Eating Event Recognition Using Accelerometer, Gyroscope, Piezoelectric, and Lung Volume Sensors

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The Authors have submitted a concise manuscript focusing on improvements in detecting food intake events that is a very important aspect in the diagnosis of eating disorders. They explore the possibility to enhance the known methods by using complex sensors system comprising an accelerometer, gyroscope, respiratory inductance plethysmography, and piezoelectric transducers. The proposed approach allows to create the model of the recognition process on the basis of detecting three kinds of activities: eating gestures, chewing and swallowing food. While the presented model and the achieved results of the calculations may not constitute groundbreaking scientific reports; nevertheless, they demonstrate a well-planned and executed research experiment. I have to admit that all components of the manuscript are included (although in minimal form):

- very short introduction and literature review; I think it should be supplemented;

- accurate theoretical explanation;

- calculations based on experimental measurements; I think that the manuscript should be supplemented with measurements obtained from sensors;

- in my opinion the “Summary” Section should be provided.

Generally, the paper is well written. Nevertheless, the authors have committed some little editorial mistakes.

Additional remark:

The experiment was performed solely in controlled settings. I recommend further testing the system in real-life scenarios to assess its performance in actual human behaviours. If the proposed solution cannot be properly found in real-world situation, then creating such the theoretical model may lose its practical significance.

Although the method of applicating the Huawei watch and the RID sensor is clearly determined by the type of used devices, additional explanations - e.g. in the context of previous publications – are needed in relation to the selected location of the piezoelectric sensor.

The obtained results of F1-score are worse or comparable than those reported in the referenced publications. Although there is a clear improvement in the results when using all four sensors, the accuracy achieved in recognizing eating events is comparable to other studies despite the use of the much more complicated measurement system. Therefore, this issue should be discussed in more detail.

There are also no considerations for cases of chewing and swallowing that are not directly related to food intake. The question arises: Is the presented model resistant to such events? In my opinion, it should be, given the multiplication of sources of features for the SVM model. However, this would need to be demonstrated.

Comments on the Quality of English Language

Minor editing of English language required.

Author Response

Dear Reviewer,

I want to express my sincere gratitude for the time and effort you dedicated to reviewing my research paper. Your feedback and comments were very valuable.

I find you made valid points and took your time to address them. I have strived to provide clearer explanations and have expanded on certain aspects to improve the overall quality of the paper. Your constructive criticism is much appreciated. Attached you will find the manuscript as a pdf with changes highlighted in yellow.

Regarding the introduction: The introduction has been expanded (see highlighted text), with more information on cited literature.

For clarity, I have added 2 x 2 figures showing measurement data of the smartwatch (gyroscope) and the RIP sensor. The gyroscope figures give insight to what differences one might see between the two different types of eating gestures. The RIP figures show two breathing cycles before and after normalisation such that the reader is shown what RIP data looks like and what effect normalisation has on breathing cycles.

Regarding the controlled setting in which this experiment was performed, I added new text to the discussion, specifically the last two paragraphs.

I added 2 sources underlining the piezoelectric sensor measuring location.

Regarding the F1-scores compared to the referenced literature: The aim of this research was primarily to find out if combining sensors improved eating event detection. The added value therefore does not lie in reaching the highest test scores, but more in the combining of sensors. Next to this, each study has different parameters such as types of food, amount of food, degree of free-living conditions and types of sensors used. I added a paragraph in the discussion about this (3rd paragraph).

About chewing and swallowing actions that are not directly related to food intake: reading out loud and other jaw movements were non-eating events for chewing and regular (non-eating) swallows and irregular breathing patterns were non-eating events for swallowing. I added the following text for more clarity:

For the classification of swallowing, the non-eating events are regular swallows unrelated to eating and irregular breathing due to other actions performed.

AND

Testing in free-living conditions will add more varied non-eating events, introducing additional events similar to the eating events. However, due to the multimodal nature of this setup with three sensors focusing on different steps of the dietary cycle, it should be easier to distinguish these non-eating events from eating events.

I hope I have been able to clarify your points of attention and improvement. Thank you for your time.

Kind regards,

Sigert Mevissen

(In consultation with the other authors)

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

This research paper focused on developing an automatic dietary monitoring (ADM) sensor system that includes a smartwatch with an accelerometer and gyroscope for detecting eating gestures, a piezoelectric sensor for chewing detection, and a respiratory inductance plethysmographic (RIP) sensor for food swallowing detection. This system aims to provide insights into eating behaviors, allowing the simplification and objectification of eating habit monitoring.

