A Novel Model for Noninvasive Haemoglobin Detection Based on Visibility Network and Clustering Network for Multi-Wavelength PPG Signals

Comments 1: This study presents a novel approach to noninvasive hemoglobin (Hb) detection using visibility graph (VG) transformations and clustering-based feature extraction from multi-wavelength photoplethysmographic (MW-PPG) signals. The study introduces a classification-regression stacking framework that demonstrates superior predictive performance compared to traditional models. The authors validate their method using a clinical dataset comprising 608 signal samples from 152 volunteers and compare its accuracy to existing methods and devices. However, the study does have some drawbacks regarding the size of the data set and practicality. For example, it would make the paper more solid if authors can include more subjects across broader Hb ranges, including anemia patients, to improve model robustness. Also, how to make sure to use consistent contact pressure in a less controlled environment. In conclusion, I suggest this paper for “Accept with Minor Revisions” in Algorithms.

Response 1: First, we thank you for suggesting ‘acceptance with minor modifications’ for publication in Algorithms. We are pleased that you have acknowledged our new approach to non-invasive haemoglobin (Hb) detection from multi-wavelength photovoltaic volumetric pulse wave (MW-PPG) signals using visibility graph (VG) transformation and clustering-based feature extraction, and the superior predictive performance it demonstrates over conventional models. We agree with some of the shortcomings and issues you have raised and will expand the dataset and collect more data from anemic patients to optimize the algorithmic model and enhance its robustness of the model in the future. Regarding your question ‘How to ensure that consistent contact pressure is used in less controlled environments.’ We are currently using an app to visualize the contact pressure and remind the subject to keep it constant, in further improvements we will have a better solution to keep the contact pressure constant, thank you for your review and suggestions.

Comments 1: Do you have a data repository? If so, please provide access details or mention if the date is available upon request.

Response 1: Yes, we have a data repository. The data supporting our study is available upon request due to privacy concerns. If you are interested in accessing the data, please contact the corresponding author a liangyongbo@guet.edu.cn for further details.

Comments 2: In the introduction, it would be helpful to be more specific about the cited references. For example, instead of using vague terms like ‘good results’, ‘promising results’, or ‘favorable outcomes,’ try to include specific findings or quantitative outcomes to strengthen the argument.

Response 2：

Thank you very much for pointing out the problem, we agree with you and we have used a numerical quantitative narrative for the specific contributions to the literature review in the introductory section of this manuscript, with specific changes in lines 43 - 79. in line 43, we changed it to Appiahen et al. used 527 images of test subjects' palms to screen for anemia in 2023 and achieved classification results with 99.26% accuracy using a Bayesian classifier. and in line 46, we changed it to The results show that the prediction framework can achieve an MAE of 1.19 g/dL. And so on.

Comments 3: Formatting

Citation format: please ensure that the citation format is consistent throughout the manuscript. For example, Line 40, it should be ‘Appiahene et al.’ instead of  ‘Peter Appiahene, Justice Williams Asare, et al.’

Title capitalization: keep consistency in the capitalization of titles in the references section. e.g. reference # 1 (# 8, # 29) ‘Can Non-Invasive Spectrophotometric…’ should either have the first word capitalized or lowercase, depending on your style guide.

DOI consistency: references # 1-8 have CrossRef DOIs, but the rest of the references do not. Please check and update this.

Abbreviation: ensure that abbreviations such as SVM-LightGBM, MAE, and EMSE are spelled out the first time they appear in the abstract and elsewhere in the manuscript.

Response 3：Thank you very much for pointing out the problem, we agree with you and we have changed the error on line 41 into Appiahen et al. In addition, the case of references has been checked and modified line by line, with specific changes in lines 540 to 623. On the issue of abbreviations, we checked and made changes. The first occurrence of abbreviations was spelled out, with specific changes in lines 23-25 and line 418.

4. Response to Comments on the Quality of English Language

Point 1: Line 123-125 contain a sentence that repeats the previous one: ‘The FSR400 is the contact pressure sensor in this study…’. Please revise it for clarity.

Response 1: Thank you for your careful observation and for pointing out this careless error, we agree and have corrected the duplicate error, the specific changes are in lines 132-133.

Point 2: Consider using ‘gender’ to replace ‘sex’ for greater clarity and terminology.

Response 2: Thank you for this suggestion, we agree and have amended the term in lines 200 and 212, as well as the term in Figure 4.

Point 3: Line 283 Table 2 should be Table 1, and line 350 Table 4 should be Table 2. Pleae update the numbering accordingly.  

Response 3: Thank you for pointing out the error, which we agree with and have corrected, specifically in lines 305 and 373.

Point 4: Line 350-351, the sentence ‘According to Table 4, when the parameter is set to 0.1, the Modular parameter reaches a significant peak exceeding 0.6 before stabilizing within the range of 0.2 to 0.4.’ is unclear. Could you clarify what the ‘parameter of 0.1’ and ‘modular parameter of 0.6’ refer to in this context? More explanation would help.

Response 4: Thank you for raising this query, we agree and have rephrased the paragraph, which now reads: According to Table 2, when the modularity detection parameter in the Louvain algorithm is set to 0.1, that is, when extensive clustering is detected, the network modularity index reaches a significant peak of more than 0.6, and then the modularity detection parameter gradually increases, the network modularity index is gradually narrowed, and finally stabilized in the range of 0.2 to 0.4, which is the result of an index with a more significant modularity.

Point 5: Figure 8 (B) confusion matrix, consider adding the labels for true/false positives and true/false negatives. This will make it easier to interpret the results visually.

Response 5: Thank you for this suggestion, we agree and have added Figure 5 with the definitions of each row and column of the confusion matrix and the probability labels for true positives and false positives.