MS-CANet: Multi-Scale Subtraction Network with Coordinate Attention for Retinal Vessel Segmentation

Round 1

Reviewer 1 Report (Previous Reviewer 3)

Overall this is an exciting paper  “MS-CANet: Multi-Scale Subtraction Network with Coordinate Attention for Retinal Vessel Segmentation”; this paper develops several approaches to improve the segmentation results for multiple datasets and then compared them with other state-of -arts.

 

 

Comments to the Author

 

1-     The article still needs more effort to be published. The novelty and the contribution are still confusing and need to be clear and precise; it seems that there is not enough contribution in this paper compared with previous work shown in reference, it offers several approaches result with the excellent result without explaining why and how those results good enough compared with other approaches.

2-     The author needs to show mathematically how their approach has better results than others.

 

3-     The number of experiments is statistically still insufficient to provide precise results in this article.

 

4-     Some language errors are observed that should be considered, such as small letters starting at the begging of the paragraph.

Author Response

Thank you for reviewing. Please see the attached file for details.

Author Response File: Author Response.pdf

Reviewer 2 Report (New Reviewer)

Interesting work carried out by the authors in the manuscript. Authors expected to respond the following queries before proceeding further.

1.      Review of literature is not sufficient.

2.      Recent references are needed to be included

3.      Introduction part can be more elaborative.

4.      The novelty of the proposed work needs to explain clearly in the abstract itself

5.      Please provide flowchart to explain the process carried out in the manuscript

6.      The size of the dataset needs to be explained clearly

7.      What is the significance of evaluation metrics used in the work? It needs to be explained clearly.

8.      Please provide confusion matrix for easy understanding

9.      Proofread the entire manuscript for English corrections

 

Author Response

Thank you for reviewing. Please see the attached file for details.

Author Response File: Author Response.pdf

Reviewer 3 Report (New Reviewer)

The authors propose a novel deep learning based approach, called multi-scale subtraction network (MS-CANet), to segment the vessels in retinal vessel images.

For what concerns the technical point of view, the paper is well described and clear. In particular, the results are well presented and the assessment procedure is exhaustively described.

However, some concerns arise about both the experimental design and also the paper structure. They are listed above.

1) The abstract is very detailed, which is in fact a good point since it allows for directly diving into the work. However I believe the authors should avoid reporting the numerical results already in the abstract.

2) I do not fully understand the editorial choice for the paper, i.e., the journal "Symmetry". I would have expected some more justification, in addition to addressing the problem of asymmetry of small blood vessels. I suggest the authors to better emphasize their chioce.

3) The authors devote a lot of space to the description of the evaluation metrics they used, while I think a list with appropriate references would have been enough.

4) The authors only take into account deep learning based approaches for comparison and for the state-of-the-art revision. On the one hand, it is known that deep learning based methods do work efficiently, so they should always be mentioned when referring to image processing and in particular image segmentation. On the other hand, I believe it is too common that researchers fall in the convenient and trending triviality of randomly combining deep-learning based methods to address segmentation tasks, not mentioning or ignoring what are the other approaches available. Also, I think that the context of image segmentation should be better pictured, in particular when submitting to a journal like this one where readers may not be expert in such a field, given the wide range of suitable topics. For the authors convenience, a list of work that could be cited in the introduction and used for further comparison follows. The authors may not limit themselves to that, but also expand it.

     [1] Panagiotakis, C., Papadakis, H., Grinias, E., Komodakis, N., Fragopoulou, P., & Tziritas, G. (2013). Interactive image segmentation based on synthetic graph coordinates. Pattern Recognition, 46(11), 2940-2952.

     [2] Zhao, Q. H., Li, X. L., Li, Y., & Zhao, X. M. (2017). A fuzzy clustering image segmentation algorithm based on hidden Markov random field models and Voronoi tessellation. Pattern Recognition Letters, 85, 49-55.

     [3] Vargas-Muñoz, J. E., Chowdhury, A. S., Alexandre, E. B., Galvão, F. L., Miranda, P. A. V., & Falcão, A. X. (2019). An iterative spanning forest framework for superpixel segmentation. IEEE Transactions on Image Processing, 28(7), 3477-3489.

     [4] Filali, H., & Kalti, K. (2021). Image segmentation using MRF model optimized by a hybrid ACO-ICM algorithm. Soft Computing, 25(15), 10181-10204.

     [5] Kucybała, I., Tabor, Z., Ciuk, S., Chrzan, R., Urbanik, A., & Wojciechowski, W. (2020). A fast graph-based algorithm for automated segmentation of subcutaneous and visceral adipose tissue in 3D abdominal computed tomography images. Biocybernetics and Biomedical Engineering, 40(2), 729-739.

