Hamming Code Strategy for Medical Image Sharing
Round 1
Reviewer 1 Report
Dear Authors,
Thank you for your submission of a manuscript regarding utility of the Hamming code in medical image sharing and its security.
The idea of using the Hamming code is indeed interesting and unique and I appreciate your experiments and extensive examples.
Having said that, the manuscript has many flaws, especially considering it's main idea and aim of the study. You suggest that Hamming code implementation should work great with medical image sharing. I understand the idea and the concept of having two shadows that have no value without each other and I agree that it theoretically adds another security layer. However, you failed to prove it in practice. Results and most specifically, figures are very uninformative. Poor resolution of images makes them illegible. You don't show any specific features that have been "lost" in the process of creating a shadow.
Moreover, after reading your introduction as well as discussion/conclusions a reader might have the understanding that you proved a fast & computationally-efficient solution for medical image sharing. But there is no data at all on what are the sizes of created shadow volumes? What is the time to generate shadows for whole volumes? How does it compare to methods presented in Related works? What is the advantage of your approach?
It is also questionable whether there is even a need for this sort of solution in radiology right now. In conclusions, you say that your scheme is suitable for hospitals and clinics. Except that it's not. Applying your idea would require all vendors and providers of PACS systems to totally change whole environment. Or require small institutions to write their own PACS systems which is even less effective. This solution also does not comply with DICOM standard. I don't see how Authors see this solution being implemented in clinical practice.
The manuscript requires extensive proofreading and editing. The introduction is written in unprofessional manner and suggests that further text will focus on implementation on mobile devices.
Happy Holidays and all the best in the New Year.
Author Response
Please see attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The paper reports an easy way to maintain confidential information for medical image. If the method is alternative to classical ones based on secret encryption, in my opinion the chaos base encryption systems must be presented. But it is a personal opinion.
A very efficient manner for encryption could be derived by the strange behaviour of HR neuron networks. The HR dynamics is not numerically strong to implement also in digital form.
The author could keep take into account the following paper tio introduce in the references:
Physics Letters, Section A: General, Atomic and Solid State PhysicsVolume 266, Issue 1, 14 February 2000, Pages 88-93
Slow regularization through chaotic oscillation transfer in an unidirectional chain of Hindmarsh-Rose models(Article)
La Rosa, M.a,d,e, Rabinovich, M.I.a, Huerta, R.a,b,f Abarbanel, H.D.I.a,c, Fortuna, L.e
I suggest to remark this point in the introduction page 2 row 4.
Moreover I appreciate the paper for the results and for the manner of the results presentation.
Author Response
List of Revisions
Authors: Li Li, Ching-Chun Chang, Junlan Bai, Hai-Duong Le, Chi-Cheng Chen, and TeenHang Meen
Title: Hamming Code Strategy for Medical Image Sharing
We greatly appreciate the valuable comments of the reviewers; these comments have significantly improved the quality of our submitted manuscript. Based on those suggestions, the manuscript has been revised accordingly. The major revisions are briefly described as follows:
Response to the suggestions of Review 2:
The paper reports an easy way to maintain confidential information for medical image. If the method is alternative to classical ones based on secret encryption, in my opinion, the chaos base encryption systems must be presented. But it is a personal opinion. A very efficient manner for encryption could be derived by the strange behaviour of HR neuron networks. The HR dynamics is not numerically strong to implement in digital form.
The author could keep taking into account the following paper to introduce in the references
Physics Letters, Section A: General, Atomic and Solid State Physics Volume 266, Issue 1, 14 February 2000, Pages 88-93. Slow regularization through chaotic oscillation transfer in an unidirectional chain of Hindmarsh-Rose models(Article) La Rosa, M.a,d,e, Rabinovich, M.I.a, Huerta, R.a,b,f Abarbanel, H.D.I.a,c, Fortuna, L.e
I suggest remarking this point in the introduction page 2 row 4.
Moreover, I appreciate the paper for the results and for the manner of the results presentation.
Response:
Many thanks for your positive comments and valuable suggestions. As the referee’s saying goes, either the chaos-based encryption or the neuron networks based encryption is a very efficient manner for image encryption. Theoretically, those encryption methods provide a rather stronger security level, however, they are computationally expensive and take time to execute on a mobile device. But it is undoubtedly a worthy subject of study in the near further, such as reversible data hiding in encrypted images.
