Hybrid CLAHE-CNN Deep Neural Networks for Classifying Lung Diseases from X-ray Acquisitions

Round 1

Reviewer 1 Report

Dear authors. Thanks for your manuscript. 

Here are some points to consider:

Authors must review all manuscripts and correct the numbers and references of figures and tables. 

a. Is figure 6 strictly necessary in the manuscript? Because the information described in that it is very general- 

b. In line 211, there isn't information about the number of figures you mention. The same problem is with table numbers 271, 283, and 292. 

c. in line 212, please revise the grammar. There is a point missed in the text. 

d. Please, be clear and consistent with the use of abbreviations and their meaning. In your manuscript, between 160-161 lines, the term CNN meaning is  "Convolution Neural Spectral Networks ."But in other sections, the "CNN" is referenced as a convolutional neural network. 

e. In lines 322, and 323, it is mentioned that "convolutional neural network (CNN) and CLAHE system the result increased to 91%," but those results are not present in the tables you provide in the manuscript. Please revise and update your information. 

Thanks.

Author Response

Authors must review all manuscripts and correct the numbers and references of figures and tables. 

Thank you for your comment.

Response: We checked all the given tables and figures.

1. Is figure 6 strictly necessary in the manuscript? Because the information described in that it is very general- 

Response: Yes, lets keep it because it shows the scheme of the proposed system.

292. In line 211, there isn't information about the number of figures you mention. The same problem is with table numbers 271, 283, and 292. 

Response: Revised

1. in line 212, please revise the grammar. There is a point missed in the text. 

Response: Revised.

1. Please, be clear and consistent with the use of abbreviations and their meaning. In your manuscript, between 160-161 lines, the term CNN meaning is  "Convolution Neural Spectral Networks ."But in other sections, the "CNN" is referenced as a convolutional neural network. 

Response: We revised the given abbreviations in the whole paper.

1. In lines 322, and 323, it is mentioned that "convolutional neural network (CNN) and CLAHE system the result increased to 91%," but those results are not present in the tables you provide in the manuscript. Please revise and update your information. 

Response: The information is revised as given in Table 6.

Reviewer 2 Report

The authors proposed a diagnosis system for lung disease. They proposed CLAHE with machine and deep learning models to solve this problem. The work can improved by the following comments:

1- The proposed deep learning model is trained and validated with only 30 epochs, please discuss.

2- The authors designed the deep learning model. Please clarify why did you choose this architecture?

3- A comparison between the proposed models and the works in the literature is needed.

4- Please discuss the hyperparameters of the proposed models.

5- The paper is needed to be revised by a native speaker.

Author Response

The authors proposed a diagnosis system for lung disease. They proposed CLAHE with machine and deep learning models to solve this problem. The work can improved by the following comments:

1. The proposed deep learning model is trained and validated with only 30 epochs, please discuss.

Thanks very much for the valuable note, which indicates the reviewer’s high scientific standing and his attention to the smallest details.

The following paragraph has been added to the manuscript in section 4.3.

The dataset employed in this research, which comprises 21,164 images, consists of 3,615 COVID images, 6,012 lung opacity images, 1,345 viral pneumonia images, and 10,192 normal images. On 80% of them, the model was trained, while the remaining 20% were used for testing. Training of layers was performed at 2,000 images per step. Training on all classes was run for 500 steps, or 5 epochs, since training of the final layers has been converged for all classes. Testing was performed after every step using the test images, and the best performing model was kept for analysis. After 30 epochs (iterations through the entire dataset) of the model, the training was ended because there had been no further improvements in accuracy.

Furthermore, the following tables have been added in section 4:

• Table 3: Results of Accuracy Classification and Algorithm Evaluation for VGG-19
• Table 4: Results of Accuracy Classification and Algorithm Evaluation for CNN
• Table 5: Comparison of proposed method with other methods reported in the literature

2. The authors designed the deep learning model. Please clarify why did you choose this architecture?

Many thanks for highlighting this point.

Typically, the results obtained from SVM were average 67.2%. Therefore, it is necessary to resort to more effective learning methods for accurate classification. Deep learning algorithm has proven to be a more proficient system compared to SVM algorithm in classifying the predicted classes. However, deep learning networks involve a large number of input data if compared to SVM. The more data that is fed into the network, it will better generalize and correctly make predictions with fewer mistakes. The presence of high-speed computers and high storage capacity to train large neural networks encourage to resort to deep learning.

The above paragraph has been added to the manuscript in section 4.

