Generating Synthetic Images for Healthcare with Novel Deep Pix2Pix GAN

Round 1

Reviewer 1 Report

In this paper, the authors proposed deep pix2pix generative adversarial networks (GAN) for generating synthetic medical images. The result has shown that the proposed approach can generate a new synthetic medical image from a different image with more accuracy than that of the other models. However, the following problem need to be improved.

(1) The English of paper is poor, please polish it again.

(2) The introductions in Sec.1 are a bit less to give the background of the proposed method.

(3) Each variable in the formula needs to be described.

(4) What is the contribution of U-Net?

(5) The variables in the text should be consistent with those in the formula.

(6) It would be better giving the pseudo code of the proposed algorithm.

(7) More experiments are needed to illustrate the advantages of the proposed algorithm.

Author Response

I would like to thank you for devoting precious time to read our paper and giving comments , please see attached documents. 

Author Response File: Author Response.pdf

Reviewer 2 Report

Structure of proposed generators is not well explained.

Please elaborate clearly, is it novel or extended or copied.

Entire manuscript requires through corrections.

Please add error tables after comparing your work with published works.

Where are acknowledgments?

Why very few references?

Author Response

I would like to thank you for devoting precious time to read our paper and giving comments , please see attached documents. 

Author Response File: Author Response.pdf

Reviewer 3 Report

This paper proposed deep pix2pix generative adversarial networks (GAN) for generating synthetic medical images. Although the concept proposed in the paper is innovative in the medical field, the writing quality of the article still needs to be greatly improved.

1.       The research background is too simplistic. It is suggested that more academic literature should be used to support and strengthen the discussion on the difficulties of collecting samples in the current medical field.

2.       In addition to Generative Adversarial Networks (GAN) that can be used for image processing, there should be more different deep learning image processing techniques. It is recommended to mention in the article and briefly explain the differences with GAN.

3.       The author should fully describe the experimental architecture and the relevant settings of the experiment

4.       The in-text description of Figure 2 is too simplistic

5.       In Figure 3, it is mentioned that the pix2pix model has the best experimental results, while the cyclegan model is the worst. However, if the reader is not a medical expert in the field, it is impossible to directly know what is a good identification result or a poor identification result from the pictures. The author should provide more objective data to explain the experimental results and explain why the pictures generated by Transferred images are good image conversion results.

The contribution of this paper is insufficient. Since the experimental results are discussed for only one type of medical imaging, readers cannot know whether the proposed method is applicable to other types of medical imaging.

Author Response

I would like to thank you for devoting precious time to read our paper and giving comments , please see attached documents. 

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The respond is OK.

Author Response

please see attached file

Author Response File: Author Response.pdf

Reviewer 2 Report

Article does not provide novelty.

Kindly improve the study. In algorithm 1, hyperparameter is 1, then how are the results satisfactory? Please justify.

Figure 2 is still absurd. 

My previous comments were deliberately ignore. 

Author Response

please see attached file

Author Response File: Author Response.pdf

Reviewer 3 Report

This paper proposed deep pix2pix generative adversarial networks (GAN) for generating synthetic medical images. In response to the reviewer's question, although the author has responded and corrected some articles, the correction results still have many defects, and the quality of the article still needs to be improved.

1. The author should give a more complete description of the current difficulties in medical image collection in the introduction chapter

2. The experimental process, software and hardware architecture, and system architecture are still not clearly explained in the experimental chapter

3. The main problem of this paper is that only using a single medical image as the experimental sample cannot verify whether the algorithm mentioned in the paper is effective. It is suggested that the author should conduct experiments on different medical images to verify the effectiveness of the proposed method

 

There is still a lot of room for improvement in the writing of this paper (for example, in the Related works chapter, a table can be used to organize the advantages, disadvantages and characteristics of related research). In addition, the experimental structure and environment should be more clearly explained so that readers can understand the rationality of the experimental design. It is suggested that the author should further explore the performance comparison between the proposed algorithm and other algorithms under the data of different medical images to highlight the academic and technical breakthrough of this paper.

Author Response

please see attached file

Author Response File: Author Response.pdf

Round 3

Reviewer 2 Report

Manuscript is improved but it still lacks novelty.

Although I asked twice, yet the authors have not provided the source 

of data. This is not ethical.

Please provide source of data with proper justification before 

bibliography.

Revise table 1.

You used 9 experiments, 3 models and 3 methods but table results are not clear to the reader. Is it error? Is it tumor size?

Carefully relabel the columns of the table please. 

Thanks.

Author Response

please see attached file 

Author Response File: Author Response.pdf

Reviewer 3 Report

This paper proposed deep pix2pix generative adversarial networks (GAN) for generating synthetic medical images.  This paper has been revised according to the comments of the reviewers, and it is recommended to accept this paper.

Author Response

please see attached file 

Author Response File: Author Response.pdf