Deep Learning Models for COVID-19 Detection

Round 1

Reviewer 1 Report

I would suggest to remove "in Smart Cities" part from the title and first three sentences from the abstract since, although related, there is no contribution to this area.  Thus, title would be more accurate.

In few different parts of the paper it is mentioned that data augmentations are applied. I believe elaborating this can be beneficial. i.e. which data augmentations are applied.  

Section 3.2. "Figure 3." was supposed to be "Figure 2."?

Section 3.6. "The single CNN model provides probability values for COVID-19 and non-COVID-19 CT scans."

Although, outputs of the Softmax adds up to 1, I believe its output can not be interpreted as probability. Thus, authors should either back up this claim either by experiment or by citing the relevant paper/experiment or remove/modify this sentence.

 

 

Author Response

I would suggest to remove "in Smart Cities" part from the title and first three sentences from the abstract since, although related, there is no contribution to this area.  Thus, title would be more accurate.

• We removed.

In few different parts of the paper it is mentioned that data augmentations are applied. I believe elaborating this can be beneficial. i.e. which data augmentations are applied.

• We applied image rotation technique to augment traiining set. We also added this explanation in Section 3.3.

Section 3.2. "Figure 3." was supposed to be "Figure 2."?

• We changed Figure 3 to Figure 2.

Section 3.6. "The single CNN model provides probability values for COVID-19 and non-COVID-19 CT scans." Although, outputs of the Softmax adds up to 1, I believe its output can not be interpreted as probability. Thus, authors should either back up this claim either by experiment or by citing the relevant paper/experiment or remove/modify this sentence.

• We removed the sentence.

Author Response File: Author Response.pdf

Reviewer 2 Report

In this manuscript, the authors developed a deep learning model for COVID-19 detection in smart cities. It is important and also necessary topics, which has potential application. The study was preformed well. A couple of issues should be addressed:

In abstract, the last sentence should be clarified.

Many techniques, including RT-PCR, have been widely developed and used in COVID-19 screening and detection, there is a recent review summarizing the major process on COVID-19 detection and vaccine treatment, which entitled “Biotechnological Perspectives to Combat the COVID-19 Pandemic: Precise Diagnostics and Inevitable Vaccine Paradigms” Cells 2022, 11(7), 1182; https://doi.org/10.3390/cells11071182. The author should read this review and cite it in the introduction.

In the conclusion, “Table ?? lists the accuracy results of the proposed CNN networks”?

Author Response

In this manuscript, the authors developed a deep learning model for COVID-19 detection in smart cities. It is important and also necessary topics, which has potential application. The study was preformed well. A couple of issues should be addressed:

In abstract, the last sentence should be clarified.

• We clarified the sentence.

Many techniques, including RT-PCR, have been widely developed and used in COVID-19 screening and detection, there is a recent review summarizing the major process on COVID-19 detection and vaccine treatment, which entitled “Biotechnological Perspectives to Combat the COVID-19 Pandemic: Precise Diagnostics and Inevitable Vaccine Paradigms”Cells2022,11(7), 1182;https://doi.org/10.3390/cells11071182. The author should read this review and cite it in the introduction.

• We cited the review in the introduction section.

In the conclusion, “Table ?? lists the accuracy results of the proposed CNN networks”?

• We corrected the Table 3 in the conclusion section.

Author Response File: Author Response.pdf

Reviewer 3 Report

The authors proposes a novel data-efficient deep network for the identification of COVID-19 on CT images by using both synthetic and augmented data,  claiming higher performance than classic deep learning models for COVID-19 detection. Although the topic is relevant to the current pandemic moment, the manuscript fails to provide some methodological details and properly discuss the findings, besides requiring improvement of English language as listed below in the order of appearance in the text:

> PAGE 1
- In the sentence "Previous versions of this virus identified include SARS and MERS", note that SARS and MERS are names of diseases, not their etiological agents (SARS-CoV and MERS-CoV, respectively) - anyway, the etiological agent of COVID-19 (i.e., SARS-CoV-2) is more closely related to SARS-CoV than to MERS-CoV.
- Replace "COVID-19" with "SARS-CoV-2" in the sentence "X-ray images show the COVID-19 infection areas on the human lungs".
- Correct "modelling" to "modeling" in the sentence "Convolutional neural networks (CNNs) are known as powerful models for image modelling".

