Statistical Validation of Synthetic Data for Lung Cancer Patients Generated by Using Generative Adversarial Networks

Round 1

Reviewer 1 Report

Dear authors,

 

thank you very much for handing in the manuscript. First, I have to say, that I am not a native speaker, therefore I will not comment on your English, as I think I am not qulified enough for that. Still, I think there are some things to talk about, from a structure and content perspective. See below my feedback:

 

Overall:

In my opinion, the manuscript is quite poor in its current state. The topic is great, also from a content perspective it is promising, but the current state of the manuscript is poor. I have the impression, that the manuscript was written by one person, and the manuscript was handed in without the review of any of the co-authors. The reason why I think, that has happened is listed below:

Introduction:

I am sure you have expected this. It is something anybody sees without digging deep into the manuscript. This is one of the reasons, why I think, what I have mentioned above. The literature that you refer to in your introduction is not nearly enough for a scientific approach. Especially since a major part of your references are papers of your own. I understand that this is reasonable at some point, as the manuscript covers the next step in your project. At the same time you should consider referencing other papers that are reasonably close to the things you have done. I will not advise you on which papers you should include. Still, there are several papers out there, that use loss-functions or experts for validation in heath care (also for cancer recognition) --> So, to support your argument in e.g.:

"For this reason, metric is usually replaced by subjective ‘expert decision’, oncologists in this case."

you should reference some papers, that did that. The whole introduction is full of such claims, that are not supported by a reference.

This is not how you should write an Introduction. This is not how you should approach such a problem in a project. First you do the literature review, not after you have done something in your project.

In my opinion, this has to be done before, any further review cna be done on the manuscript. Still though, I have read the manuscript and want to give you feedback on the rest as well:

 

Materials and methods:

You tell parts of the results in the materials and methods (line 84). You should not do that. Try to be clear and concise. This means, that you should follow a logical path throughout your paper. In my opinion you do not follow a logical path, but you jump between topics and sections. Don't do that. (Again, this is something you see, when you read the paper, which is why I think you have not done it)

The graphs that you show in the section are not well done. You call it a histogram. Is the y-axis given in counts or %? I could not find a hint about it. And if it is %, why do we have a total of approx. 25% in figure 1 (left, As you have not noted the  (a), (b), (c) or alike, there is no other way to tell you which figure I mean). For a pontential reader it is quite annoying not to know these kind of things. (Again, this is something you see, when you read the paper, which is why I think you have not done it)

 

Results, Discussion, Conclusion:

Transforming the data, so it fits the GAN approach is quite clever in my opinion. Others have done it as well. You should support this approach in your discussion by referencing others doing similar things with a GAN approach (also outside of health care, there is quite a lot of research).

I also think the discussion is done quite poorly. It also reflects the missing review of other literature. A few points have been discussed though and especially the limitations are part of it. I like that.

From a results perspective, there is still room for improvement (e.g. Performance fo Discriminator on synthetic data in comparison to real one). Nevertheless, the results are done quite well (except the graphs, add more information for better referencing them in the text)

 

I hope the review does not sound to harsh. Bottom line. Do a better job from a scientific and documenting point of view, content wise is ok.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

In this paper (entitled " Statistical Validation of Synthetic Data for Lung Cancer Patients Generated by using Generative Adversarial Networks"), authors propose statistical decision-making as a validation tool: is the model good enough to use? Does the model pass rigorous hypothesis testing criteria?. However, there are some issues that should be addressed before this work can be accepted. The detailed comments are given as follows:

1.        There are grammatical errors in the paper. Please proofread them carefully and correct them.

2.        When preprocessing the data, what method is used for normalization?

3.        In Figure 2, some fonts are too large, and some fonts are too small, so it is recommended that the author modify them.

4.         the formulas should be numbered.

5.        It is suggested that the author provide the derivation and proof process of the formula, or explain the origin of the formula in detail, such as the loss function.

6.        How is the performance index of the discriminator selected?

7.        Can the author provide the Chi-square hypothesis testing process?

8.        The reference list can be enhanced. Some recent intelligent methods have potential to deal with the problem.

[1] "Synthetic CT generation of the pelvis in patients with cervical cancer: a single input approach using generative adversarial network." IEEE access 9 (2021): 17208-17221.

[2] How to Simplify Search: Classification-wise Pareto Evolution for One-shot Neural Architecture Search, arXiv preprint arXiv:2109.07582

[3] Evolutionary computation for solving search-based data analytics problems, Artificial Intelligence Review 54 (2), 1321-1348

[4] "Generative adversarial networks in digital pathology and histopathological image processing: A review." Journal of Pathology Informatics 12.1 (2021): 43.

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Dear authors,

 

again, thank you for handing in the manuscript. It improved a lot in my opinion. All my arguments have been heard in a good way. I still think, that there could have been a better literature review, but if it is ok with the editor it is ok with me. Congratulations, I will recommend to accept the paper in its current form.

Author Response

Dear reviewer, thank you again for your consideration.

Reviewer 2 Report

My main concerns have been addressed.

Author Response

Dear reviewer, thank you again for your consideration.