     [6] Trombini, M., Solarna, D., Moser, G., & Dellepiane, S. (2023). A goal-driven unsupervised image segmentation method combining graph-based processing and Markov random fields. Pattern Recognition, 134, 109082.

I do believe this work has potential interest and impact to be considered for publication, and I think that addressing all the aformentioned issues the paper will significantly improve so that it can be published.

Author Response

Thank you for reviewing. Please see the attached file for details.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report (Previous Reviewer 3)

NA

Author Response

I would like to express my sincere gratitude for your invaluable contributions and insightful comments throughout the review process. Your detailed feedback has been instrumental in improving the quality of our manuscript, and we are truly grateful for your time and effort. Your constructive criticism and thoughtful suggestions have helped us to refine our work and clarify our ideas. We greatly appreciate your dedication and expertise, which have undoubtedly strengthened our research. Thank you once again for your valuable feedback and for your commitment to maintaining the highest standards of academic rigor. We look forward to your continued support and collaboration in the future.

Reviewer 2 Report (New Reviewer)

In my view, the authors are expected to respond to the following queries before proceeding further,

The authors are requested to provide the confusion matrix and they are also expected to provide an explanation towards the size of the dataset adopted. The size of the dataset is enough? Please explain.

 

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 3 Report (New Reviewer)

I thank the authors for their reply to all the concerns I raised. I have no further comments, in my opinion the paper can be accepted.

Best luck for your work.

Author Response

Thank you for considering our responses to the concerns you raised about our paper. We appreciate your time and effort in reviewing our work, and we are glad that you have no further comments. We are honored and grateful that you believe our paper is worthy of acceptance. Once again, we want to express our gratitude for your valuable feedback, which helped us improve the quality of our paper. We will continue to strive for excellence in our future research. 

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

Authors have developed MS-CANet for retinal vessels segmentation from the color fundus images. The idea is appreciated but the paper need to be rechecked on the following points.

The description about the model need to be improved for better understanding by the reader. Many cases typo and grammatical mistakes are found. Few of are mentioned.  Page 2 line no 56. "In addition" was written twice. Page 4 line number 156 module is written twice. 

Methodology discusses in Section 2.2 need to be improved with description about each part in the block. 

Figures are tiny and need to be enlarged for better visualization. 

Proper organization and discussion about the tables are missed. 

The work need to be compared with the recent published work and mentioned below. 

Mithun Kumar Kar, Debanga Raj Neog, and Malaya Kumar Nath, "Retinal Vessel Segmentation using Multi-scale Residual Convolutional Neural Network (MSR-Net) Combined with Generative Adversarial Networks", Circuits, Systems and Signal Processing: Springer Nature, pp: **, vol. **, no. **, September 2022. https://doi.org/10.1007/s00034-022-02190-5

 

References need to be uniform. 

Reviewer 2 Report

1. The gaps in the existing works need to be consolidated and discussed at the end of Introduction section.

2. Statistics regarding dataset can be given as a table

3. Fig 1 is too small. Clarity can be improved.

4. Multiscale attention is already applied in the existing works. Authors need to cite and discuss how their work is different from the following works. 

Multiscale attentional residual neural network framework for remaining useful life prediction of bearings

Multi-scale attention network for image super-resolution

Ischemic Lesion Segmentation using Ensemble of Multi-Scale Region Aligned CNN

5. Details of hyperparameter tuning need to be presented.

6. Plots of convergence for loss and accuracy need to be presented. 

Reviewer 3 Report

1-     The Math model is not clearly mentioned in the article. The updating network's parameters, such as the weight and others need to be shown in this article. The network structure must also be included to show the complexity of this approach.

 

 

2-     Is there any limitation that can be considered in this work such as the image size?

 

 

 

3-     The author claim in the abstract that” Most improved methods based on U-shape networks for retinal vessel segmentation use element-wise addition or dimension-concatenate to fuse the differences between different network layers. However, these operations easily generate redundant information, which will weaken the complementarity between features of different layers and neglect the location information of feature maps, resulting in inaccurate retinal vessel localization and blurred vessel edges. To address this, would you clarify how to evaluate this problem? Need to explain more detail and show that in the paper.

 

4-     The author needs to state the complexity of the algorithm since the results are almost the same with another study except for a small difference as shown in most of the results tables, so if the complexity is expensive, then it is not worth using this approach.

 

 

5-      As mentioned in the dataset section, the dataset is still not huge to train the network, how does the author train the network with this number of images? Please state this point.