Meanwhile, according to the referee’s suggestion, we have added some citations [6-12] related to data sharing or encryption based privacy protection on mobile devices. And a more detailed literature review has been addressed in the revised manuscript.
[Modified] (In Introduction)
In the recent decades, some researches related to protecting privacy information for mobile services on different techniques, such as the redesign of the architecture of network [6], the non-interactive privacy-preserving protocol for image similarity computation [7], the data sharing protocol by using a new cryptographic primitive named online/offline attribute-based proxy re-encryption, and the transform key technique [8], have been developed. Also, some encryption methods based on cryptography, such as homomorphic encryption [9], elliptic curve cryptography based encryption [10] and chaotic oscillation theory based encryption [11, 12], were designed and applied to provide the confidentiality of patients’ health information.
[Modified] (In References)
[6] Zhang, K.; Yang, K.; Liang, X.; Su, Z.; Shen, X.; Luo, H. H. Security and privacy for mobile healthcare networks: from a quality of protection perspective. IEEE Wireless Commun., 2015, 22, 104-112.
[7] Zhang, L.; Jung, T.; Liu, C.; Ding, X.; Li, X. Y.; Liu, Y. Pop: Privacy-preserving outsourced photo sharing and searching for mobile devices. International conference on distributed computing systems. IEEE, Columbus, USA, 2015, 308-317.
[8] Shao, J.; Lu, R.; Lin, X. Fine-grained data sharing in cloud computing for mobile devices. 2015 IEEE conference on computer communications (INFOCOM), Kowloon, Hong Kong, 2015, 2677-2685.
[9] Ibtihal, M.; Hassan, N. Homomorphic encryption as a service for outsourced images in mobile cloud computing environment. Cryptography: Breakthroughs in research and practice. IGI Global, 2020, 316-330.
[10] Shankar, T. N.; Sahoo, G.; Niranjan, S. Image encryption for mobile devices. International conference in communication control and computing technologies, IEEE, Ramanathapuram, India, 2010, 612-616.
[11] La Rosa, M.; Rabinovich, M. I.; Huerta, R.; Abarbanelac, H. D. I.; Fortuna, L. Slow regularization through chaotic oscillation transfer in an unidirectional chain of Hindmarsh–Rose models. Physics Letters A, 2000, 266, 88-93.
[12] Wang, F.; Ding, J.; Dai, Z.; Peng Y. An application of mobile phone encryption based on Fibonacci structure of chaos. 2010 Second world congress on software engineering. IEEE, Wuhan, China, 2010, 2, 97-100.
Author Response File: 
Author Response.doc
Round 2
Reviewer 1 Report
Dear Authors,
I would like to thank you for your prompt and very extensive revision. The manuscript has improved significantly and I am much more convinced about the results and its clinical value.
Especially looking at Table 3, which compares original images to Hamming(7,4) and Hamming(15,11). This is a crucial and very convincing figure. I still have a few doubts about potential selection bias. But, given that this is just a feasibility study, I am satisfied with this figure as it absolutely shows potential of Hamming code for anonymizing images.
I have just a few minor comments.
First of all, I appreciate adding information about execution time & comparison in Table 5. You mention using GPU for computing, however there is no information about the setup. Is the 0.6s execution time for cluster-level GPUs? It would be nice to mention your hardware used for computation. Also, is reported time of ~0.6 sec for 512x512 or 138x138 images? Is it execution on full 3D volume or a single 2D slice?
In section 5.1. you said that you used "X-ray scans". Did you mean computed tomography (as shown in Figures and Tables)?
Moreover, manuscript still requires English proofreading. There are multiple grammar errors, the best example being abstract sentence: "With low computational cost, suitable for Tablet, pamphlets and other mobile devices".
I also noticed that you refer to Figures 3 and 4 on page 11 ("The original images and their shadows are shown in Figures 3 and 4 for algorithms..."). I believe you mean Table 3 (or 1-3)?
Good luck
Author Response
List of Revisions
Authors: Li Li, Ching-Chun Chang, Junlan Bai, Hai-Duong Le, Chi-Cheng Chen, TeenHang Meen
Title: Hamming Code Strategy for Medical Image Sharing
We greatly appreciate the valuable comments of the reviewers; these comments have significantly improved the quality of our submitted manuscript. Based on those suggestions, the manuscript has been revised accordingly. The major revisions are briefly described as follows:
Response to the suggestions of Review 1:
I would like to thank you for your prompt and very extensive revision. The manuscript has improved significantly and I am much more convinced about the results and its clinical value.
