Automatic Classification of the Severity of Knee Osteoarthritis Using Enhanced Image Sharpening and CNN

Round 1

Reviewer 1 Report

 

1.      The aim and contributions of the proposed technique are not stated in the abstract.

2.      Authors wrote that “we achieved a mean accuracy of 91.03 percent, an improvement of 19.03 percent over the earlier accuracy of 72 percent by using the original knee x-ray images for the detection of OA with five grading”. I have never seen such improvement of 19.03% only with image preprocessing. The mathematical background of used preprocessing algorithms is not explained.

3.      The dataset is imbalance between different KL grading (i.e., 5 classes). Therefore data augmentation is required to increase the size of dataset and make it balanced. It can also reduce the over-fitting problem.  

4.      Related work’s survey must have its own section including the research gap and a table that shows the techniques, dataset, and findings (accuracy). Besides, give a general diagram of the proposed technique in section introduction.

5.      Put a label for each type in figure 1 and 3 before and after processing.

6.      In figure 2, system architecture indicates that segmentation and contour detection have applied before image enhancement. I did not see any flowchart or algorithm for segmentation and contour detection.

7.      The cross validation for splitting data is not faire 90% for training and 10% for testing. Try 80:20 and 70:30.

8.      Give the accuracy and loss curves before and after processing.

9.  Why ResNet based-CNN has used? Why not DesnseNet?, which is more efficient and accurate.  

10.  The deep learning used model in this work is not well described. Do some tuning of hyper-parameters such as (batch size, epochs, learning rate, layers variation and so on).

11.  The confusion matrix in figure 7 contains 176 images, which is 10% of the whole dataset. Is it correct? Same for figure 8. Please verify both.

12.  More recent papers must add to the reference section.

13.  Did you consider the accuracy comparison of your proposed model and others?

Comments for author File: Comments.pdf

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

General comments

The article Automatic Classification of the Severity of Knee Osteoarthritis using Enhanced Image Sharpening and CNN  is well written and well organized. The research topic is original and addresses an important issue involving the diagnosis of osteoarthritis of the knee using machine learning. The study at hand compares the correlations of unaided ratings with clinical severity  against correlations of AI aided ratings with clinical severity. The methodology of the adopted method was described in a concise and clear manner. The results have been discussed in a correct manner allowing for easy interpretation.  The discussion presented in the paper seems to be too short and does not relate the results obtained by the authors to the results obtained in papers on similar topics. Conclusions are presented in a concise manner and are related to the results obtained. The list of references contains only 25 items which is far too small a number for a scientific paper. The paper requires editorial correction and adaptation to the requirements of the journal in terms of formatting and citation.  In my opinion, it seems necessary to make minor corrections to allow a more thorough understanding of the topic. The following are my comments.

Minor comments:

Introduction

The incidence of osteoarthritis is influenced by many factors, such as work, sports participation, musculoskeletal injuries, obesity and gender. Information about this, along with the necessary literature, should be added to the first paragraph of the introduction. Authors may find some useful information in the works: DOI 10.1016/S0140-6736(19)30417-9; DOI10.3390/app11041552; https://doi.org/10.1136/annrheumdis-2013-204763; DOI 10.3390/app10238312;

The introduction should be expanded to include short information on typical diagnostic methods (MRI, CT, X-ray, ultrasound) including physical examination, as well as alternative methods such as vibroartrography or phonoartrography. Authors may find some useful information in the works: https://doi.org/10.1016/j.cpet.2018.08.004; https://doi.org/10.1111/j.1617-0830.2006.00063.x; DOI 10.3390/app9194102; https://doi.org/10.1016/j.berh.2016.09.007; doi: 10.35784/acs-2022-14; https://doi.org/10.3390/s22062176; https://doi.org/10.3390/s22103765;

Please increase the emphasis on the essence and purpose of the research conducted and the potential of the results obtained.

Please make the relevant additions with the necessary literature. This will allow you to better understand the topic and highlight the essence of the issue at hand.

Materials and Methods

Subsection 2.6 Evaluation Parameters should be expanded to include more detailed descriptions of the parameters used to compare classifiers. Please add the necessary descriptions with appropriate sources. Results

Results and Discussion

The discussion should be significantly expanded with a detailed reference of the results obtained to the results of other authors in the works of which a similar method was proposed. A detailed summary and comparison of the results in a table would help to emphasize the importance of the present work. It would also be worthwhile to relate the results obtained to the results of diagnostic accuracy obtained in other methods such as physical examination, MRI, CT, X-ray or vibroarthrography. It seems reasonable to add information on the limitations of the proposed method and the authors' plans for further research and potential directions for the development of machine learning-based diagnostics.

The following is a list of publications that you may find useful and may be a valuable addition to the information presented so far in the discussion.

https://doi.org/10.3390/s22062176;

 https://doi.org/10.3390/s22103765;

https://doi.org/10.1007/s10439-022-02913-4

https://doi.org/10.1038/s41598-022-07092-9

https://doi.org/10.1016/j.jbspin.2022.105370

https://doi.org/10.1007/s11547-022-01476-7

https://doi.org/10.1016/j.cmpb.2022.106963

After making the appropriate additions, the article may be accepted for publication. The results obtained can find practical application in daily clinical practice.

 

 

 

 

 

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

1. Related work’s survey must have its own section including the research gap and a table that shows the techniques, dataset, and findings (accuracy). Besides, give a general diagram of the proposed technique in section introduction.

2.  Give the accuracy and loss curves before and after processing.

Author Response

Please see the attachment

Author Response File: Author Response.pdf