> PAGE 2
- Adjust "Covid-19" to "COVID-19" in the sentences "inadequate data might hamper the usage of an artificial intelligence-based model for Covid-19 detection", "data-efficient CNN models are built on small sets of available images and allow fast modelling of the disease to diagnose Covid-19", "data-efficient models might make a significant contribution to the rapid diagnosis of Covid-19 during a pandemic" and "A method is proposed for fusing synthetic and augmented CT scans for generating enhanced CNN models for Covid-19 detection".
- Replace "COVID-19 virus" with "chests of COVID-19 patients in the sentence "The proposed model builds on augmented and synthetic CT images of the COVID-19 virus".
- Since it is still the introduction section, delete the result-related sentence "The improved CNN performance over classic CNN models is reported in Table 3".

> PAGE 3
- All effectively used images from the COVID19-CT and Mosmed datasets must be identified for reproducibility purposes.
- The method section must be improved for clarity and to contain some missing resources such as the softwares used for performance evaluation.

> PAGE 4
- Adjust "Covid-19" to "COVID-19" in the panel labels of Figures 2 and 3.
- All components of the optimization function must be described, not only G(x) and D(z).

> PAGE 5
- Delete "COVID-19 Prediction" after the sentence "Fine-tuning is achieved by freezing all convolutional layers and adapting the last fully connected layer for COVID-19 and non-COVID-19 classification".
- In the equation identified by the number 3, correct "Sensetivity" to "Sensitivity".
- Replace "if" with "of" in the sentence "Table 3 reports the performances if the AlexNet, VGG, ResNet-18, ResNet-50, MobileNetV2, and DensNet-121 deep learning models".

> PAGE 6
- In Figure 4, adjust the legends of the four panels (a, b, c and d) in order to show the CNN models in the same order and turn the line colors of the last three more distinctive to avoid confusion.
- To avoid repetition, the subsection 4.3 (Comparison of the Proposed Methods) could be integrated with the two previous subsections (i.e., 4.1 and 4.2). 

> PAGE 7
- Replace "mode" with "model" in the sentence "Furthermore, the MobileNetV2 model built on augmented and synthetic data outperformed MobileNetV2 mode".
- Replace "build on both" with "only build on" in the sentence "Figure 4 (c) the ROC curves of CNN models, which build on both augmented CT Scans of the Mosmed dataset".
- Replace "COVID19-CT" with "Mosmed" in the sentence "Figure 4 (d) the ROC curves of CNN models, which build on both augmented and synthetic CT Scans of the COVID19-CT dataset". 
- The discussion section must be rewritten to actually discuss the data (e.g., by comparing the results with those from other studies available in the scientific literature using similar techniques), as they were just recapped.
- Delete the sentences "The method created several CNN networks" and "Table ?? lists the accuracy results of the proposed CNN networks" from the conclusion section.

Author Response

The authors proposes a novel data-efficient deep network for the identification of COVID-19 on CT images by using both synthetic and augmented data, claiming higher performance than classic deep learning models for COVID-19 detection. Although the topic is relevant to the current pandemic moment, the manuscript fails to provide some methodological details and properly discuss the findings, besides requiring improvement of English language as listed below in the order of appearance in the text:

> PAGE 1
- In the sentence "Previous versions of this virus identified include SARS and MERS", note that SARS and MERS are names of diseases, not their etiological agents (SARS-CoV and MERS-CoV, respectively) - anyway, the etiological agent of COVID-19 (i.e., SARS-CoV-2) is more closely related to SARS-CoV than to MERS-CoV.
- Replace "COVID-19" with "SARS-CoV-2" in the sentence "X-ray images show the COVID-19 infection areas on the human lungs".

• We replaced COVID-19 with SARS-CoV-2.

- Correct "modelling" to "modeling" in the sentence "Convolutional neural networks (CNNs) are known as powerful models for image modelling".

• We correct modelling to modeling.

> PAGE 2
- Adjust "Covid-19" to "COVID-19" in the sentences "inadequate data might hamper the usage of an artificial intelligence-based model for Covid-19 detection", "data-efficient CNN models are built on small sets of available images and allow fast modelling of the disease to diagnose Covid-19", "data-efficient models might make a significant contribution to the rapid diagnosis of Covid-19 during a pandemic" and "A method is proposed for fusing synthetic and augmented CT scans for generating enhanced CNN models for Covid-19 detection".

• We corrected all Covid-19 to COVID-19.

- Replace "COVID-19 virus" with "chests of COVID-19 patients in the sentence "The proposed model builds on augmented and synthetic CT images of the COVID-19 virus".

• We repalced COVID-19 virus with chests of COVID-19 patients.

