A Survey on Deep-Learning-Based Diabetic Retinopathy Classification
Round 1
Reviewer 1 Report
Diabetic retinopathy is undoubtedly a serious threat to the modern population. So, the topic of the paper is relevant and up-to-date. The paper is well written, but need some changes before publication:
1. Abstract does not point the matter of the article. It should be rewrite considering the presentation what reader can expect in following pages. Now, it is only a brief introduction to the topic.
2. The numbers before sections’ is not necessary (as I remember MDPI template). Can be removed. Also, style of section names is inconsistent (e. g. second section is in capitals).
3. The authors focus only on fundus camera imaging. It should be pointed clearly because also good results are achieving using OCT imaging with CNNs. It worth mentioning that OCT has become an indispensable modality of investigation in the assessment of diabetic retinopathy.
4. Accuracy is not a good parameter for unbalanced datasets, the better would be F1-score.
5. Figure 1. the quality of text is poor while printing. Should be improved.
Author Response
Diabetic retinopathy is undoubtedly a serious threat to the modern population. So, the topic of the paper is relevant and up-to-date. The paper is well written, but need some changes before publication:
- Abstract does not point the matter of the article. It should be rewrite considering the presentation what reader can expect in following pages. Now, it is only a brief introduction to the topic.
Response: We have rewritten the abstract accordingly.
- The numbers before sections’ is not necessary (as I remember MDPI template). Can be removed. Also, style of section names is inconsistent (e. g. second section is in capitals).
Response: We referred to some other MDPI articles and found numbers before sections. Hence, we followed the same in this paper. We have made the section names consistent. Thanks for pointing this out.
- The authors focus only on fundus camera imaging. It should be pointed clearly because also good results are achieving using OCT imaging with CNNs. It worth mentioning that OCT has become an indispensable modality of investigation in the assessment of diabetic retinopathy.
Response: We had not included studies that used OCT images for diabetic retinopathy diagnosis with CNNs since the number of such studies is less. Moreover, OCT images are more high-resolution images that are similar to ultrasound images. Whereas, fundus images provide a view of the retina similar to what can be seen by human eyes. Due to this, the studies done on both differ from each other.
- Accuracy is not a good parameter for unbalanced datasets, the better would be F1-score.
Response: We strongly agree with this comment. But, unfortunately, F1-score is unavailable in most of the papers that were reviewed due to which we had chosen accuracy.
- Figure 1. the quality of text is poor while printing. Should be improved.
Response: We have replaced Figure 1 with a better-quality figure.
Reviewer 2 Report
The article entitled “A Survey on Deep Learning based Diabetic Retinopathy Classification” is well-written and, from my point of view, would be of interest for the readers of Diagnostics. In spite of these and before its publication I would recommend the following changes:
Line 36: please check if all the acronyms are defined before using them in the text. Like in the case of AI.
My main concern about the study is why the articles mentioned were included in this study and why quite possible others were not included. In other words this kind of topic should be treated like a systematic literature review. Please see some articles that would serve you as example:
Todorov, I.B.; Sánchez Lasheras, F. Forecasting Applied to the Electricity, Energy, Gas and Oil Industries: A Systematic Review. Mathematics 2022, 10, 3930. https://doi.org/10.3390/math10213930
Kissi, J.; Kusi Achampong, E.; Kumasenu Mensah, N.; Annobil, C.; Naa Lamptey, J. Moving towards Digitising COVID-19 Vaccination Certificate: A Systematic Review of Literature. Vaccines 2022, 10, 2040. https://doi.org/10.3390/vaccines10122040
Alruwaili, A.; Alanazy, A.R.M. Prehospital Time Interval for Urban and Rural Emergency Medical Services: A Systematic Literature Review. Healthcare 2022, 10, 2391. https://doi.org/10.3390/healthcare10122391
It would also be useful for the audience and future researchers if a guide for the future research is provided: how this research could be used concretely to open new pathways? Is it possible to provide some examples and possible directions for future research?
Author Response
Reviewer #2:
The article entitled “A Survey on Deep Learning based Diabetic Retinopathy Classification” is well-written and, from my point of view, would be of interest for the readers of Diagnostics. In spite of these and before its publication I would recommend the following changes:
Line 36: please check if all the acronyms are defined before using them in the text. Like in the case of AI.
