A Federated Learning Framework for Breast Cancer Histopathological Image Classification

Round 1

Reviewer 1 Report

The research paper proposes a framework for a centralized and secure system to train and test distributed datasets on medical imaging. Models on medical dataset may work better on a specific dataset, however training and testing these models on external datasets is a challenge due to security and privacy concerns. This research tries to solve this problem by providing a federated system where models can be uploaded, trained and tested using secure servers. The framework was tested using 4 different models on BreaKHis dataset and performance was measured using F1, Kappa and other metrics. Overall the paper is well organized. The experiment and results section describe the framework  adequately. One area where the paper can be improved is the details on the performance of the system in terms of processing time, response times, information about data store, susceptibility to security attacks. These are not serious flaws though and the paper does provide insight into an interesting topic. As a result the review recommends a decision of Accept.

Author Response

Dear Reviewer,

Thank you very much for reviewing our manuscript. We have carefully read the thoughtful comments from you and found that the suggestions are helpful for us to improve our manuscript.

 

One area where the paper can be improved is the details on the performance of the system in terms of processing time, response times, information about data store, susceptibility to security attacks. 

Response: We appreciate and agree with your suggestion. We tried to organize above information into the manuscript. However, our experiment did not measure these aspects scientifically in the pilot phase. We have added this point in the conclusion section as future work, it is very important for the next indeed. Many thanks! 

Reviewer 2 Report

Kindly allow me to congratulate the author(s) for the work done. It seems to me a relevant, interesting and very necessary work. However, for its better visibility and readability, I suggested some comments as below:

In abstract add the results of this study briefly in one statement.

In introduction there are missing references, it need to enrich the introduction with more related quotation. many statement without references? 

All acronyms should have their first letter capitalized, at least the first time they appear, and state the statement first then followed by abbreviation in bracket (in the introduction and throughout the text).

The Results section is good and acceptable 

Lack limitations and nd are needed to identify the future study recommendations for the research community correctly.

It is necessary to review the REFERENCES, to adjust them to the journal rules, and try your best to include 3-5 current and recent references, from 2020 to 2022, on the same subject. Almost all references before 2020.

Author Response

Dear Reviewer, 

We have studied the valuable comments from you, and tried our best to revise the manuscript.  The point to point responds to the comments are listed as following:

 

In abstract add the results of this study briefly in one statement.

Response: Thank you for your suggestion. We have added a brief statement of the results in line 9.

 

In introduction there are missing references, it need to enrich the introduction with more related quotation. many statement without references? 

Response: Thank you for your careful review. In introduction section, we have added reference No. 14,15,16,17,18,19,20 to support the related statements. 

 

All acronyms should have their first letter capitalized, at least the first time they appear, and state the statement first then followed by abbreviation in bracket (in the introduction and throughout the text).

Response: Many thanks for your careful reading. All acronyms have been revised.

 

The Results section is good and acceptable 

Response: Thanks for your approval. 

 

Lack limitations and nd are needed to identify the future study recommendations for the research community correctly.

Response: Thanks a lot for the comment. We have rewritten the last paragraph to present current limitation and future recommendations. 

 

It is necessary to review the REFERENCES, to adjust them to the journal rules, and try your best to include 3-5 current and recent references, from 2020 to 2022, on the same subject. Almost all references before 2020.

Response: Thank you for your careful work. We have updated the references with 9 recent works from 2020 to 2022. 

Reviewer 3 Report

The manuscript presents a study conducted for establishing a federated learning framework for medical image diagnosis. The proposed method is an alternative to overcome the difficulties in providing the datasets from multiple institutions as well as the issues regarding the patient privacy and data confidentiality regulations. The developed method was applied for the dataset of breast cancer histopathological images (BreaKHis) in a classification task as the practical case. Four deep learning models including ResNet-152, DenseNet-20, MobileNet-v2-100, and EfficientNet-b7 were used to diagnosis of the images. And five evaluation metrics namely F1 measure, diagnostic odds ratio, test accuracy at image level, test accuracy at patient level, and Kappa criteria were conducted on the test dataset to assess the performance of the method.

I should say that the manuscript is really well organized, the presentation is very good, and the English writing is OK as it is easily readable. Also, all the adopted methodologies with regards to the model development and the evaluation metrics are alright, and the results are clearly presented.

Therefore, to me, the manuscript is accepted in the present format.

Author Response

Dear Reviewer, 

Thank you for the affirmation and encouragement.  We have read the whole manuscript once again and made any necessary editorial corrections. We take this opportunity thank you again for your review.