Computer-Aided Detection of Hypertensive Retinopathy Using Depth-Wise Separable CNN
, Qaisar Abbas 3, Junhua Yan 2, Ayyaz Hussain 4, Kashif Shaheed 5 and Abdul Rauf Baig 3,*Round 1
Reviewer 1 Report
Authors have proposed computer-aided systems to advance depth-wise separable convolutional neural network with residual connection and a linear support vector machine.
1- In the abstract section, I would suggest that the author should provide the point and quantitative advantages of classification results.
2- The main contributions of this paper should be further summarized and clearly demonstrated.
3- Some new references should be added to improve the literature review—for example, https://doi.org/10.3390/app12115500; https://doi.org/10.1117/1.JMI.7.3.034501; https://doi.org/10.48550/arXiv.1912.09621.
Author Response
Original Manuscript ID: ID: applsci-2002519
Original Article Title: Computer-aided Detection of Hypertensive Retinopathy Using Depthwise Separable CNN
To: Editor in Chief,
MDPI, Applied Sciences
Re: Response to reviewers
Dear Editor,
Many thanks for insightful comments and suggestions of the referees. Thank you for allowing a resubmission of our manuscript, with an opportunity to address the reviewers’ comments.
We are uploading (a) our point-by-point response to the comments (below) (response to reviewers), (b) an updated manuscript with yellow highlighting indicating changes, and (c) a clean updated manuscript without highlights (PDF main document).
By following reviewers’ comments, we made substantial modifications in our paper to improve its clarity and readability. In our revised paper, we represent the improved manuscript.
We have made the following modifications as desired by the reviewers:
Best regards,
Corresponding Author,
Dr. Abdul Rauf Baig (On behalf of authors),
Professor.
Reviewer 1:
Comment - (1) In the abstract section, I would suggest that the author should provide the point and quantitative advantages of classification results.
Response 1: As suggested by reviewer #1, we have the abstract as:
“
On average, the CAD-HR achieved sensitivity (SE) of 94%, specificity (SP) of 96%, accuracy (ACC) of 95% and area under the receiver operating curve (AUC) of 0.96.
“
Those changes can be easily seen in the revised paper. Thank you to clear this problem in our paper.
Comment - (2) The main contributions of this paper should be further summarized and clearly demonstrated.
Response 2: As suggested by reviewer #1, we have add more emphasis on the main contributions to this paper as:
“
- This paper develops a new CAD system to recognize HR based on a residual con-nection and depth-wise separable convolution neural network (DSC) with linear support vector machine (LSVM) to address the problems of limited dataset, high computational complexity, and the lack of lightweight and efficient feature descriptors.
- To effectively extract features from HR and non-HR photos, a pre-trained CNN model that is lightweight and based on a depth-wise separable convolution neural network is developed. Our utilization of a depth-wise separable convolution neural network for feature extraction in a HR classification challenge is a first to our knowledge. To deal with a real-time context, the proposed feature extraction is more extensive.
- For automatic HR classification, we employ a 75%-25% train-test split utilizing the linear SVM machine learning classifier. The efficiency and performance of Linear SVM make it a popular choice, especially when working with limited data sets.
- Extensive experiments are conducted by employing several statistical metrics on two publicly available and one proprietary benchmarks, namely DRIVE, Di-aRetDB0, and Imam-HR. A full comparative study comparing the suggested strategy to other existing DL approaches is presented.
- The proposed CAD-HR system is outperformed compared to other transfer learning (TL) based architectures to recognize HR.
”
Thank you to clear this problem in our paper.
Comment - (3) Some new references should be added to improve the literature review—for example, https://doi.org/10.3390/app12115500; https://doi.org/10.1117/1.JMI.7.3.034501; https://doi.org/10.48550/arXiv.1912.09621..
Response 3: It is difficult to add those references because they are not related to our topic and all required references have been already incorporated to the paper.
Reviewer 2 Report
Computer-aided Detection of Hypertensive Retinopathy Using Depthwise Separable and residual-based CNN
- In line 3, revise “residual-based”.
- Readdress the in-text citations, for example, in lines 7, 3, 8, 10, etc.
- Revise the multiple uses of “hypertensive retinopathy (HR)”. Similarly, some other acronyms and their full versions.
- In line 94, double-check the acronym (DMLs),
- In line 105, define the acronym (DSC-LSVM) before using it.