Especially looking at Table 3, which compares original images to Hamming (7,4) and Hamming(15,11). This is a crucial and very convincing figure. I still have a few doubts about potential selection bias. But, given that this is just a feasibility study, I am satisfied with this figure as it absolutely shows potential of Hamming code for anonymizing images.
I have just a few minor comments.
First of all, I appreciate adding information about execution time & comparison in Table 5. You mention using GPU for computing, however there is no information about the setup. Is the 0.6s execution time for cluster-level GPUs? It would be nice to mention your hardware used for computation. Also, is reported time of ~0.6 sec for 512x512 or 138x138 images? Is it execution on full 3D volume or a single 2D slice?
Response:
Many thanks for your positive comments and valuable suggestions. In this paper, all experiments were implemented by MATLAB R2017a, and the simulation environment for experiments was an Intel(R) Core(TM) i5–8500 v5 Hexa-core processor with 8 GB of RAM. Correspondingly, the average execution time of 0.69s is conducted in aforementioned hardware and software platform. Meanwhile, the execution time of creating shadows and reconstructing original images listed in Table 4 are obtained at the same condition, where all images are 8-bit depth gray-scale images (a single 2D slice) with the size of 512×512. Implementing Hamming code on graphics processing unit (GPU) [30] can speed up the process 99 times compared with normal sequential approach, and ensure fast response to those applications that are time-sensitive. In the near further, we will try to do this job in practical applications.
Many thanks for your valuable comments again. A more detailed revision has been addressed in the revised manuscripts.
[Modified] (In Subsection 5.1)
In addition, all experiments were implemented by MATLAB R2017a, and the simulation environment for experiments was an Intel(R) Core(TM) i5–8500 v5 Hexa-core processor with 8 GB of RAM.
[Modified] (In Subsection 5.1)
It should be noted that all images used in Table 4 are 8-bit depth gray-scale images (each one is a single 2D slice) with the size of 512512.
In section 5.1. you said that you used "X-ray scans". Did you mean computed tomography (as shown in Figures and Tables)?
Response:
Many thanks for your valuable comments. The medical images used in Tables 1 and 2 are modality of computed tomography. As for the medical images used in Table 3, one is computed tomography (the former) and the other belongs to other modality (http://www.dicom-solutions.com/modalities.php)
[Modified] (In Subsection 5.1)
In our experiments, first, four pairs of 138138 medical images of scale X-ray scans were selected, which are the modality of computed tomography.
[Modified] (In Subsection 5.1)
Among which, the former image is computed tomography and the other belongs to other modality.
Moreover, manuscript still requires English proofreading. There are multiple grammar errors, the best example being abstract sentence: "With low computational cost, suitable for Tablet, pamphlets and other mobile devices".
Response:
Many thanks for your valuable suggestions. According to the suggestion, a technical writer whose first language is English has proofread the revised paper. A more detailed revision has been addressed in the revised manuscripts.
[Modified] (In Abstract)
With low computational cost, the proposed scheme is suitable for tablet, pamphlets and other mobile devices.
[Modified] (In Section 3)
Suppose that the parity-check matrix is H= [ImQ], where Im is an m´m identity matrix and Q is an m-tuple with weight two or more.
[Modified] (In Subsection 4.1)
From [p’1 p’2 p’3 d’1 d’2 d’3 d’4], the correct codeword [p1 p2 p3 d1 d2 d3 d4] is recovered, and thus the value of (x1 x2 x3) is obtained. Finally, the original pixels are recovered.
[Modified] (In Subsection 4.1)
No error has 50% (∵7/16+1/16) probability.
I also noticed that you refer to Figures 3 and 4 on page 11 ("The original images and their shadows are shown in Figures 3 and 4 for algorithms..."). I believe you mean Table 3 (or 1-3)?
Response:
Many thanks for your valuable suggestions. We are very sorry for our negligence of the rough usage of Figures 3 and 4. We have corrected it, and a more detailed revision has been addressed in the revised manuscripts.
[Modified] (In Introduction)
The original images and their shadows are shown in Tables 1 and 2 for algorithms using (7,4) Hamming and (15,11) Hamming, respectively.
Author Response File: Author Response.doc
