A Low-Power ECG Processor ASIC Based on an Artificial Neural Network for Arrhythmia Detection

Round 1

Reviewer 1 Report

The work seems to be very good provided authors need to revise the manuscript for possible publication.

1.      Authors mentioned “Traditional methods for detecting arrhythmia require professional skills and equipment, and patients need to go to the hospital for examination”. What are those traditional methods mention few names and their limitations.

2.      Authors reported “Some processors are assisted by artificial intelligence algorithms to perform on-chip classification and detect heart diseases [4-11]”. What are those processors be specific. Furthermore, group citations need to be strictly avoided. Authors need to explain individual contributions of your referred literature.

3.      Authors need to revise the Figures, as they use many abbreviations and need to be explained either in Figure or in the manuscript description for ease of readers.

4.      Authors need to clearly define the type of neural network and algorithms used to train it.

5.      The hidden layered neural network consists of 8 neurons and how those neurons has been selected. Does the authors conduct any parametric study.

6.      Authors need to provide detailed insights of the following:

a.      Type of architecture and the basis for selection

b.      How many data sets used for training and whether those data has been normalized or not to limit the numerical overflows.

c.      What are the different transfer functions used to train the NN

d.      What is the stopping criteria or error goal or epochs

e.      How many testing data used

f.       What was the training and testing error with respect to R²

7.      Authors proposing their work for real time processing and requires ANN to predict or respond immediately to have accurate control or monitoring. Kindly let authors revise with respect to computation time of ANN.

8.      More discussion on Table 2, could help the readers on how this work will benefit the readers.

9.      Authors need to explain the scope for future work by highlighting the limitations of yours will be beneficial.  

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

General comment:

The manuscript introduces an ASIC intended for fast and effective arrhythmia detection using ECG signals. The proposed architecture works around dedicated signal processing stages and a lightweight ANN for signal classification. The work is relevant in the field of integrated circuits coupled with machine learning biomedical applications. Furthermore, the proposal is well-motivated and represents an advance in its area of study. The results are some weak In general, the manuscript is fluent, and original. I have some points that should be addressed.

Comment 1:

Revise the units format. The standard is value+space+units. Also, KHz, should be kHz.

 

Comment 2:

Why did the ANN architecture was selected?

 

Comment 3:

Though the proposed ASIC is well detailed, the manuscript lacks results for ECG signal acquisition and processing in real world. At least, a proof-of-concept is needed to support the main findings of the work.

 

Comment 4:

Regarding the confusion matrix in Fig. 8. There is no mention of the accuracy and sensitivity of the classification task.

 

Comment 5:

Which is the main advantage of the proposed ASIC compared with commercial available chips such as the MAX30003?

Check for grammatical errors and typos.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Current manuscript entitled “A Low-Power ECG Processor ASIC Based on Artificial Neural Network for Arrhythmia Detection” by “Zhang et al” deliberated on the Early detection of arrhythmia based on artificial neural network. The parameters required for classification are configurable to facilitate the updating of the algorithm model. The proposed ECG processor ASIC is implemented using 55nm CMOS technology, occupying an area of 0.33 mm2. This design consumes 12.88μW at 100KHz clock frequency, achieving a classification accuracy of 96.69%. The implementation results indicate that this ASIC provides a good solution for lowcost ECG monitoring. The manuscript seems good and can be accepted after addressing the following comments.

1.      The structure of this paper needs to be improved.

2.      Provide the full form of FIR in figure 2 caption.

3.      Insufficient summary and analysis of research issues in the Introduction.

4.      The Abstract should contain answers to the following questions: What problem was studied and why is it important? What methods were used? What are the important results? What conclusions can be drawn from the results? What is the novelty of the work and where does it go beyond previous efforts in the literature? Add the main findings and objective of the current study in the abstract.

5.      The paper should be carefully revised for punctuation, grammar, spelling mistakes and sentences structuring.

6.      Clear statements of the novelty of the work should also appear briefly in the Abstract and Conclusions sections.

 

 

Minor editing of English language required

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

I am not satisfied with the Figure 6, please add bias, transfer functions used, weights and so on in the network they appear. 

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx