Federated Learning to Safeguard Patients Data: A Medical Image Retrieval Case

Round 1

Reviewer 1 Report

This paper does not enhance the existing body of knowledge in the given subject area. The purpose of this manuscript is not clear and the content is not consistent with the purpose of the paper.

1. The Introduction section needs to be improved. 
2. The information from the literature has not been appropriately and sufficiently acknowledged in the introduction section.
3. The related work section should be enhanced by discussing more papers related to Federated Learning. 
4. The author should clearly explain the contributions. 
5. The idea of the proposed approach presented in section 4 is not clear. Please explain in a logical sequence by redrawing the workflow (Fig.1). 

Author Response

First, we would like to thank the Editor and the anonymous Reviewers for their helpful comments, which allowed us to improve the quality of our paper. We addressed all the requests and suggestions of the Reviewers. We identified the following comments, together with the changes we did to comply with them.

Reviewer 1

Comment 1: The Introduction section needs to be improved. 

Answer:  

We would like to express our gratitude to the reviewer for his advice. As he suggested, Section 1 has been improved by adding two paragraphs. The first is where further depth has been added to the discussion of the General Data Protection Regulation (DGPR), and the second one is where the paper's drawbacks have been deepened and explained.

Comment 2: The information from the literature has not been appropriately and sufficiently acknowledged in the introduction section.

Answer:

We want to thank the reviewer for pointing this topic out. We introduced in the introduction more citations:

i) Ceroni, A.; Gadiraju, U.K.; Fisichella, M. Improving Event Detection by Automatically Assessing Validity of Event Occurrence in Text. In Proceedings of the Proceedings of the 24th ACM International Conference on Information and Knowledge Management, CIKM 2015, Melbourne, Eds. ACM, 2015, pp. 1815–1818. https://doi.org/10.1145/2806416.2806624. 

ii) Banabilah, S.; Aloqaily, M.; Alsayed, E.; Malik, N.; Jararweh, Y. Federated learning review: Fundamentals, enabling technologies, and future applications. Information Processing & Management 2022, 59, 103061.

iii) Nguyen, D.C.; Ding, M.; Pathirana, P.N.; Seneviratne, A.; Li, J.; Poor, H.V. Federated learning for internet of things: A comprehensive survey. IEEE Communications Surveys & Tutorials 2021, 23, 1622–1658

iv) Nguyen, D.C.; Pham, Q.V.; Pathirana, P.N.; Ding, M.; Seneviratne, A.; Lin, Z.; Dobre, O.; Hwang, W.J. Federated learning for smart healthcare: A survey. ACM Computing Surveys (CSUR) 2022, 55, 1–37.

v) 2018 reform of EU data protection rules.

vi) Pfitzner, B.; Steckhan, N.; Arnrich, B. Federated learning in a medical context: a systematic literature review. ACM Transactions on Internet Technology (TOIT) 2021, 21, 1–31.

Comment 3: The related work section should be enhanced by discussing more papers related to Federated Learning. 

Answer:

We want to thank the reviewer for pointing this topic out. We introduced more citations, such as:

Yang, Q.; Fan, L.; Yu, H. Federated Learning: Privacy and Incentive; Vol. 12500, Springer Nature,2020.

Lim, W.Y.B.; Luong, N.C.; Hoang, D.T.; Jiao, Y.; Liang, Y.C.; Yang, Q.; Niyato, D.; Miao, C. Federated learning in mobile edge networks: A comprehensive survey. IEEE Communications Surveys & Tutorials 2020, 22, 2031–2063.

Mbonihankuye, S.; Nkunzimana, A.; Ndagijimana, A. Healthcare data security technology:HIPAA compliance. Wireless communications and mobile computing 2019, 2019.

Nguyen, D.C.; Pham, Q.V.; Pathirana, P.N.; Ding, M.; Seneviratne, A.; Lin, Z.; Dobre, O.; Hwang,W.J. Federated learning for smart healthcare: A survey. ACM Computing Surveys (CSUR) 2022,55, 1–37.

Comment 4: The author should clearly explain the contributions. 

Answer: 

We would like to express our gratitude to the reviewer for his advice. Section 1 has been improved by adding our contributions in privacy, data quality, scalability, and in the medical domain. 

Comment 5:  The idea of the proposed approach presented in section 4 is not clear. Please explain in a logical sequence by redrawing the workflow (Fig.1). 

Answer:

We would like to express our gratitude to the reviewer for his advice. Section 4 has been improved. In particular, the figure is more descriptive and specifies which arrow refers to local training, classification, local parameter sending, and global model sharing to be better connected with the text.

Reviewer 2 Report

-        The results must include other performance metrics such as confusion matrix and precision for all the scenarios.

-        The dataset partition for testing and validation must be implemented and discussed to evaluate overfitting.

-        The conclusion must include a comprehensive discussion on the interpretability of the federative learning model’s results, drawbacks, and limitations.

Author Response

First, we would like to thank the Editor and the anonymous Reviewers for their helpful comments, which allowed us to improve the quality of our paper. We addressed all the requests and suggestions of the Reviewers. We identified the following comments, together with the changes we did to comply with them.

Reviewer 2

Comment 1:  The results must include other performance metrics such as confusion matrix and precision for all the scenarios.

Answer:

Thanks to reviewer for pointing this out. In Section 5, confusion matrices  have been introduced. Also, in the same section, a table has been added indicating the precision.

Comment 2: The dataset partition for testing and validation must be implemented and discussed to evaluate overfitting.

Answer:

Thanks to the reviewer for pointing this out. The implementation of splitting the dataset for testing and validation purposes is covered in Section 5 of this document. It is specified that 60% of the data is reserved for the training procedure, 20% of the data is reserved for the validation set useful to avoid overfitting and the remaining 20% is reserved for the test set to evaluate the performance of the trained network.

Comment 3: The conclusion must include a comprehensive discussion on the interpretability of the federative learning model’s results, drawbacks, and limitations.

Answer: 

Thanks to the reviewer for pointing this out. The conclusion paragraph has been extended highlighting the main drawbacks and limitations required when dealing with FL. 

Round 2

Reviewer 1 Report

The authors have addressed the reviewer's comments successfully in the revised version. I proposed accepting the paper.

Reviewer 2 Report

The article can be accepted in its current form.