Artificial Intelligence-Driven Mobile Platform for Thermographic Imaging to Support Maternal Health Care

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The research topic is worth investigating and the paper has good research merits. However, various parts require enhancement. Please consider my comments below for your revision.
Comment 1. Abstract:
(a) Provide numeric results for your work.
(b) Summarize the performance comparison between your work and the existing methods.
(c) Discuss the research implications.
Comment 2. Keywords: Some of the terms (deep learning and mobile) are too general. Please consider adding more terms to better reflect the scope of the paper.
Comment 3. Refer to the journal’s template to prepare the list of references.
Comment 4. As a standard research article, instead of a review-type article, citing 86 sources is not appropriate. Priority for removing references goes to references that are beyond the recent 5-year timeframe.
Comment 5. Section 1 Introduction:
(a) Correct “(L4, L5, S1)” to “(L4, L5, and S1)”.
(b) If the authors would like to provide examples, please ensure that at least two examples are provided. For example, “such as Convolutional Neural Networks (CNNs)” is insufficient.
(c) In three research contributions, the authors cited various existing methods. Please clarify if you followed the existing methods or if you modified the methods. In addition, provide numeric results and percentage improvement to justify your statement.
Comment 6. Section 2 Related Work:
(a) Add an introductory paragraph before Subsection 2.1.
(b) Ensure that the literature review provides a concise summary of the methodologies, results, and limitations of the existing methods. Your work should aim to address the limitations.
Comment 7. Section 3 Materials and Methods:
(a) Add an introductory paragraph before Subsection 3.1.
(b) Enhance the resolution of all figures. Zoom in on your file to (200%) to confirm that no content is blurred.
(c) Justify “standard resolution of 512 × 512 pixels”
(d) Figures 2 to 4: Provide the size and setting of each component. In addition, explain how you fine-tuned your model.
(e) The content related to “lightweight” was insufficient.
Comment 8. Section 4 Experimental Set-Up:
(a) Justify “Adam optimizer with a batch size of 36 for up to 60 epochs”.
(b) The ratio of 70:20:10 of the Train-valid-test validation needs to be justified.
Comment 9. Section 5 Results and Discussion:
(a) What are the results of fine-tuning each model?
(b) When there are multiple subfigures in a figure, please update the captions of the figure to describe each subfigure.
Comment 10. A performance comparison between your work and existing works is missing.

Author Response

See attached pdf

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Dear Authors

The paper explore an important area and the use of infrared thermal imaging is relevant. The paper is presented well, has a good literature review, methodology is is explained mostly well, the resulted are clearly outlined and discussed.

I am ware the paper already is detailed, however my recommendation is to provide the parameters of the AI models used to allow the methodology to be complete.

 

 

Author Response

See attached pdf

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

Thanks for the invitation to review this work. The manuscript presents a technically sound integration of mobile thermal imaging, deep learning segmentation (U-Net/ResUNet), and non-rigid registration (B-spline FFD) for LEA monitoring. Strengths include rigorous benchmarking of loss functions and mobile deployment via TFLite. Limitations are fairly acknowledged, though prospective validation remains pending.

 

1. Page 2, Line 25:"provide a straighforward feedback" → straightforward (spelling error).

Page 3, Line 40: "would constitute a significant advancement in obstetric anaesthesia" → obstetric anesthesia (spelling consistency; "anaesthesia" [British] vs. "anesthesia" [American]). Recommend consistent usage throughout (American English preferred in scientific publishing).

Page 5, Line 94: "collectively define the technical gap" → collaboratively define (more precise wording).

Page 5, Line 132: "accessibility, reproducibility,and objective non-invasive monitoring" → Missing oxford comma. Add comma: accessibility, reproducibility, and objective.

Text Clarity & FlowIntroduction (Page 3, Lines 15–16): "Plantar thermography, focusing on the soles of the feet, is particularly informative as the feet..." → Repetition of "feet" disrupts flow. Rephrase as: "...as this region...".

Abstract (Page 2, Line 26): "By integrating thermal imaging with deep learning and mobile deployment..." → Overuse of "platform." Consider: "...proposed system provides a practical...".

Limitations (Page 18, Line 602): "potential misalignments of even a few millimeters" → misalignments as small as.

2. Figure 9: Caption "Dermatome contour edge inference and registration" is ambiguous. Replace with "Warped dermatome contours after registration". Graphic labels lack anatomical specificity: Add "L4," "L5," "S1," "S2"to canonical/registered dermatomes.
3. Figure 10: Y-axis label "correlation" is vague. Specify: "Normalized Correlation".

X-axis: Add units "(# Iterations)".

4. Figure 4 (DeepLabV3+):Architecture diagram lacks decoder path detail. Mention in caption: "Illustrative schematic highlighting ASPP module; decoder omitted for brevity".

1. Text (Page 3) specifies "L4, L5, S1" dermatomes, but Figure 9 implies S2 inclusion. Clarify S2 role or reconcile inconsistency. Conclusion, Line 621: "robust quantitative performance" → Tone down. Use "competitive segmentation performance" (Dice=0.94→good but not infallible).

6. Section 3.1 uses "ThermalFeet"dataset, but later (Page 8) calls it "Feet Mamitas." Standardize to one name.
7. FFD Registration (Page 11):

Similarity metric S(∙) is underspecified. Explicitly state NCC is used.

Initialization strategy not described: Mention "Affine transformation for coarse alignment prior to FFD".

8. Page 6, Lines 82: "lightweight registration methods[...] suitable for mobile deployment" → Quantify "lightweight" (e.g., mean execution time/device specs). Contributions (Page 5, Lines 118–126): Fourth contribution "mobile deployment" rephrases third. Merge or reword uniquely.Dataset Attribution: Public dataset "Feet Mamitas" requires a formal citation ([77], Aguirre-Arango et al.).
9. Term Consistency:

"Deep learning" vs. "deep-learning" → Unify hyphenation (DL = adjective: deep-learning).

"Labor/Labour" → Use labor consistently (American English in text/bibliography).

10. Consider including citations of Exploration, 2024, 4, 20230146;

 

Author Response

See attached pdf

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Most comments have been properly addressed. I have several comments on the revised paper:
Comment 1. Abstract: Since multiple existing methods should be compared, the authors should state the percentage of improvement by the proposed method. In addition, other performance evaluation metrics should be reported (in addition to Dice).
Comment 2. The authors can consider adding more terms (at most 10) in the keywords because these help the readers to download and potentially cite the paper.
Comment 3. Regarding the research contributions, provide more performance evaluation metrics and ranges of improvement compared to the existing methods.
Comment 4. Tables 1 and 2: The metrics being reported should align with those in the experiment (in later sections).
Comment 5. Table 5: Why only t_0, t_1, t_5, t_10, t_15, and t_20 being reported? There are many more variables.
Comment 6. Figures 2 to 7: Details of each component are missing. Please provide the size/setting. The caption does not provide full details.

Author Response

See attached pdf

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

Thanks for the invitation to review this work. The authors have solved the previous concerns. The article is recommended for publication. 

Author Response

See attached pdf

Author Response File: Author Response.pdf