Additionally, to support the idea quantitatively, we added figure 9 in section 4.1. The figure shows the average performance of SVM.

3. A comparison between the proposed models and the works in the literature is needed.

Many thanks for advising this point, which had a role in demonstrating the distinction of the proposed work.

The following table has been added, which compares the proposed method with the literature.

4. Please discuss the hyperparameters of the proposed models.

Many thanks for highlighting this point.

The following paragraph has been added to the manuscript in section 4.3.

To tune the hyperparameters, the first hyperparameter to tune is the number of neurons in each hidden layer. Then, the rest of the hyperparameters are tuned, which include: optimizer, learning rate, batch size, and epochs. The final step is to tune the number of layers. The accuracy may be affected by various layers. A result with fewer layers may be underfitting, whereas one with too many layers may be overfitting. For the hyperparameter-tuning demonstration, we set the number of neurons in every layer to be the same. The number of neurons ranged from 5 to 100. One of the hyperparameters is the learning rate of the optimizer. The learning rate controls the step size for a model to reach the minimum loss function. A lower learning rate increases the likelihood of discovering a minimum loss function. A higher learning rate makes the model learn quicker. There are 7 optimizers to choose from. We chose the linear optimizer Adam with a learning rate of 0.001. Another hyperparameter is the batch size. Batch size is the amount of training sub-samples for the input, which prevents the model from receiving all of the training data at once. The learning process moves more quickly with the smaller batch size and more slowly in larger batches. The batch size was set to 1000. The last hyperparameter is epochs, which is the number of times a complete dataset is run through the neural network model. That is, one epoch denotes one forward and backward pass of the training dataset through the neural network. In this experiment, the training was ended after 30 epochs of the model because there had been no further improvements in accuracy.

5. The paper is needed to be revised by a native speaker.

We revised the language of the whole paper

Thanks to the reviewer for his keen in the level of language in the research that he is reviewing. Sure, it is done.

Reviewer 3 Report

In this work, the authors proposed a CLAHE and CNN based model to classify X-Ray chest images. The results provided by the proposed model are satisfactory. There are some issues which need to be handled by the authors

1.     Some of the sentences are very large. For example- first sentence of abstract is of more than 4 lines. It needs to be split into two- three sentences.

2.     Many abbreviations are repeated many time. CLAHE has been abbreviated 5-6 times in the paper.

3.     Remove ??. It is also present many times in the paper.

4.      Use word image into place of photo/picture throughout the paper.

5.     Generate experimental results by taking different size of training and testing data.

6.     Discuss your CNN model properly along with parameters tuned.

Author Response

In this work, the authors proposed a CLAHE and CNN based model to classify X-Ray chest images. The results provided by the proposed model are satisfactory. There are some issues which need to be handled by the authors

1. Some of the sentences are very large. For example- first sentence of abstract is of more than 4 lines. It needs to be split into two- three sentences.

Done, with thanks.

2. Many abbreviations are repeated many time. CLAHE has been abbreviated 5-6 times in the paper.  Remove ??. It is also present many times in the paper.

Done. Thank you very much for the careful note

3. Use word image into place of photo/picture throughout the paper.

Done, with thanks.

4.  Generate experimental results by taking different size of training and testing data.

Thanks very much for this note.

In the experiments presented in this manuscript, we followed the state-of-the-art approach, which is called the 80-20 rule. Using this rule, the dataset was divided in the form where the randomly selected 80% of data are used for training while the left over 20% of data are used for testing.

However, the result section has been supported by adding the following:

Figure 9: Comparing the accuracy of different correlation of the four lung diseases using CLAHE and SURF features with SVM learning

Table 3: Results of Accuracy Classification and Algorithm Evaluation for VGG-19

Table 4: Results of Accuracy Classification and Algorithm Evaluation for CNN

Table 5: Comparison of proposed method with other methods reported in the literature

Table 6. Summary of Results Accuracy Classification

5.    Discuss your CNN model properly along with parameters tuned.

Thanks very much for the valuable note, which indicates the reviewer’s high scientific standing and his attention to the smallest details. The same note has been addressed by the first reviewer.

The following paragraphs have been added to the manuscript in section 4.3.

The dataset employed in this research, which comprises 21,164 images, consists of 3,615 COVID images, 6,012 lung opacity images, 1,345 viral pneumonia images, and 10,192 normal images. On 80\% of them, the model was trained, while the remaining 20\% were used for testing.

To tune the hyperparameters, the first hyperparameter to tune is the number of neurons in each hidden layer. Then, the rest of the hyperparameters are tuned, which include: optimizer, learning rate, batch size, epochs, and layers.

