Next Article in Journal
Experimental Study on Flexural Properties of Polyurethane–Cement Composites under Temperature Load
Next Article in Special Issue
Perturbation-Based Explainable AI for ECG Sensor Data
Previous Article in Journal
Extensible Hierarchical Multi-Agent Reinforcement-Learning Algorithm in Traffic Signal Control
Previous Article in Special Issue
Computer-Aided Detection of Hypertensive Retinopathy Using Depth-Wise Separable CNN
 
 
Article
Peer-Review Record

Applying Convolutional Neural Network in Automatic Assessment of Bone Age Using Multi-Stage and Cross-Category Strategy

Appl. Sci. 2022, 12(24), 12798; https://doi.org/10.3390/app122412798
by Ching-Tung Peng 1,*, Yung-Kuan Chan 2, Yeong-Seng Yuh 3 and Shyr-Shen Yu 1
Reviewer 1: Anonymous
Reviewer 2: Anonymous
Reviewer 3:
Appl. Sci. 2022, 12(24), 12798; https://doi.org/10.3390/app122412798
Submission received: 21 October 2022 / Revised: 6 December 2022 / Accepted: 7 December 2022 / Published: 13 December 2022
(This article belongs to the Special Issue Recent Advances in Deep Learning for Image Analysis)

Round 1

Reviewer 1 Report

This paper builds a multi-stage and cross-category CNN for bone age assessment. It is an interesting topic. However, it lacks comparative experiments with other state-of-the-art methods, such as the References [16]-[27]. It is important to prove the authors’ contributions. CNN and FNN can be found in any textbook, which does not need to be introduced. In addition, the authors do not explain in detail how to label the dataset, and if the authors can public the dataset and code, it will be more helpful for the research society. 

Author Response

  1. We add a comparison with 4 methods, such as section 4.6 (lines 540-548 of the revised version).
  2. We deleted most of the descriptions of CNNs (in Sections 2.1.1-2.1.2).
  3. The labeling method is based on the GP method by two doctors, and we added an explanation in the 289-290 lines of the revised version.

Author Response File: Author Response.docx

Reviewer 2 Report

Bone age assessment is very important to check if the children grow normally or not. Aiming at solving the disadvantages of the Greulich-Pyle Atlas method and the Tanner-Whitehouse method, the authors built an automatic bone age assessment system using the convolutional neural network (CNN). They also try to improve the accuracy of the CNN with a cross-stage strategy. However, it should not be published until the authors addressed the following points:

1. Why the authors only use two-step stage classification strategy ? Is it better if they use a multi-stage strategy ? It would be more useful for the readers to understand the advantage of the proposed method if the author can increase three-stage classification performance in Table 5.

2.  It is very hard to recognize those numbers or words in the pictures, such as Figure1 and Figure 3. Can the authors increase the font size ? They are two blurry.

3. The author must double check the typing errors since there are a lot of errors in the context. For example, in line 261 S(x) epresents……; In our data, the max (S(x)) and max(S(x))-Min(S(x))… shown in Table 2. However, there is not the value of max(S(x))-Min(S(x)) instead of Min(S(x)).

     The equation 4 should be corrected as,

                                                                                                                (4)

     In addition, the authors didn’t explain the meaning the in the equation 6, ,in the equation 8, ACC. in the equation 9.

     From line 322 to 323, BA shouldn’t be a bold type. Furthermore, there must be a .

4. There are several faulty wordings in the context. These problems make it difficult for the readers to understand the meaning of the full text. For example, there is a super long sentence from line 362-369. It is indeed a difficult task to understand what the authors would like to say.

     Besides, there is a very poor expression from 388-390. It is really tough to understand, …our system can deal well with children at any stage.

Comments for author File: Comments.pdf

Author Response

We performed the three-stage experiment with 2 configurations and updated Table 6 to illustrate line 467-479 of the revised version.
We have redrawn Figure 1 and Figure 3 to make them clearer.
(1)We corrected the error in Table 2. (2)in equation 6 we added a description in lines 401-402 of the revised version. (3)Equation 8-9 we modify in lines 430-434  of the revised version.

(1)We rewrite lines 362-369 in lines 503-212 of the revised version. (2)In lines 388-390 of the original version, we want to describe that we can have good predictive ability in children at all stages. We rewrite them in revised version lines 537-538.

Author Response File: Author Response.docx

Reviewer 3 Report

1、 If the training datasets are got from GP or TW results, the CNN will still have human made error (line 58). Then you cannot say that this study can avoid the human made error.

2、 There are too many descriptions for the method and study route in Section 1, move it to Section 3. And in Section 1, the bone age assessment research progress needed to be added. According to the multi-step and cross stage method in this study, is there any background or progress in previous research? What pain points have been solved?

3、 Give the full name of ROI

4、 Line 143-180 are previous research and progress, move them to Section 1

5、 Give the full name of MSCS in the main body

6、 There are many repeated descriptions in Section 3.2 to 3.4

7、 Move the validation/evaluation method into Section 3

8、 The meaning of “one/two-step” in Section 4.3 is different from “Step 1-3” in Section 3.3-3.5. It’s easy to make confusion, it’s better to change a word/statement. What’s more, in table 5, the “stage” is non-corresponding with “step” in the text above.

9、 “[8]” is used as subject, object or pronoun for so many times in this paper, which is nonstandard in paper writing.

10、           Is it “column of table 7” or “row in table 7” in line 365?

11、           Please give specific configuration of config.1 and config.2 in table 7.

12、           Which is the evaluation standard in Section 4.4? “staged classification accuracy” or “age assessment performance”Please make it clear. All “stage” or “step” configuration process are served for the final age assessment, so just compare the accuracy/MAE of age assessment. It is recommended that list all configuration and step choosing and accuracy/MAE comparison with a table plus a figure in Section 4.

Author Response

Our ground truth is judged by two physicians (additional descriptions are in lines 289-290 of the revised version). Human errors only exist between them. After the model training is completed, there will be no human errors caused by other physicians' judgments. Therefore, we believe that human error can still be reduced by model judgment. So we changed the term "no human error" to "reduce human error"(line 62 in the revised version).
We have modified the article configuration according to your suggestion, for the cross-category configuration is our proposed strategy in this article, we also try to describe why it is beneficial in the experiments in the article.
In Figure 1 we added them.
We have moved them thanks for the suggestion.
On line 190 of the revised version, we added his full name.
We appropriately delete the repeated part in sections 3-2.
We have moved them to a revised version of sections 3-8.
We revisited all the tables and they did have errors, we have fixed them, thank you very much for reminding us.
In the full article, we replace "Ref.[8]" to "[8]".
They are true "row", we have fixed them.
We add the description in lines 522-526 of the revised version.
We did get the "stage" or "step" wrong, we fixed them in the table.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

 The authors have responded to most of my comments, but the English still needs to be proofread by native speakers before publication.

Author Response

I have used the MDPI English editing service and have finished editing.
Thank you for your advice.

Author Response File: Author Response.docx

Reviewer 2 Report

All the points I listed have been fixed, I don't have any more questions.

Author Response

We revised the article following the suggestion of other reviewers to use native English speakers for editing.
The revised version is attached here.
Thanks again reviewer for your suggestion.

Author Response File: Author Response.docx

Reviewer 3 Report

No more comment.

Author Response

We revised the article following the suggestion of other reviewers to use native English speakers for editing.
The revised version is attached here.
Thanks again reviewer for your suggestion.

Author Response File: Author Response.docx

Back to TopTop