A Novel Implicit Neural Representation for Volume Data

Round 1

Reviewer 1 Report

The authors proposed a method to use Lanczos downsampling before using SIREN to compress the medical images, to overcome the problems of low compression rate and huge GPU memory consumption with a large deep network size. The author's method significantly reduces the training time and GPU memory consumption while maintaining comparable PSNR of the reconstructed images.

1. It seems like the major innovation of this study is the use of Lanczos to downsample the images before using SIREN to compress them? Haven't previous methods used downsampling method?

2. The title of the manuscript needs to be improved, as the term novel representation does not provide any useful information. In addition, the term novel representation in other parts of the text have to be modified accordingly.

3. line 3, "reconstructing low-quality medical images" has to be removed. Otherwise, the authors need to explain why the image quality of the previous INR-based reconstruction method is poor in the introduction section.

4. line 6, How did the authors conclude that the proposed method can improve the compression and reconstruction of super-resolution images without relevant experiments?

5. What is the difference between the three coordinates in the text, voxel coordinates(line 28), UV coordinates(line 73)，3d coordinates(line 104)?

6. lines 45-49, The contribution of the study has to be improved, and the architecture used should be clarified.

7. Sections 2.2 and 2.3, What is the difference between the deep learning methods for medical images and those for super-resolution images? If the authors did not perform experiment with super-resolution images, section 2.3 can be removed.

8. Figure 1 needs to be improved and added with information about the specific methods used by each module.

9. The difference between SIREN-with and SIREN-without is whether using Lanczos downsampling. Why the PSNR of the SIREN-with method is slightly higher SIREN-without method (Tables 4 and 5, Figure 13)?

10. Figure 2 needs to be improved. In particular, the font size should be increased, because the right size of Figure 2 is currently unreadable even when zoomed in.

Author Response

Dear Editor and Reviewers 

Many thanks for the editor’s helpful comments and the reviewers’ valuable questions and suggestions. We have carefully revised the manuscript according to the comments, questions and suggestions. The responses to the editor and reviewers attached as a PDF file. Please see the attachment.

Hope we clearly answer the concerns of the editor and reviewers. 

Any further queries, let me know, please.

Regards,

Armin Sheibanifard

Author Response File: Author Response.pdf

Reviewer 2 Report

The paper need serious revisions.

Improve all images.

Improve acknowledgments.

Improve conclusions.

Too many graphs need to be adjusted in supplementary  file please.

What novelty does this study bring?

Please cite similar research such as works on medical imaging.

Author Response

Dear Editor and Reviewers 

Many thanks for the editor’s helpful comments and the reviewers’ valuable questions and suggestions. We have carefully revised the manuscript according to the comments, questions and suggestions. The responses to the editor and reviewers are attached as a PDF file. Please see the attachment.

Hope we clearly answer the concerns of the editor and reviewers. 

Any further queries, let me know, please.

Regards,

Armin Sheibanifard

Author Response File: Author Response.pdf

Reviewer 3 Report

authors proposed an  interesting  topic，the abstract summarises very well everything that has been done throughout the document.  however there exist some issues the authors should be considered.

1)Too more contents are used to summarized, thedetails about the representation should be described in the section. readers need to the method and results for this representation.

2) the problem has been solved in other ways by other researchers, but it is not clear that the results of existed algorithms， one or two Latest compression algorithms should be added in experiments.  

Author Response

Dear Editor and Reviewers 

Many thanks for the editor’s helpful comments and the reviewers’ valuable questions and suggestions. We have carefully revised the manuscript according to the comments, questions and suggestions. The responses to the editor and reviewers are attached as a PDF file. Please see the attachment.

Hope we clearly answer the concerns of the editor and reviewers. 

Any further queries, let me know, please. 

Regards,

Armin Sheibanifard

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

1. Point 9: Although the authors have explained the differences between SIREN-with and SIREN-withour methods, they still have not explained why the method using the Lanczos downsampling technique did not lose information, but instead enhanced it? The downsampling may have lost little information, but how was the enhancement achieved? The authors need to elaborate on this point.

2. The authors did not mark where the manuscript was modified when replying to the reviewer’s comments and did not highlight the modification in the revision. It brings difficulties to the revision reviewing.

Author Response

Dear Editor and Reviewers

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

I guess you can improve the work by discussing results in the light of past studies. Please work on my previous comments as currently it all looks like a book on mri

Author Response

Dear Editor and Reviewers

Please see the attachment.

Author Response File: Author Response.pdf