Response: Thanks for pointing this out. We have defined all the acronyms in the abstract now.
My main concern about the study is why the articles mentioned were included in this study and why quite possible others were not included. In other words this kind of topic should be treated like a systematic literature review. Please see some articles that would serve you as example:
Todorov, I.B.; Sánchez Lasheras, F. Forecasting Applied to the Electricity, Energy, Gas and Oil Industries: A Systematic Review. Mathematics 2022, 10, 3930. https://doi.org/10.3390/math10213930
Kissi, J.; Kusi Achampong, E.; Kumasenu Mensah, N.; Annobil, C.; Naa Lamptey, J. Moving towards Digitising COVID-19 Vaccination Certificate: A Systematic Review of Literature. Vaccines 2022, 10, 2040. https://doi.org/10.3390/vaccines10122040
Alruwaili, A.; Alanazy, A.R.M. Prehospital Time Interval for Urban and Rural Emergency Medical Services: A Systematic Literature Review. Healthcare 2022, 10, 2391. https://doi.org/10.3390/healthcare10122391
Response: We strongly agree with this comment. But since there are innumerable studies in this area we decided to survey some papers and provide a briefing about the latest methods and datasets that are being used for diabetic retinopathy diagnosis. For this purpose, we attempted to include some papers which have used publicly available datasets and those which use the same metrics for evaluation.
It would also be useful for the audience and future researchers if a guide for the future research is provided: how this research could be used concretely to open new pathways? Is it possible to provide some examples and possible directions for future research?
Response: Thank you for the suggestions. We have included a new section named future directions.
Reviewer 3 Report
The authors presented the results of an analytical review of Deep Learning-based Diabetic Retinopathy methods classification. Strictly speaking, this review cannot be classified as systematic. However, therein lies its advantage. Readers will be able to navigate the modern methods of image recognition in the field of diagnosis of Diabetic Retinopathy, choose the appropriate method and data sets.
I have only a few minor suggestions for the design of the manuscript.
1. Explanations of tables and figures in the text are located too far from the tables and figures themselves, which makes it difficult to read. I propose to supplement the text with explanations of the tables and figures immediately before them.
2. I did not understand why the authors used bold and underline for some values in tables 3 and 4.
3. The use of links instead of method names in Figures 5 and 6 makes it difficult to read. I propose to supplement the links with the names of the methods.
Author Response
Reviewer #3:
The authors presented the results of an analytical review of Deep Learning-based Diabetic Retinopathy methods classification. Strictly speaking, this review cannot be classified as systematic. However, therein lies its advantage. Readers will be able to navigate the modern methods of image recognition in the field of diagnosis of Diabetic Retinopathy, choose the appropriate method and data sets.
I have only a few minor suggestions for the design of the manuscript.
- Explanations of tables and figures in the text are located too far from the tables and figures themselves, which makes it difficult to read. I propose to supplement the text with explanations of the tables and figures immediately before them.
Response: We have modified the positions of tables and figures accordingly.
- I did not understand why the authors used bold and underline for some values in tables 3 and 4.
Response: We have modified the table captions to include this.
- The use of links instead of method names in Figures 5 and 6 makes it difficult to read. I propose to supplement the links with the names of the methods.
Response: Thank you for the comments. Most of the proposed methods used deep learning backbones to in their implementation. But they did not specify a name for their methods. We have modified Figures 5 and 6 by adding the reference number and the name of the authors.
Round 2
Reviewer 1 Report
Dear authors, thank you for improving the article.
There is only one thing left. According to my previous remark number 3. OCT images are different from ones from fundus camera, that is obvious. The matter is to point that fundus camera is not only way for diabetic retinopathy classification. Please add relevant paragraph.
Author Response
Reviewer comments:
There is only one thing left. According to my previous remark number 3. OCT images are different from ones from fundus camera, that is obvious. The matter is to point that fundus camera is not only way for diabetic retinopathy classification. Please add relevant paragraph.
Thank you for reviewing our paper. As suggested we added a paragraph in page 2 about OCT images.
Reviewer 2 Report
After the changes performed by the authors, the article is ready for its publication.
Author Response
Reviewer comments:
After the changes performed by the authors, the article is ready for its publication.
Thank you for reviewing our paper.