For the hyperparameter-tuning demonstration, we set the number of neurons in every layer to be the same. The number of neurons ranged from 5 to 100. One of the hyperparameters is the learning rate of the optimizer. The learning rate controls the step size for a model to reach the minimum loss function. A lower learning rate increases the likelihood of discovering a minimum loss function. A higher learning rate makes the model learn quicker. There are 7 optimizers to choose from. We chose the linear optimizer Adam with a learning rate of 0.001. Another hyperparameter is the batch size. Batch size is the amount of training sub-samples for the input, which prevents the model from receiving all of the training data at once. The learning process moves more quickly with the smaller batch size and more slowly in larger batches. The batch size was set to 1000. The next hyperparameter is epochs, which is the number of times a complete dataset is run through the neural network model. That is, one epoch denotes one forward and backward pass of the training dataset through the neural network. Training of layers was performed at 2,000 images per step. Training on all classes was run for 500 steps, or 5 epochs, since training of the final layers has been converged for all classes. Testing was performed after every step using the test images, and the best performing model was kept for analysis. In this experiment, the training was ended after 30 epochs of the model because there had been no further improvements in accuracy. For the layer hyperparameter, we used five layers. The first layer was convolution (16). The second layer, convolution (32), max pooling and batch normalization. The third layer, convolution (64), max pooling and batch normalization, The fourth layer is convolution (128) and max pooling. Here, we applied dropout (0.2) in the fourth layer to improve the test results. The fifth layer convolution (256), max pooling and dropout (0.2). Finally, we applied flatten, dense (256), dropout (0.15) and again dense.  The accuracy may be affected by various layers. A result with fewer layers may be underfitting, whereas one with too many layers may be overfitting.

Reviewer 4 Report

The paper proposes a hybrid architecture of CLAHE and deep convolutional network for the classification of lung diseases. The experimental results indicate that the suggested hybrid architecture outperforms traditional methods by roughly 20% in terms of accuracy.

The Introduction provides sufficient background and include relevant references. I recommend authors to formulate one goal of their paper and presents the currents goals as its sub-goals.

Authors are reviewed enough in number and up to date literarе sources. the gap in the field.

The methodology is appropriate and very well described. 

Results are clearly presented and visualized through charts and tables. Authors are missed the number of one table in the Discussion section (Line 301: The accuracy results on training set are illustrated in table ??. ).

Conclusions are supported by results.

Author Response

The paper proposes a hybrid architecture of CLAHE and deep convolutional network for the classification of lung diseases. The experimental results indicate that the suggested hybrid architecture outperforms traditional methods by roughly 20% in terms of accuracy.

The Introduction provides sufficient background and include relevant references. I recommend authors to formulate one goal of their paper and presents the currents goals as its sub-goals.

Authors are reviewed enough in number and up to date literarе sources. the gap in the field.

The methodology is appropriate and very well described. 

Results are clearly presented and visualized through charts and tables. Authors are missed the number of one table in the Discussion section (Line 301: The accuracy results on training set are illustrated in table ??. ).

Conclusions are supported by results.

Thank you for the positive comment. We also revised the whole paper to be clearer for the readers.

Round 2

Reviewer 1 Report

Thanks for your improved paper. 

Author Response

Thank you

Reviewer 2 Report

Good job

Author Response

Thank you

Reviewer 3 Report

The paper has been improved a lot. The authors were suggested to take different ratios of training and testing data size which is still missing from the paper. The authors should include some figures or tables by taking 60-40, 70-30, 90-10 and some other combinations of data partitions. It will increase the significance of the research.

Author Response

The paper has been improved a lot. The authors were suggested to take different ratios of training and testing data size which is still missing from the paper. The authors should include some figures or tables by taking 60-40, 70-30, 90-10 and some other combinations of data partitions. It will increase the significance of the research.

Thanks very much for this note.

Figure 12 shows the accuracy of using the CNN network in the dataset, taking different training and validation partition combinations. The results we obtained were as follows: 92%, 91%, 85% and 82% when using partitions 90-10, 80-20, 70-30 and 60-40, respectively. However, we followed the state-of-the-art approach called the 80-20 rule. The results of the proposed method for
each class are clarified in Table 7. Also, a comparison of the proposed method with other methods reported in the literature is presented in Table 8. Table 9 presents the summary of the obtained results. Figure 12. Training and Validation Accuracy for CNN takes a different combination of data partitions.