- Since it is still the introduction section, delete the result-related sentence "The improved CNN performance over classic CNN models is reported in Table 3".

• We deleted the sentence.

> PAGE 3
- All effectively used images from the COVID19-CT and Mosmed datasets must be identified for reproducibility purposes.
- The method section must be improved for clarity and to contain some missing resources such as the softwares used for performance evaluation.

• We added 3.8 Softwave section and explained the softwave library.

> PAGE 4
- Adjust "Covid-19" to "COVID-19" in the panel labels of Figures 2 and 3.

• We corrected all Covid-19 to COVID-19.

- All components of the optimization function must be described, not only G(x) and D(z).

• We described.

> PAGE 5
- Delete "COVID-19 Prediction" after the sentence "Fine-tuning is achieved by freezing all convolutional layers and adapting the last fully connected layer for COVID-19 and non-COVID-19 classification".

• We deleted.

- In the equation identified by the number 3, correct "Sensetivity" to "Sensitivity".

• We corrected.

- Replace "if" with "of" in the sentence "Table 3 reports the performances if the AlexNet, VGG, ResNet-18, ResNet-50, MobileNetV2, and DensNet-121 deep learning models".

• We corrected.

 

 

> PAGE 6
- In Figure 4, adjust the legends of the four panels (a, b, c and d) in order to show the CNN models in the same order and turn the line colors of the last three more distinctive to avoid confusion.

- To avoid repetition, the subsection 4.3 (Comparison of the Proposed Methods) could be integrated with the two previous subsections (i.e., 4.1 and 4.2).

• We integrated section 4.3 to Section 4.2

> PAGE 7
- Replace "mode" with "model" in the sentence "Furthermore, the MobileNetV2 model built on augmented and synthetic data outperformed MobileNetV2 mode".

• We replaced mode with model.

- Replace "build on both" with "only build on" in the sentence "Figure 4 (c) the ROC curves of CNN models, which build on both augmented CT Scans of the Mosmed dataset".

• We replaced.

- Replace "COVID19-CT" with "Mosmed" in the sentence "Figure 4 (d) the ROC curves of CNN models, which build on both augmented and synthetic CT Scans of the COVID19-CT dataset".

• We replaced.

- The discussion section must be rewritten to actually discuss the data (e.g., by comparing the results with those from other studies available in the scientific literature using similar techniques), as they were just recapped.

• We add first paragraph in the discussion section. We compare results with the similay work.

- Delete the sentences "The method created several CNN networks" and "Table ?? lists the accuracy results of the proposed CNN networks" from the conclusion section.

• We deleted the sentences.

Author Response File: Author Response.pdf

Reviewer 4 Report

I have to state that I have a positive opinion about this study in consequence of its importance. However, I would like to draw the attention of the authors to the important points that need to be corrected in the article. Before acceptance, the following points must be incorporated.

1. The abstract needs to be more technical.
2. Add also latest applications of computational methods in the healthcare domain. Consult the following papers:

A bibliometric analysis of corona pandemic in social sciences: A review of influential aspects and conceptual structure

A proficient approach to forecast COVID-19 spread via optimized dynamic machine learning models

Disease diagnosis system using IoT empowered with fuzzy inference system

COVID-19 detection based on lung CT scan using deep learning techniques

Classification of Human Face: Asian and Non-Asian People

3. The authors justify and explain the evaluation criteria.
4. The results section needs to be enhanced in the context of tables and figures explanation.
5. The discussion and conclusion are too short it needs improvement.
6. The authors justify and explain why their model achieved such high results.
7. Summarize the experimental results with some numeric values in the discussion sufficiently.
8. A comparison of previous research must be done as in table 1 of referred paper: “Corporate Bankruptcy Prediction: An Approach Towards Better Corporate World”.
9. It is better to summarize the major findings of the paper within one-two sentence without experimental results.

Overall speaking, the innovation points and main contributions of this paper need to be carefully reconsidered, and the innovation points should be presented more clearly and prominent in terms of word expression and methodology & experiment design.

 

Author Response

I have to state that I have a positive opinion about this study in consequence of its importance. However, I would like to draw the attention of the authors to the important points that need to be corrected in the article. Before acceptance, the following points must be incorporated.

1. The abstract needs to be more technical.

• We add a technical explanation of the methodology in the abstract section. (Then, we estimate parameters of convolutional and fully connected layers of the deep networks using synthetic and augmented data.)

