A Systematic Literature Review of 3D Deep Learning Techniques in Computed Tomography Reconstruction

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This paper presents a review of 3D deep learning techniques in computed tomography reconstruction. The topic is interesting and timely. However, there are many issues with the current manuscript.

(1) There are many important literatures on 3D CT reconstruction that are not involved in this work. I would suggest expanding the search space for including important lieratures in this review.

(2) Apparently, some papers that were reviewed are not related to CT reconstruction; they are more about diagnosis. This could be found in the population performance showing auc or classification accuracy.

(3) The presentation should be improved and carefully checked. There are many statements that are not very professional. 

(4) Fig. 1 should be refereed from one paper. Credit should be given.

(5) It would like to see some discussion with existing review papers on CT reconstruction.

(6) There are different aspects in CT reconstruction such as low-dose or sparse-view. They have different options for model selection. Please consider enhancing this paper.

(7) In addition to (6), the methods could also divided into image domain or dual-domain. 

(8) Basic tomography reconstruction should be given.

 

Comments on the Quality of English Language

It could be improved with native speakers.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

In this manuscript, the authors listed many literature sources about neural network-based tomography reconstruction methods and datasets available for the training and testing of these networks. It provides useful information for readers to know the progress of this research field. The topic of this manuscript is interesting. However, it didn’t provide an in-depth discussion about each method/research field and didn’t specify the advantages or disadvantages of these methods. Many acronyms and terminologies were not clearly explained, and the manuscript was not well organized. Overall, I recommend a major revision before submission. Below are comments:

 

1.     Many acronyms were not clearly explained and even mixed. For example, the authors use DLR to express both deep learning reconstruction and deep learning regularization. It is very confusing which technique the authors referred to in the manuscript. Also, what is the difference between DCNN and CNN? They seem the same to me.

2.     The authors mentioned many different literatures and used many acronyms to represent these methods. However, it is not clear the meaning of these acronyms. Could the authors elaborate on the difference between DLR and CNN methods? Does the DLR method use a fully connected multi-layer neural network for reconstruction? What is the general input and output of the DLR method? What about the DLIR method? The author even didn’t give the full name of this method. In addition, for GAN methods, many of them are still based on CNN-based generators/discriminators. Why did the authors list these methods separately?

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

I don't have further comments.

Comments on the Quality of English Language

some typos can be fixed after carefully checking.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf