Next Article in Journal
Virtual CANBUS and Ethernet Switching in Future Smart Cars Using Hybrid Architecture
Next Article in Special Issue
GCT-UNET: U-Net Image Segmentation Model for a Small Sample of Adherent Bone Marrow Cells Based on a Gated Channel Transform Module
Previous Article in Journal
Meta Classification Model of Surface Appearance for Small Dataset Using Parallel Processing
Previous Article in Special Issue
An Ontology-Based Framework for a Telehealthcare System to Foster Healthy Nutrition and Active Lifestyle in Older Adults
 
 
Article
Peer-Review Record

Arrhythmia Detection Based on WGAN-GP and SE-ResNet1D

Electronics 2022, 11(21), 3427; https://doi.org/10.3390/electronics11213427
by Jing Qin 1, Fujie Gao 2, Zumin Wang 2,*, Lu Liu 3,* and Changqing Ji 2,4
Reviewer 1: Anonymous
Reviewer 2:
Reviewer 3: Anonymous
Reviewer 4: Anonymous
Electronics 2022, 11(21), 3427; https://doi.org/10.3390/electronics11213427
Submission received: 29 September 2022 / Revised: 14 October 2022 / Accepted: 17 October 2022 / Published: 23 October 2022

Round 1

Reviewer 1 Report

The purpose of the manuscript is to propose a deep learning technique based on WGAN-GP and SE-ResNet1D to solve the problem of the unbalanced dataset with ECG classification. Authors claimed to achieve a good result on a dataset with small sample size and unbalanced distribution  Authors need to work on a literature survey, provide result comparison with contemporary studies, and better explain the rationale of their choices, etc. Significant revision is required before this manuscript can be accepted. The major issues as listed below:

Major points:

1.      In section 2.1 author discussed existing research using deep learning on the classification of ECG signals. Authors need to do a better literature survey since a significant amount of research has been done in recent years which showed better accuracy compared to the ones authors presented in section 2. Please emphasize studies done during 2020,2021 and 2022.  

2.      The authors mentioned 3 shortcomings in the list of ECG classifications. Authors need to use multiple significant references in each of the three points. Those references should point out that those studies contained those mentioned shortcomings.

3.      After listing down the shortcomings authors need to point out more research that does not have any of those problems. If the author thinks there is no such study available then the author needs to mention that. If the author thinks or finds such a study then the author needs to provide what difference their research contributed compared to those.

4.      Before going to lines 50-54 talking about what is proposed in the paper, the author needs to use a few sentences about why they have decided to use the techniques in their methods such as: “WGAN-GP”, ResNet1D” etc. Authors can not suddenly bring up some techniques and state that they propose to use those.  

5.      In section 2.2 authors discussed the use of GAN-based ECG classification, the references used (13 to 16) all are from 2019 or 2020. In the last two years, a large amount of research has been done using GAN with ECG classification, authors need to revise that section significantly.

6.      Section 3.1 – both GAN and WGAN-GP were proposed by other people and the author is not proposing anything different than those, so the author needs to shorten section 3.1 and use mainly references (unless authors made any changes to the GAN or WGAN-GP algorithm).

7.      The same goes for section 3.2. If those are proposed by other researchers already then authors do not need to elaborate on them. Simply putting a reference is enough. These are well-known algorithms or techniques in deep learning.

8.      Section 6.2 is not necessary as these are pretty common metrics used in research outcome, simply using references or names of the metrics were enough.

9.      Combine tables 3 and 4 into one table.

10.   Since all the techniques used by authors here are common in other ECG classifications using deep learning in existing research, authors must provide a table comparing their result metrics with other contemporary research (from the last 3-4 years).

Minor points:

1.      Figure 1 is not necessary,  readers suppose to be quite aware of a typical ECG wave.

2.      Figures 2 and 3, are these authors' pictures or taken from another place (citation)?

3.      Equation 1, use citation if it is not proposed by authors.

4.      “Bi-GRU is an improved RNN that is more suitable for acquiring features of the ECG” – reference?

5.      Check citation requirements for all equations and figures (if those are taken from other places).

6.      Citation for MIT-BIH in section 5.1?

7.      Limitations of the author's technique need to be discussed elaborately. Currently, it was discussed in

 

The final section is not sufficient. 

Author Response

We sincerely thank you for your valuable feedback that we have used to improve the quality of our manuscript.

 

Major points:

Point 1: In section 2.1 author discussed existing research using deep learning on the classification of ECG signals. Authors need to do a better literature survey since a significant amount of research has been done in recent years which showed better accuracy compared to the ones authors presented in section 2. Please emphasize studies done during 2020,2021 and 2022.

