Review Reports - Exploiting the Generative Adversarial Network Approach to Create a Synthetic Topography Corneal Image

Round 1

Reviewer 1 Report

In my opinion, although the manucript is analysing an interesting set of data, it does not fit at all in the scope of Biomolecules. No "biomolecules" have been included in the manuscript. I suggest another type of Journal for submission.

Author Response

Reviewer 1

Comments and Suggestions for Authors

Authors response:

Thank you so very much for your valuable feedback.

Reviewer 2 Report

The manuscript proposed to apply a generative adversarial network (GAN) to the cornea dataset to create the synthetic image. This could help to improve the classification accuracy. The experimental results show that the proposed structure achieves better results compared to the baseline models. Here are several comments on the manuscript:

1. The novelty of the manuscript is limited since the structure only combines the existing models (GAN and CNN). It can be improved more by opening the dataset to the Public or improving the GAN structure.

2. Please provide the comparison data before and after GAN for qualitative comparison. Please describe in detail what is the difference between the data before GAN and After GAN. For example, GAN could remove the noise and so on. It can be seen with the crop part of the image, and zoom in to see the different contours.

3. Please calculate the SSIM and PSNR score for the data before and after GAN to see the quantitative comparison.

4. Please provide the inference or processing time for the proposed system to see if it is suitable for the real-time scenario.

5. It Would be very better to use the latest baseline from 2020 to 2022, such as [1, 2, 3]. It also could help to improve the accuracy.

Reference:

[1] C.-F. R. Chen, Q. Fan, and R. Panda, “Crossvit: Cross-attention multi-scale vision transformer for image classification,” Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 357-366, 2021.

[2] W. Xu, Y. Xu, T. Chang, and Z. Tu, “Co-scale conv-attentional image transformers,” Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 9981-9990, 2021.

[3] Z. Liu, Y. Lin, Y. Cao, H. Hu, Y. Wei, Z. Zhang, S. Lin, and B. Guo, “Swin transformer: Hierarchical vision transformer using shifted windows,” Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 10012-10022, 2021.

Author Response

The authors would like to thank the respected and distinct editors and reviewers for their valuable and treasured comments and suggestions. The following points are made by the authors to address the comments and suggestions that were raised by the respected editor and reviewers:

Reviewer 2

Comments and Suggestions for Authors

The novelty of the manuscript is limited since the structure only combines the existing models (GAN and CNN). It can be improved more by opening the dataset to the Public or improving the GAN structure.

Authors response:

Thank you so very much for your valuable opinion.

Unfortunately, we are not the dataset owner and have no authorization to open the dataset to the Public.

Please provide the comparison data before and after GAN for qualitative comparison. Please describe in detail what is the difference between the data before GAN and After GAN. For example, GAN could remove the noise and so on. It can be seen with the crop part of the image, and zoom in to see the different contours.

Authors response:

Thank you so very much for your valuable opinion.

The comparison data before and after GAN for qualitative comparison is described in figure 4 and discussed in the third paragraph on page 10.

Please calculate the SSIM and PSNR score for the data before and after GAN to see the quantitative comparison.

Authors response:

Thank you so very much for your valuable opinion.

The SSIM and PSNR are calculated in table 5 on page 9, which are discussed in the first paragraph on page 10.

Please provide the inference or processing time for the proposed system to see if it is suitable for the real-time scenario.

Authors response:

Thank you so very much for your valuable opinion.

CNN classifier's Average-Time-Test (ATT) is calculated in table 6, which is discussed in the second paragraph on page 10.

It Would be very better to use the latest baseline from 2020 to 2022, such as [1, 2, 3]. It also could help to improve the accuracy.

Authors response:

Thank you so very much for your valuable opinion.

The references which are mentioned, including (Vision Transformer (ViT), Co-scale conv-attentional image Transformers (CoaT), and Swin transformer (Swin-T)) are utilized as follows:

In section 2.2 (Transfer Learning Models), the structure of the three models is described in the second paragraph on page 5.

The models with all different types of data are trained and tested, and the results are shown in Tables 2 and 3 on pages 8 and 9, respectively.

The results are discussed in section 4 (Discussion), the first and the second paragraph on page 10.

Reference:

[1] C.-F. R. Chen, Q. Fan, and R. Panda, “Crossvit: Cross-attention multi-scale vision transformer for image classification,” Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 357-366, 2021.

[2] W. Xu, Y. Xu, T. Chang, and Z. Tu, “Co-scale conv-attentional image transformers,” Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 9981-9990, 2021.

[3] Z. Liu, Y. Lin, Y. Cao, H. Hu, Y. Wei, Z. Zhang, S. Lin, and B. Guo, “Swin transformer: Hierarchical vision transformer using shifted windows,” Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 10012-10022, 2021.

Authors response:

Thank you so very much for your valuable opinion.

They are very useful and used them in the revised version.

Reviewer 3 Report

This paper, “Exploiting Generative Adversarial Network Approach, to Create a Synthetic Topography Corneal Image”

The proposed approach is shown clearly. The submission has several weak points, following points are not clear:

1 It is not clear when "Data Resampling" is applied, nor how it improves the efficiency of the compared classifiers.

2 In section 3 “Results”, line 297 says “Table 3”, I think the correct one is Table 2.

3 It is recommended to add further information for figure 2 “Structure of the proposed system”

In section 2.3 you explain “Generating Synthetic Cornea Images”, I recommend presenting it as an algorithm as well.

5 Give more information about the values of the parameters used in the classifiers. They could be presented in a table.

Author Response

Reviewer 3

Comments and Suggestions for Authors

This paper, “Exploiting Generative Adversarial Network Approach, to Create a Synthetic Topography Corneal Image”

The proposed approach is shown clearly. The submission has several weak points, following points are not clear:

1 It is not clear when "Data Resampling" is applied, nor how it improves the efficiency of the compared classifiers.

Authors response:

Thank you so very much for your valuable opinion.

We clarify the Data Resampling usage and explain its effeteness on the classifier performance. Please see the text in the results section on lines 322-326.

2 In section 3 “Results”, line 297 says “Table 3”, I think the correct one is Table 2.

Authors response:

Thank you so very much for your valuable opinion.

The table numbers have been corrected on page 9.

3 It is recommended to add further information for figure 2 “Structure of the proposed system”

Authors response:

Thank you so very much for your valuable opinion.

The structure of the proposed system is explained in detail in section 2.3 (Generating Synthetic Cornea Images) on page 6, lines 214-254.

4 In section 2.3 you explain “Generating Synthetic Cornea Images”, I recommend presenting it as an algorithm as well.

Authors response:

Thank you so very much for your valuable opinion.

Response: The pseudocode is added on page 5

5 Give more information about the values of the parameters used in the classifiers. They could be presented in a table.

Authors response:

Thank you so very much for your valuable opinion.

Response: Values of the parameters are added in Table 1, on page 6

Bottom of Form

Round 2

Reviewer 2 Report

Thank you very much for answering all of the comments. Good luck.