Lung’s Segmentation Using Context-Aware Regressive Conditional GAN

Round 1

Reviewer 1 Report

The paper proposes a new way of dealing with COVID-19 detection. I would like to ask the following questions and address some remarks:
a) as all COVID-19 names are in capitals I think that it should be the same in line 17,
b) did the authors consider an ensemble approach that could also be a solution (even with the known models)?
c) how do the initial GAN settings affect the outcome?
d) what's the learning time in the case of the proposed approach compared to other approaches?

Author Response

Thank you for giving me the opportunity to submit a revised draft of my manuscript. We are thankful for the time and effort you have dedicated to providing your valuable feedback on the manuscript. Answers to the suggestions are provided in a separate file attached here and incorporated in the revised paper version.

Author Response File: Author Response.docx

Reviewer 2 Report

In Abstract section，What's DSC，AJC ？

What's the novelty of this study?

It is not enough  only provides results in Results and discussion section. Discussion need to deepen. 

What are the advantages and disadvantages of this work?

Why the Original Images of Figure 2 (a) and Figure 3 (a) are different?

Author Response

Thank you for allowing me to submit a revised draft of my manuscript. We are thankful for the time and effort you have dedicated to providing your valuable feedback on the manuscript. The point-by-point answers to suggestions are provided in the file attached and suggested changes are incorporated in the revised version of the manuscript.

Author Response File: Author Response.docx

Reviewer 3 Report

 

The paper proposes a deep learning method for lung segmentation in CT images to detect COVID-19 infection and to classify the severity of the infection. The method is based on a conditional generative adversarial network. It provides up-to-date literature overview on deep learning methods applied to detect COVID infections in CT images. The proposed method was evaluated on more than 50 thousand CT images from several datasets. The method was able to improve the results compared to other state-of-the-art deep learning methods.

I recommend introducing shortly the methods selected for the evaluation. Why were they selected?

Some further comments:

Line 24: I do not understand: the values of the presented evaluation metrics for lesions segmentation in the abstract because they belong to GAN not to the proposed CGAN according to Table 4.

Line 25: Please specify what the given evaluation metrics refer to. Do they refer also to lesions segmentation?

Line 112: The first sentence of the paragraph belongs logically to the previous paragraph.

Line 135-136: The first two sentences of the paragraph belong logically to the previous paragraph.

Line 123: Please define all acronyms in the text (e.g. AUC)

Line 146: Please use upper case first letter (L) in the literature reference: Li et al.

Line 245: I do not understand: MosMedData contains 784 slices according to the text. However, it contains more than 46 thousand CT images according to Table 1.

Eq 5: Please use "|" before Pk in the formula instead of "I".

Line 285: i index is not needed in lower index of P: P_{i=} -> P =

Line 302: Mark-RCNN - > Mask-RCNN

Line 308: Figure 2 -> Figure 4

Line 348: The text mentions sensitivity but it is not shown in Table 5.

Please be consistent in using the mathematical notations:

Line 163: V(D,G) is written by using lower case v in the text while upper case V is used in Eq. 1.

Line 164: Pz is written by using upper case p while lower case p is used in Eq. 1.

I recommend using names consistently. Use only one variant to name the same principle:

- cGAN or CGAN

- UNet or U-Net

There are some spelling mistakes.

Please put . after viz in the text ("viz" occurs several times in the text).

Line 302: Please use upper case initial for Table 3.

Line 328: the best segmentation results than -> better segmentation results than

Line 339: double ","

 

 

 

Author Response

Thank you for giving me the opportunity to submit a revised draft of my manuscript. We are thankful for the time and effort you have dedicated to providing your valuable feedback on the manuscript. The point-by-point answers to suggestions is provided in the file attached and suggested changes are incorporated in the revised version of the manuscript.

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

Figure 3 and Figure 4 present results using proposed CGAN, Why using different original images for Figure 3 and Figure 4?  It will be better to use the same orignial images to present results for preprocessing, lungs, and lesions segmentation.   Use the same orginal images for Figure 2 (a) and Figure 3 (a)   in order to provide the results  before  and after (d) preprocessing.

Author Response

Thankyou for your time for providing valuable comments. A separate file is attached to answer point by point to the comments.

Author Response File: Author Response.docx