While this study seems to be interesting and combining multiple sensors could have the potential of accurately detecting eating events, the results lack a supportive conclusion. However, to enhance the quality and clarity of the paper, the following points should be addressed:

1. 1. In Figure 2: The abbreviation "RIP" should be consistently capitalized throughout the document to avoid confusion. What does BP stand for?
2.  

2. In Figure 3, additional markers or annotations along the horizontal bar is needed to improve readability and interpretability.3. The performance of the swallowing detection seems suboptimal. It is crucial to discuss the reasons for this performance and potential improvements or alternative methodologies.4. Figure 5 shows different combinations of sensors were used with different feature vector size. It is unclear whether the result is due to the change of combination or feature vector size.
5. The limited sample size of six subjects is a significant constraint on the study's applicability. The paper should discuss the implications of this limitation on the generalizability of the results and propose strategies for future research to validate the findings with a larger cohort

Author Response

Dear Reviewer,

I want to express my sincere gratitude for the time and effort you dedicated to reviewing my research paper. Your feedback and comments were very valuable.

I find you made valid points and took your time to address them. I have strived to provide clearer explanations and have expanded on certain aspects to improve the overall quality of the paper. Your constructive criticism is much appreciated. Attached you will find the manuscript as a pdf with changes highlighted in yellow.

Regarding abbreviations and clarity of Figure 2: The abbreviations have been addressed, both in this figure and the rest of the text. Figure 2 has been explained in more detail, in the caption and the text.

Regarding Figure 3: Section 2.3 feature vectors have been re-ordered and rewritten. I added a figure (figure 8) for clarity, and rewritten caption of figure 3 (now figure 9). I think it should be much clearer now.

Regarding swallow detection performance: the reasons for this performance and potential improvements are now discussed at the end of the 4^th paragraph of the discussion.

Regarding Figure 5 and the effect of the feature vector size: The primary study outcome was to find out to what extent different combinations of sensors improved the overall classification scores. The feature vector length was a mere variable in this process. To emphasize this, I added text in the introduction and results (captions), and I rearranged the results to show the sensor combinations results before the feature vector size results.

Regarding the limited sample size: I added some text in the last paragraph of the discussion explaining the implications and strategies for further research.

I hope I have been able to clarify your points of attention and improvement. Thank you for your time.

Kind regards,

Sigert Mevissen

(In consultation with the other authors)

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

- In Section 1. Introduction: Highlight the research contributions of this work and the limitations of the existing works.

- Figure 1 needs further explanation > Define each block and sub-block in detail, e.g., define the Eating stage, then the preparation and cleaning steps. What do cyclic arrows depict?

- Add “Literature Review” section.

- Why did authors need to collect their own dataset? Why did not use any of the existing datasets? Discuss the limitations of existing datasets and the importance of the collected dataset over others.

- Section 2.1. Experimental setup: Add some figures (of different subjects wearing sensors on different body parts) to better illustrate the Experimental Setup. For privacy reasons, authors may hide the faces of the subjects and keep the main focus on worn sensors. Also, address the following comments about the dataset:

                - Mention: how many samples were collected.

                - Add a table of activities with the duration of each performed activity.

                - Describe the age, gender, and demographic info of the subjects from which the dataset is collected.

- Section 2.2. Data Processing: Sensors used in this work, have different frequency rates, clearly explain how they are adjusted.

- Figure 2: Explain why did the authors used different window sizes?

- Section 2.3. Feature vectors: needs to have major improvements.  Explain the details of how the feature vector is calculated.

- Figure 3: Explain it in a clear and easily understandable manner to make it reproducible.

- Section 2.4. Classification: Why did the authors only use the SVM model?

- Add Conclusion Section.

- There exist some grammatical mistakes, unclear sentences, and formatting issues. I cannot go into details here. However, authors must carefully fix these issues.

Comments on the Quality of English Language

There exist some grammatical mistakes, unclear sentences, and formatting issues. I cannot go into details here. However, authors must carefully fix these issues.

Author Response

Dear Reviewer,

I want to express my sincere gratitude for the time and effort you dedicated to reviewing my research paper. Your feedback and comments were very valuable.

I find you made valid points and took your time to address them. I have strived to provide clearer explanations and have expanded on certain aspects to improve the overall quality of the paper. Your constructive criticism is much appreciated. Attached you will find the manuscript as a pdf with changes highlighted in yellow.