 

Response 1:

Thanks for your advice. We have rewritten the literature review in section 2.1 and the references is now mainly from the 2022.

 

Point 2: The authors mentioned 3 shortcomings in the list of ECG classifications. Authors need to use multiple significant references in each of the three points. Those references should point out that those studies contained those mentioned shortcomings.

 

Response 2:

Thanks for your suggestion. We have added relevant references to each shortcoming in our resubmitted manuscript.

 

Point 3: After listing down the shortcomings authors need to point out more research that does not have any of those problems. If the author thinks there is no such study available then the author needs to mention that. If the author thinks or finds such a study then the author needs to provide what difference their research contributed compared to those.

 

Response 3:

Thanks for your advice. We have illustrated how our study differs from existing studies in lines 50-53.

 

Point 4: Before going to lines 50-54 talking about what is proposed in the paper, the author needs to use a few sentences about why they have decided to use the techniques in their methods such as: “WGAN-GP”, ResNet1D” etc. Authors can not suddenly bring up some techniques and state that they propose to use those. 

 

Response 4:

Thanks for your suggestion. We have added the reasons for using WGAN-GP and SE-ResNet, now it is in lines 54-57.

 

Point 5: In section 2.2 authors discussed the use of GAN-based ECG classification, the references used (13 to 16) all are from 2019 or 2020. In the last two years, a large amount of research has been done using GAN with ECG classification, authors need to revise that section significantly.

 

Response 5:

Thanks for your advice. We have rewritten the literature review in section 2.2 and the references is now mainly from the 2022.

 

Point 6: Section 3.1 – both GAN and WGAN-GP were proposed by other people and the author is not proposing anything different than those, so the author needs to shorten section 3.1 and use mainly references (unless authors made any changes to the GAN or WGAN-GP algorithm).

 

Response 6:

Thanks for your suggestion. We have simplified section 3.1, retaining only a brief introduction and references to WGAN-GP.

 

 

Point 7: The same goes for section 3.2. If those are proposed by other researchers already then authors do not need to elaborate on them. Simply putting a reference is enough. These are well-known algorithms or techniques in deep learning.

 

Response 7:

Thanks for your advice. We have also simplified section 3.2, keeping only a brief introduction and references to ResNet and SeNet.

 

Point 8: Section 6.2 is not necessary as these are pretty common metrics used in research outcome, simply using references or names of the metrics were enough.

 

Response 8:

Thanks for your suggestion. We have deleted section 6.2 and described the evaluation indicators used in a single sentence in the last of section 6.1(lines 286-287).

 

Point 9: Combine tables 3 and 4 into one table.

 

Response 9:

Thanks for your advice. We have combined Table 3 and Table 4 into one table, which is now Table 3 in our resubmitted manuscript.

 

Point 10: Since all the techniques used by authors here are common in other ECG classifications using deep learning in existing research, authors must provide a table comparing their result metrics with other contemporary research (from the last 3-4 years).

 

Response 10:

Thanks for your suggestion. We have added a new table: Table 4, which compares the performance of the proposed method with that of the last three years of researchs on deep learning methods.

 

Minor points:

Point 1: Figure 1 is not necessary, readers suppose to be quite aware of a typical ECG wave.

 

Response 1:

Thanks for your advice. We have removed Figure 1 from our resubmitted manuscript.

 

Point 2: Figures 2 and 3, are these authors' pictures or taken from another place (citation)?

 

Response 2:

Thanks for your comments. As we have simplified section 3.2, these two pictures have also been removed.

 

Point 3: Equation 1, use citation if it is not proposed by authors.

 

Response 3:

Thanks for your advice. As we have simplified section 3.1, this equation has also been removed.

 

Point 4: “Bi-GRU is an improved RNN that is more suitable for acquiring features of the ECG” – reference?

 

Response 4:

We sincerely appreciate the valuable comments. We have added the reference, which is “[32] Zhang, X.; Li, R.; Dai, H.; Liu, Y.; Zhou, B.; Wang, Z. Localization of myocardial infarction with multi-lead bidirectional gated 407

recurrent unit neural network. IEEE Access 2019, 7, 161152–161166.”.

 

Point 5: Check citation requirements for all equations and figures (if those are taken from other places).

 

Response 5:

Thanks for your advice. We have checked all equations and figures, all the equations are presented in this paper and all the figures are drawn by us.

 

Point 6: Citation for MIT-BIH in section 5.1?

 

Response 6:

We sincerely thank the reviewer for careful reading. We have added the reference, which is “[4] Moody, G.B.; Mark, R.G. The impact of the MIT-BIH arrhythmia database. IEEE Engineering in Medicine and Biology Magazine 350 2001, 20, 45–50.”.