2. Add also latest applications of computational methods in the healthcare domain. Consult the following papers:

A bibliometric analysis of corona pandemic in social sciences: A review of influential aspects and conceptual structure

A proficient approach to forecast COVID-19 spread via optimized dynamic machine learning models

Disease diagnosis system using IoT empowered with fuzzy inference system

COVID-19 detection based on lung CT scan using deep learning techniques

Classification of Human Face: Asian and Non-Asian People

• We added all papers.

3. The authors justify and explain the evaluation criteria.

• We use accuracy, sensitivity and specificity and area under the curce for evaluation.

4. The results section needs to be enhanced in the context of tables and figures explanation.

• We add explanations in the fist paragraph of the disscution section.

 

5.The discussion and conclusion are too short it needs improvement.

• We add one paragraph (fisrt paragraph) to disscution section.

• we add one paragraph (fisrt paragraph) to conclusion section.

6 The authors justify and explain why their model achieved such high results.

• We achieved accurate models by using syntetic data. Syntetic data represents all posible casses of CVOID-19 virus on CT images.

7 Summarize the experimental results with some numeric values in the discussion sufficiently.

• We summarized results and compare other work in the fist paragraph of the discussion section.

8 A comparison of previous research must be done as in table 1 of referred paper: “Corporate Bankruptcy Prediction: An Approach Towards Better Corporate World”.

9 It is better to summarize the major findings of the paper within one-two sentence without experimental results.

• We found that we can generate accurate deep networks using limited data. We achieved this using syntetic and augmentation techniques.

 

 

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

The revised version of manuscript entitled "Deep Learning Models for Covid-19 Detection" was improved by the authors, but only partially considered my requests on its original version. Some issues - once again described below in order of appearance in the manuscript - still need to be addressed by the authors:

> PAGE 1
- In the sentence "Previous versions of this virus identified include SARS and MERS", note that SARS and MERS are names of diseases, not their etiological agents (SARS-CoV and MERS-CoV, respectively) - anyway, the etiological agent of COVID-19 (i.e., SARS-CoV-2) is more closely related to SARS-CoV than to MERS-CoV.

> PAGE 3
- All effectively used images from the COVID19-CT and Mosmed datasets must be identified for reproducibility purposes.

> PAGE 4
- Description of components of the optimization function requires revision for proper correspondence.

> PAGE 6
- In Figure 4, adjust the legends of the four panels (a, b, c and d) in order to show the CNN models in the same order and turn the line colors of the last three more distinctive to avoid confusion.
- To avoid repetition, the subsection 4.2 (Proposed Data-Efficient Method) could be integrated with the subsection 4.1 (Classic Deep Learning Method) as a single subsection entitled "Comparison between Classic Deep Learning Method and Proposed Data-Efficient Method". 

> PAGE 7
- The discussion section must be improved to discuss the data by comparing the results with those from multiple studies available in the scientific literature using similar techniques, since only one of them was explored.

Author Response

The authors proposes a novel data-efficient deep network for the identification of COVID-19 on CT images by using both synthetic and augmented data, claiming higher performance than classic deep learning models for COVID-19 detection. Although the topic is relevant to the current pandemic moment, the manuscript fails to provide some methodological details and properly discuss the findings, besides requiring improvement of English language as listed below in the order of appearance in the text:

> PAGE 1
- In the sentence "Previous versions of this virus identified include SARS and MERS", note that SARS and MERS are names of diseases, not their etiological agents (SARS-CoV and MERS-CoV, respectively) - anyway, the etiological agent of COVID-19 (i.e., SARS-CoV-2) is more closely related to SARS-CoV than to MERS-CoV.

• We corrected the sentense.

> PAGE 3
- All effectively used images from the COVID19-CT and Mosmed datasets must be identified for reproducibility purposes.

• We could not understand this comment. Can you be more clear please?

> PAGE 4
- Description of components of the optimization function requires revision for proper correspondence.

• We added section 3.7 and explained optimization values.

> PAGE 6
- In Figure 4, adjust the legends of the four panels (a, b, c and d) in order to show the CNN models in the same order and turn the line colors of the last three more distinctive to avoid confusion.

• We removed the Figure.

- To avoid repetition, the subsection 4.2 (Proposed Data-Efficient Method) could be integrated with the subsection 4.1 (Classic Deep Learning Method) as a single subsection entitled "Comparison between Classic Deep Learning Method and Proposed Data-Efficient Method".

• We corrected.

> PAGE 7
- The discussion section must be improved to discuss the data by comparing the results with those from multiple studies available in the scientific literature using similar techniques, since only one of them was explored.

• We added Table 4 and we compare the proposed method with two other studies.