A Semantic Segmentation Method with Emphasis on the Edges for Automatic Vessel Wall Analysis
Round 1
Reviewer 1 Report
This paper presents an interesting work on developing a new deep learning network, named as EVSegNet to segment arterial vessel wall from MR images. This network has one regular stream, and one edge stream. The method has been explained well, with adequate test images, and the performance, when compared to some other deep learning network, shows that EVSegNet performs the best. Thus, in my opinion, this paper will be useful for the audiences from image processing, deep learning, or medical background. The current quality is up to the standard for a journal publication.
However, there are very minor comments from me, to further improve the quality of the paper.
1. The abstract is well written and concise. It contains all required
information; background, method, results and conclusion. However, I
am not familiar with the sectioned abstract. If this is the format
of the journal, then, it is okey. Else, the authors should remove
the section's headers from the abstract.
2. The introduction is well written, where the authors start with the
MR vessel wall imaging followed to methods of segmentation, then the
introduction to the deep learning. It is good to see that the
review includes references up to year 2020. However, it would be
better if the authors could include some more recent related
references, especially from year 2021 and 2022.
3. At the end of Section 1, it would be better if the authors presented
the outline of the paper.
4. The authors proposed a new network named EVSegNet. Yet, it would be
better if the full name of this method is also provided. For
example, "Edge Vessel Segmentation Network (EVSegNet)".
5. All mathematical symbos should be typed in Italic. For examples, on
page 3, symbols H, W, and C should be presented in Italic.
6. On lines 115, 116, 117 and 119, symbol d2, should remove '^'
symbol, and '2' should use superscript.
7. To make it more accurate, in equations (3) and (6), the authors
should specify whether it is log_2, or log_10.
Author Response
Point-by-point response to the comments of the referee(s) and editor on applsci-1789449 titled “A Semantic Segmentation Method with Emphasizing on the Edges for Automatic Vessel Wall Analysis”
We would like to thank the editor and reviewers for the constructive comments. Our response to each comment is shown below in italics.
Comments for the Authors:
Reviewer #1：
This paper presents an interesting work on developing a new deep learning network, named as EVSegNet to segment arterial vessel wall from MR images. This network has one regular stream, and one edge stream. The method has been explained well, with adequate test images, and the performance, when compared to some other deep learning network, shows that EVSegNet performs the best. Thus, in my opinion, this paper will be useful for the audiences from image processing, deep learning, or medical background. The current quality is up to the standard for a journal publication.
However, there are very minor comments from me, to further improve the quality of the paper.
1. The abstract is well written and concise. It contains all required information; background, method, results and conclusion. However, I am not familiar with the sectioned abstract. If this is the format of the journal, then, it is okey. Else, the authors should remove the section's headers from the abstract.
Response: Thanks for your suggestion. The abstract is now revised as the usual structure according to the suggestion.
2. The introduction is well written, where the authors start with the MR vessel wall imaging followed to methods of segmentation, then the introduction to the deep learning. It is good to see that the review includes references up to year 2020. However, it would be better if the authors could include some more recent related references, especially from year 2021 and 2022.
Response: Thanks for your suggestion. The 2021 reference “[11] Alblas, D. et al. " Deep Learning-Based Carotid Artery Vessel Wall Segmentation in Black-Blood MRI Using Anatomical Priors." arXiv e-prints, 2021.” and 2022 reference “[12] Zhou, H. et al. “Intracranial Vessel Wall Segmentation with Deep Learning Using a Novel Tiered Loss Function to Incorporate Class Inclusion.” 2022 IEEE 19th International Symposium on Biomedical Imaging (ISBI), 2022, pp. 1-4, doi:10.1109/ISBI52829.2022.9761428.” are now added to the third paragraph of “Introduction” section.
3. At the end of Section 1, it would be better if the authors presented the outline of the paper.
Response: Thanks for your suggestion. The outline is now added to the end of Section 1.
4. The authors proposed a new network named EVSegNet. Yet, it would be better if the full name of this method is also provided. For example, "Edge Vessel Segmentation Network (EVSegNet)".
Response: The full name of this method is provided in the third sentence of “Abstract” Section and first sentence of “Network Architecture” Section.
5. All mathematical symbos should be typed in Italic. For examples, on page 3, symbols H, W, and C should be presented in Italic.
Response: Thanks for your suggestion. All mathematical symbos are now revised in Italic.
6. On lines 115, 116, 117 and 119, symbol d2, should remove '^' symbol, and '2' should use superscript.
Response: Thanks for your suggestion. The '^' symbol is now removed and revised to use superscript.
7. To make it more accurate, in equations (3) and (6), the authors
should specify whether it is log_2, or log_10.
Response: Thank you for pointing out this. The log is now revised to log_e in equations (3) and (6).
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The aim of this research was to evaluate the EVSegNet by adding Resnet, DUC, HDC, edge stream block, to develop a precise semantic segmentation method with emphasizing on edges for automated segmentation of arterial vessel wall and plaque. The study could be considered for further processing after a revision, based on the following points:
Abstract
· Incomplete sentences. For example in method section: A total of 124 subjects’ MR vessel wall images. What does it mean?
Module Hybrid Dilated Convolution (HDC) Module
Explain it in more details. What is it? What is the parameters r and k?
· Figure 3 and Figure 4 are the same with different captions. What is the difference?
Further proofreading/editing is recommended
Author Response
Point-by-point response to the comments of the referee(s) and editor on applsci-1789449 titled “A Semantic Segmentation Method with Emphasizing on the Edges for Automatic Vessel Wall Analysis”
We would like to thank the editor and reviewers for the constructive comments. Our response to each comment is shown below in italics.
Comments for the Authors:
Reviewer #2：
The aim of this research was to evaluate the EVSegNet by adding Resnet, DUC, HDC, edge stream block, to develop a precise semantic segmentation method with emphasizing on edges for automated segmentation of arterial vessel wall and plaque. The study could be considered for further processing after a revision, based on the following points:
1. Abstract:
Incomplete sentences. For example in method section: A total of 124 subjects’ MR vessel wall images. What does it mean?
Response: Thanks for pointing out the mistake. The sentence “A total of 124 subjects’ MR vessel wall images” is now revised as “A total of 124 subjects’ MR vessel wall images were used to train, validation, and test the model using deep learning.”.
2. Module Hybrid Dilated Convolution (HDC) Module
1） Explain it in more details. What is it? What is the parameters r and k?
Response: Sorry for the unclear expression about this. The Hybrid Dilated Convolution (HDC) model is built by stacking the dilated convolution kernels with a different rate, r is the dilation rate and k is kernel size. These are now added to the fifth sentence and tenth sentence in the “Hybrid Dilated Convolution (HDC) Module” section for clearer.
2）Figure 3 and Figure 4 are the same with different captions. What is the difference?
Response: Sorry for the mistake, Figure 4 is now revised. The difference between Figure 3 and Figure 4 is that the dilation rate of Figure 3 is the same, r-2, and the dilation rates of Figure 4 are different, r=1,2,3, respectively.
Author Response File: Author Response.pdf