Regarding the introduction: The introduction has been expanded (see highlighted text), with more information on cited literature.

Regarding Figure 1: The figure has been simplified. In the text above the figure, I added some text to explain the dietary cycle.

Regarding the introduction: The introduction has been expanded (see highlighted text), with more information on cited literature.

Regarding collecting a new data set: I added a paragraph in the introduction (from line 76)

Regarding section 2.1: I have included a figure explaining the sensor setup in general, as well as two images showing how the sensors are worn. Next to this, a table with subject characteristics was included. Finally, a table with activities performed during the experiment and their (approximated) duration is included.

Regarding the point on section 2.2: This is addressed in the last paragraph of section 2.2.

Regarding the different window sizes: I emphasized the window sizes are based on windows in previous work. Added a sentence second paragraph of 2.3 feature vectors explaining where the difference in window sizes comes from.

Regarding the point on clarity of section 2.3 and the following comment on figure 3: Section 2.3 feature vectors have been re-ordered and rewritten. I added a figure (figure 8) for clarity, and rewritten caption of figure 3 (now figure 9). I think it should be much clearer now.

Regarding the utilisation of SVM: I added sources to section 2.4 explaining it was based on previous work

I added a conclusion, fixed grammar errors and rephrased unclear sentences.

I hope I have been able to clarify the points of attention and improvement. Thank you for your time.

Kind regards,

Sigert Mevissen

(In consultation with the other authors)

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Thanks to the authors for their efforts. The paper has been improved, and questions have been addressed. In my opinion, the paper can be published in Sensors in its present form. The authors answer my comments to a satisfactory extent. The manuscript has been significantly modified, and I believe it is now much more suitable for publication. I have no further comments.

Comments on the Quality of English Language

In general, the paper is well-written. Nevertheless, the authors have committed some minor editorial and grammatical mistakes.

Author Response

Dear Reviewer,

Thank you for taking the time and effort to review the manuscript again. I am delighted to hear that you consider it worthy of publication.

Kind regards,

Sigert Mevissen

Reviewer 2 Report

Comments and Suggestions for Authors

The authors have thoroughly addressed the questions and enhanced the manuscript's quality. However, I recommend relocating some of the figures to supplementary material to streamline the main text

Author Response

Dear reviewer,

Thank you for taking the time and effort to review the manuscript again. I am delighted to hear that you think the manuscript’s quality is enhanced.

We have taken another look at the figures in the manuscript. The first manuscript contained 5 figures. The other two reviewers requested more figures to provide a more complete picture of all steps taken within the study. Deleting them now would mean disregarding their earlier comments. Next, we think all figures now included are required for full understanding and therefore we decided not to delete any figure. I hope you can agree with us.

Kind regards,

Sigert Mevissen

Reviewer 3 Report

Comments and Suggestions for Authors

The paper has been improved as compared to the previous version. However, it still needs more improvements to make it publishable. 

Authors did not address the following key issues previously highlighted:

 

- In Section 1. Introduction: Highlight the research contributions of this work and the limitations of the existing works.

- Add “Literature Review” section.

 

Also, the existing references are old. The authors must review and include some latest research articles.

Comments on the Quality of English Language

 Minor editing of English language is required. Keep the sentences precise, and simple.

Author Response

Dear Reviewer,

I want to thank you again for commenting on the proposed research paper. I have tried my best to address your comments.

• In Section 1. Introduction: Highlight the research contributions of this work and the limitations of the existing works.
• Add “Literature Review” section.

I have restructured the introduction into sections Background, Related work and Proposed sensor system. In the Proposed sensor system section I addressed the limitations of current work and how this proposed sensor system contributes to tackling those limitations. This can be seen in the highlighted part of the section. I also added some (more recent) references in the Related work section.

• Also, the existing references are old. The authors must review and include some latest research articles.

This is a valid point. I reviewed all articles and replaced or supplemented the old references if newer references could be found. 6 changes were made in total, mainly in the background section.

• Comments on the Quality of English Language: Minor editing of English language is required. Keep the sentences precise, and simple.

I have re-evaluated the complete research paper with a native speaker. I have not highlighted these changes, because many changes were only minor or there are no new parts (merely restructuring the sentence).

I hope I have satisfactorily addressed your comments.

Kind regards,

Sigert Mevissen (in consultation with the other authors)