All reviews of published articles are made public. This includes manuscript files, peer review comments, author rebuttals and revised materials. Note: This was optional for articles submitted before 13 February 2023.
Peer reviewers are encouraged (but not required) to provide their names to the authors when submitting their peer review. If they agree to provide their name, then their personal profile page will reflect a public acknowledgment that they performed a review (even if the article is rejected). If the article is accepted, then reviewers who provided their name will be associated with the article itself.
Author has addressed reviewer comments properly. Thus I recommend publication of the manuscript.
[# PeerJ Staff Note - this decision was reviewed and approved by Xiangjie Kong, a PeerJ Section Editor covering this Section #]
The authors have addressed my previous comments. The revised manuscript is much improved and ready for publication.
In Table I and Table II, we see different levels of noise from 30dB to 40dB. Could you add details and describe clearly how the different noises were simulated?
"The projections were contaminated with noise between 30-40 dB." This is too brief for other researchers to reproduce the experiments.
n/a
This manuscript is ready for publication, pending minor revision.
The authors made all my revision requests. There is no additional revision.
The authors made all my revision requests. There is no additional revision.
The authors made all my revision requests. There is no additional revision.
no comment
The answers given to the criticisms made in the previous review are sufficient and appropriate. I have no other comment.
The answers given to the criticisms made in the previous review are sufficient and appropriate. I have no other comment.
The answers given to the criticisms made in the previous review are sufficient and appropriate. I have no other comment.
Accept
Good
Yes
NA
Based on the referee reports, I recommend a major revision of the manuscript. The author should improve the manuscript, taking carefully into account the comments of the reviewers in the reports, and resubmit the paper.
**PeerJ Staff Note:** Please ensure that all review and editorial comments are addressed in a response letter and that any edits or clarifications mentioned in the letter are also inserted into the revised manuscript where appropriate.
**Language Note:** The review process has identified that the English language must be improved. PeerJ can provide language editing services - please contact us at copyediting@peerj.com for pricing (be sure to provide your manuscript number and title). Alternatively, you should make your own arrangements to improve the language quality and provide details in your response letter. – PeerJ Staff
This manuscript presents a self-supervised “projection to projection” method for low-dose CT. While the paper is easy to read, I have a few concerns with the current manuscript.
1. Even the FBP images are not terrible in this study. We would like to see the results of proposed method with even lower dose levels (sparser views, noisier projections), so that we know the full potential of the proposed approach.
2. "parallel beam projections were generated using Radon transform; image resolution was selected as 512x512; detector resolution was selected as 512x1 pixels". These settings are too ideal and far from reality. It is unclear whether this method is practical for clinical use.
3. The authors compared the performance of the proposed method with several iterative reconstruction approaches. It is also important to include state-of-the-art deep learning approaches as a comparison.
1. Is the trained mapping f(theta) different for different datasets because of the noise/sparsity level? Could the authors shed more light on this aspect?
2. In figure 3g and figure 4g (not figure 5g), why are there many streak and patchy artifacts? Especially, figure 4g, Projection2Projection trained on human CT dataset, the resulting image is still worse than figure 4f.
n/a
Here are some comments that I think the authors should address to improve the paper.
-Why did you just compare the method with the analytical methods? Briefly explain the motivation for selecting the following algorithms for comparison. FBP, SART, SART+TV, SART+BM3D. The method can also compared with a state-of-the-art unsupervised or supervised deep neural network-based method.
- Please clarify how data are split for training, validation and testing for reproducibility of the algorithm
- Do you think the proposed network is hyperparameter dependend considering the clinical usability ?
- What was your criterion while selecting the certain perturbation ?
- Writing should be standardized and checked.
- Captions of all figures should be rearranged correctly.
no comment
no comment
You addressed an important topic but need language corrections (proofreading should be done). Authors need attention to be paid to the clarity of expression and readability. (especially the use of references within the text; please check the citation format of the journal. eg. Line 50-51 etc.)
The first time you use the term, you can acronym it and continue using the acronym thereafter. Also, please do not introduce an acronym unless you will use it a minimum of three or four times. Unfortunately, you used many more acronyms, but some of them were not explained(eg. FBP etc.), and some were used once in the text, especially in the related work section (you do not need to give the acronym if you will not use it again).
What are the keywords of the study?
Please replace the paragraph at the end of the Introduction that explains your new approach (the first sentence is not a good place to start). Also discuss your work in comparison to your preliminary work published at the conference. What's the difference or what did you add? What is the important contribution of this study to the literature? From what I've examined, it appears to be a repeat of a similar study; Therefore, please emphasize its differences and innovation.
Also, the Figure 1 you used is also available in your other publication, so it would be better to reference it.
What is the size of each dataset, is it sufficient for CNN algorithms?
How do you split the training and reconstruction part or can you use k-folding?
The comparison metrics are PSNR and SSSIM. What are these and how can they be calculated?
Additionally, the iteration rate and learning rates you use are determined by where. After making a comparison and getting results with different values, it should be stated whether the most optimal values are, if so.
The paper identify clearly implications for research, and these implications reasonably good consistent with the findings and conclusions of the paper. Although the results are presented, I still do not feel fully convinced. In discussion part, need support for the arguments and claims.
In my opinion, since the continuous use of the pronoun "we" in the written language causes informal writing, passive uses will make the text more readable so that it is more formal.
See the attachment.
See the attachment.
See the attachment.
See the attachment.
All text and materials provided via this peer-review history page are made available under a Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.