- Revise the sentence in lines 150 and 151.
- Revise the structure connection in lines 152 and 153.
- The paragraph from line 140 through line 190 is excessively long; you should divide it.
- Revise the sentence in lines 174 and 175.
- In lines 181 and 182, “... the results were correct.” How and based on what metrics?
- In line 214, “They discovered the much higher accuracy in that study.” The sentence is ambiguous, consider revising it.
- In Figure 2, remove the swirly line under Retinograph.
- In line 424, “Figure 9b”, I believe you meant 6b. Accordingly, add a and b to Figure 6 and include them in the caption as well.
- In line 633, revise “Figure 18”.
- In line 678, revise “…has been chosen…”
- In line 679, revise “Table 8”.
- In line 684, revise “Dl”.
- In line 701, revise “conversely,…”
- In line 707, revise “Table 9”.
- Revise the sentence in lines 764 and 765.
- In line 782, which figure do you mean by “Figure 20”?
- In line 785, revise “Tables 7 and 8”.
General comments:
1. In sub-section 4.2, the provided cost calculation is machine-dependent. I would recommend generalizing the computation or elaborating on the calculation details for future comparisons.
2. Revise the presentation and contents of algorithms 1 and 2. They should be formalized in a more appropriate way for algorithm description.
3. Apart from time complexity, what are the limitations of your work? Kindly, address them in the discussion.
4. There are several state-of-the-art pre-trained DL algorithms such as VGG16 & 19, ResNet50, AlexNet, etc. can be also used for comparison purposes. The main reason for suggesting such DL models is that they have been trained on a huge number of images from various databases and classes. Therefore, for robust validation, please expand your experimental work to include a few of them in your assessment.
5. The similarity index of the study is 23%, although the quotes and bibliography are excluded. (According to the iThenticate)
Comments for author File: 
Comments.pdf
Author Response
Reviewer #2:
Comment - (1) - In line 3, revise “residual-based”.
- Readdress the in-text citations, for example, in lines 7, 3, 8, 10, etc.
- Revise the multiple uses of “hypertensive retinopathy (HR)”. Similarly, some other acronyms and their full versions.
- In line 94, double-check the acronym (DMLs),
- In line 105, define the acronym (DSC-LSVM) before using it.
- Revise the sentence in lines 150 and 151.
- Revise the structure connection in lines 152 and 153.
- The paragraph from line 140 through line 190 is excessively long; you should divide it.
- Revise the sentence in lines 174 and 175.
- In lines 181 and 182, “... the results were correct.” How and based on what metrics?
- In line 214, “They discovered the much higher accuracy in that study.” The sentence is ambiguous, consider revising it.
- In Figure 2, remove the swirly line under Retinograph.
- In line 424, “Figure 9b”, I believe you meant 6b. Accordingly, add a and b to Figure 6 and include them in the caption as well.
- In line 633, revise “Figure 18”.
- In line 678, revise “…has been chosen…”
- In line 679, revise “Table 8”.
- In line 684, revise “Dl”.
- In line 701, revise “conversely,…”
- In line 707, revise “Table 9”.
- Revise the sentence in lines 764 and 765.
- In line 782, which figure do you mean by “Figure 20”?
- In line 785, revise “Tables 7 and 8”.
Response 1: As suggested by reviewer #2, we have updated all the sentences, which are having grammatical mistakes and typos errors through the paper. Yes, you are right, there were English writing problems in the first version of the paper but, it has been improved in terms of writing. We have used professional software to improve the writing. We have carefully read the whole paper and improve it. You can find these changes in the revised paper through word tracking.
Thank you to clear this problem in our paper.
Comment - (2) In sub-section 4.2, the provided cost calculation is machine-dependent. I would recommend generalizing the computation or elaborating on the calculation details for future comparisons.
Response 2: Yes, you are right. the provided cost calculation is machine-dependent. I would recommend generalizing the computation or elaborating on the calculation details for future comparisons. We have added and calculate the generalized complexity and mentioned in section 4.2 as:
“
This computational time complexity is generalized in terms of the training of the network. In the development of the CAD-HR system, there are four blocks with residual connections that can be represented as i, j, k, and l, with t training examples and n epochs. The result was O (n x t (i + j + k + l)). This time complexity can be reduced by using tensor processing units (TPUs), which are provided by the Google cloud. In practice, the TPUs achieved substantial speedups for DL models and utilized less power. This point of view will be addressed in future work as well.