 

Point 7: Limitations of the author's technique need to be discussed elaborately. Currently, it was discussed in the final section is not sufficient.

 

Response 7:

Thanks for your comments. We have added more discussion of limitations of our technique in section 7.

Reviewer 2 Report

Colleagues Qin et al. describe in their manuscript, entitled "Arrhythmia Detection Based on WGAN-GP and SE-ResNet1D" about two ECG classification methods, that showed better performance on automated ECG evaluation, that could have the potential to be a useful diagnostic tool to assist cardiologist in the diagnosis of arrhythmias.

The manuscript is well written, but from the standpoint as a cardiologist I have a few comments:

Used ECG data sets: Please describe this sample size. What patients? ECG analysis and evaluation made by whom?

Figure 8. Sample ECG still quite unphysiological as P wave is very "peaky". ST-segment depression would have the potential in a real patient in more than two neighboured leads to be pathological (ischemia).

Table.1 Please define Fusion or give examples.

Author Response

We sincerely thank you for your valuable feedback that we have used to improve the quality of our manuscript.

 

Point 1: Used ECG data sets: Please describe this sample size. What patients? ECG analysis and evaluation made by whom?

 

Response 1:

Thanks for your comments. We have added a description of the ECG data set in section 5.1.

 

“The experimental dataset we used was the MIT-BIH arrhythmia database, open-sourced by MIT, which has been widely used by the academic community since it was made public in 1978 and has become the most widely used small ECG dataset in academic research today. The MIT-BIH Arrhythmia Database contains 48 half-hour excerpts of two-channel ambulatory ECG recordings, obtained from 47 subjects studied by the BIH Arrhythmia Laboratory. Twenty-three recordings were chosen at random from a set of 4000 24-hour ambulatory ECG recordings collected from a mixed population of inpatients (about 60%) and outpatients (about 40%) at Boston's Beth Israel Hospital; the remaining 25 recordings were selected from the same set to include less common but clinically significant arrhythmias that would not be well-represented in a small random sample. All ECGs are automatically annotated by the machine with the R-peak positions and calibrated by two cardiologists.”(section 5.1)

 

Point 2: Figure 8. Sample ECG still quite unphysiological as P wave is very "peaky". ST-segment depression would have the potential in a real patient in more than two neighboured leads to be pathological (ischemia).

 

Response 2:

Thanks for your comments. The previously selected sample image was not well enough and we have reselected the ECG image, it is now Figure 5 in our resubmitted manuscript.

 

Point 3: Table.1 Please define Fusion or give examples.

 

Response 3:

We sincerely appreciate the valuable comments. We have added footnotes to Table 1 to illustrate the definition of Fusion.

 

“Fusion denotes fusion of ventricular and normal beat”(Table 1)

Reviewer 3 Report


Comments for author File: Comments.pdf

Author Response

We sincerely thank you for your valuable feedback that we have used to improve the quality of our manuscript.

 

Point 1: The author should improve the introduction to make more clear about the importance of the proposed methods in this study.

 

Response 1:

Thanks for your comments. We have improved the introduction in our resubmitted manuscript. The new introduction highlights the contributions of our work and the importance of the proposed methods in our study.

 

Point 2: Modify all captions of the figures describe clearly the meaning of figures. The detailed description will help the reader get a better understanding of the figures (i.e. Figure 5. Generation results of different stages: (a) what is it? (b) what is it, etc.).

 

Response 2:

We sincerely appreciate the valuable comments. We have refined the captions of all images to ensure that they are described accurately and in detail.

Reviewer 4 Report

The paper has an interesting topic. There are used two networks and designed to detect arrhythmia on ECG signals.

The simulations results were compared with others three existing methods and they showed that the proposed algorithm has the best performances in terms of precision, recall and F1 measure. This method can be useful diagnostic tool to assist the cardiologist in the diagnosis of arrhythmia.

The paper is well organized, the methods are clearly presented. It technically sounds. It has a good state-of-the-art, the references are relatively new.

The results are clearly presented.

There is a small observation: there are some sentences were the authors used the same word twice (for example: proposed).

Author Response

We sincerely thank you for your valuable feedback that we have used to improve the quality of our manuscript.

 

Point 1: There are some sentences were the authors used the same word twice (for example: proposed).

 

Response 1:

We sincerely appreciate the valuable comments. We have modified this problem by using richer wordsin addition to proposed(such as developed, used and conducted etc.) in our resubmitted manuscript.

Round 2

Reviewer 1 Report

Authors have revised as suggested.

Back to TopTop