Noninvasive Blood Pressure Classification Based on Photoplethysmography Using K-Nearest Neighbors Algorithm: A Feasibility Study

Round 1

Reviewer 1 Report

The paper presents a feasibility study on the application of a weighted knn method to the problem of BP estimation from PPG, as an alternative to deep learning based approaches.

Deep learning methods are very effective, and several working examples of estimating BP (and other) information from PPG/EGC signals have been proposed so far. The authors claim that the motivations of their study are given by the "classification accuracy" and "time consumption" issues.
As the issue caused by the time required during training occurs only once, while prediction is often very fast, this point should be better motivated. Moreover, is not clear what is the problem related to the "classification accuracy". The authors should stress the above points.

The amount and quality of data are interesting, as the authors involved 219 participants. The experiments are convincing and the results are promising. The method is clear and well detailed.

Several relevant scientific works are missing in the state of the art:

Rundo, et al. An advanced bio-inspired photoplethysmography (PPG) and ECG pattern recognition system for medical assessment.

Rundo, et al. Advanced bio-inspired system for noninvasive cuff-less blood pressure estimation from physiological signal analysis.

Rundo, et al. Advanced Multi-neural System for Cuff-less Blood Pressure Estimation through Nonlinear HC-features.

McCombie, et al. Adaptive blood pressure estimation from wearable PPG sensors using peripheral artery pulse wave velocity measurements and multi-channel blind identification of local arterial dynamics.

Kurylyak Y., et al. A Neural Network-based method for continuous blood pressure estimation from a PPG signal.

Yan, Y.S., et al. Noninvasive estimation of blood pressure using photoplethysmographic signals in the period domain.

Teng, X.F., et al. Continuous and noninvasive estimation of arterial blood pressure using a photoplethysmographic approach.

Kim, J.Y., et al. Comparative study on artificial neural network with multiple regressions for continuous estimation of blood pressure.

Author Response

Dear Reviewer 1

Please see the attachment

Regards

Hendrana Tjahjadi

Author Response File: Author Response.pdf

Reviewer 2 Report

the paper can be interesting and published after the revision. The main criticism to the paper is the lack of clarity on the author contribution. What it is original here is not clear. The proposed structure for the classification of blood pressure (BP) in Fig.8 is typical structure for signal classification. For example, see the publication Rabcan J., Levashenko V. at al., "Application of Fuzzy Decision Tree for Signal Classification" - IEEE Trans on Industrial Informatics, 2019. What is difference with comparison of such algorithms for signal classification except the classifier? Why the k-NN is used as classifier for this problem? Is it causes by data specifics, classifier properties?

The questions for the result comparison in Table 4. The proposed method result is shown for data set of 219 subjects. Other results are shown for data set of 121 subjects. What is result of the proposed method for the data set of 121 subjects?

Other comments:

-the "medical" part of the introduction can be shorted.

- the figure 7 for trivial explaining of k-NN classification is not very important

- what does "-" mean in table 1 and 2?

Author Response

Dear reviewer 2

Please see the attachment

Regards

Hendrana Tjahjadi

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The authors addressed all the rised concerns.

Author Response

Dear Reviewer 1

 

Thank You very much

 

Best Regards

Hendrana

Reviewer 2 Report

Authors took into account some of the comments and modify the paper, but there are still some points that need to be clarify.

All modification in the paper should be marked to see they.  The difference of algorithm for signal classification and propose should be introduced in the paper.  The comparison for proposed algorithm and other should be implemented for similar data set. Therefore the argument about the investigation in other paper is not acceptable (Reviewer #2, Concern # 4:). Pls, implement the experiment for similar data set and add result to the paper. Fig.7 is modified, but in my point of view,  the paper with the original figure should be cited.  The answer for (Reviewer #2, Concern # 6) should be in text of paper.

Author Response

Dear Reviewer 2

Please see the attachment

Best Regards

Hendrana

 

Author Response File: Author Response.pdf

Round 3

Reviewer 2 Report

The paper can be accepted