”
Thank you to point out this viewpoint.
Comment - (3) Revise the presentation and contents of algorithms 1 and 2. They should be formalized in a more appropriate way for algorithm description.
Response 3: Yes, you are right. We have revised the content of algorithm 1 and 2 to show more formalized and appropriate description. Thank you, this valuable comment.
Comment - (4) Apart from time complexity, what are the limitations of your work? Kindly, address them in the discussion.
Response 4: Yes, you are right. We have write new future works in the discussion paragraphs as suggested by you.
“
We have developed this CAD-HR system to recognize only two-stages such as HR and normal. However, the CAD-HR system has not been tested on the five-stages of the HR system. In addition, the optimization of hyper-parameters is required to fine-tune this deep-learning model. Apart from the time complexity, the overall computational time can be decreased by adding a block-based fine-tuning strategy. Also, the preprocessing step can be enhanced in terms of utilizing different color space models. The ConvMixer architecture should also be tested against the proposed model, which should be the one of the future works
”
Thank you, this valuable comment.
Comment - (5) There are several state-of-the-art pre-trained DL algorithms such as VGG16 & 19, ResNet50, AlexNet, etc. can be also used for comparison purposes. The main reason for suggesting such DL models is that they have been trained on a huge number of images from various databases and classes. Therefore, for robust validation, please expand your experimental work to include a few of them in your assessment.
Response 5: Yes, you are right. We have performed some other comparisons in the result section. Those changes can be easily seen but mentioned here in the response letter too as:
Figure 14. Comparison with other transfer learning-based (TL) CNN architectures for classification of HR based on data augmentation and preprocessing steps.
Moreover, we have compared it with other three transfer learning-based (TL) CNN architectures, such as VGG-16, VGG-19, ResNet-50, Inception-v3 and AlexNet. To assess the performance, the preprocessing and data augmentation techniques are first applied to these TL architectures. We employed data augmentation to get around the issue that most CNN architectures need a lot of labeled data for training. To perform comparisons with TL algorithms, we used, by default, hyper-parameters with 200 epochs. A visual example of the performance of TL algorithms is displayed in Figure 14. The obtained results demonstrate that our CAD-HR system is outperformed compared to other state-of-the-art TL-based CNN architectures. On the other hand, in our CAD-HR approach, we intend to deploy more dense separable CNN architectures. Furthermore, merging the deep features of many architectures is an intriguing strategy that can boost performance. The performance has significantly improved when comparing the ResNet50 to other TL models such as VGG-16, VGG-19, Inception-v3 and AlexNet. Therefore, we suggest using ResNet50 to classify HR eye-related diseases. Furthermore, we draw the conclusion that even the ResNet50 model cannot match the performance of the proposed CAD-HR model, which is trained with the optimal hyper-parameter setup. Even though TL-based approaches are often employed to improve classification task accuracy, they also significantly increase the architectural complexity of the model and might not make a meaningful difference in how well deep learning models perform when configured with the best hyper-parameters.
Thank you, this valuable comment.
Comment - (6) . The similarity index of the study is 23%, although the quotes and bibliography are excluded. (According to the iThenticate).
Response 6: Yes, you are right. We have reduced the similarity to 20% and acknowledgement and affiliations are removed then the similarity will be 18%.
Author Response File: 
Author Response.docx
Round 2
Reviewer 2 Report
I would like to thank the author for their prompt and thorough revision based on the provided comments. However, please address the following minor revisions:
1. In line 109, remove hypertensive retinopathy.
2. In line 111, Convolutional Neural Network (CNN).
3. In line 113, replace “convolution neural network” with CNN.
4. In lines 123 and 124, the structure is rather vague. You can replace it with “The proposed CAD-HR system outperforms other transfer learning (TL) based architectures in recognizing HR.”
5. In line 936, what do you exactly mean by “two-stages”? Are you referring to the classification process (HR and non-HR)? If so, then replace “two-stages” with “two classes” or “two categories”. The same goes to “five-stages” in line 937.
Author Response
Dear Reviewer,
Thank you for your valuable comments to enhance the writing of the paper. We have updated the manuscript as suggested by you. You can find these changes in the word tracking options.
Thank you once again.
