (or close to SOTA) performance on all 3.
- 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.
- fine and coarse generators
- SOTA
- tested on 3 popular 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.
- N/A
- 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 indicate that they will release the code in great detail including train/test code and pre-trained models. The community will be thankful for this
- Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review: https://miccai2021.org/en/REVIEWER-GUIDELINES.html
The paper is very well written. I have only very minor comments:
- Eqs. 1 & 2: please explain the different terms in the equation
- A nitpicky comment on style: Please try to avoid contractions such as “that’s” in scientific writing
- Please state your overall opinion of the paper
strong accept (9)
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
This is a strong contribution. The network architecture is novel, the model is evaluated on 3 popular datasets using a range of common & meaningful metrics, and the results are convincingly either SOTA or close to SOTA.
- What is the ranking of this paper in your review stack?
1
- Number of papers in your stack
5
- Reviewer confidence
Very confident
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 was reviewed by three experts. All three reviewers consistently agree the novelty and contributions of the work. The AC would recommend acceptance of the work.
- 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).
1
Author Feedback
We thank the reviewers for the suggestions and constructive comments on the paper. We will address comments made by the reviewers in the camera-ready version of the paper. However, there are some misinterpretations in the reviewer’s findings that are clarified below.
In response to Reviewer 1:
- Comment 6.2: Detailed information for training has been provided on Page 7, Subsection 3.1, lines 3-7.
- Comment 6.3:Quantitatively, the faster inference is more critical for deployment in medical applications. Hence, we mentioned the inference speed on Page 7, Subsection 3.2, line 7. Qualitatively, Mean-Intersection-over-Union and SSIM metrics are considered the gold standard for measuring Image Segmentation task over other metrics. We have shown that anatomically less error-prone, more accurate segmentation of microvessels results in better clinical applicability, as elaborated on Page 8, Subsection 3.4, lines 13-17.
In response to Reviewer 2:
- Comment 6.2 & 6.3: The motivations for each block have been presented on Page 4, Subsection 2.3, lines 1-7, and Page 4, Subsection 2.4, lines 4-7. Also, the design choices to extract both macro and micro-vessel using global and local feature extracting architecture have been explained in detail on Page 3-4, Subsection 2.1, lines 3-9 and Page 5, Subsection 2.5, lines 1-7.
- Comment 6.5: A detailed discussion on degenerative diseases that affect retinal vasculature and the motivation for a system that solves the underlying problem has been elaborated on Page 1-2, Section 1, lines 1-9.