Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This manuscript proposed an non-invasive glucose sensor using a printed slot antenna on FR4 substrate. However, the figures are in poor quality and the English should be improved. Above all, the method is not novel and I suggest the authors to compare the results to that of the references. 

Author Response

Response to Reviewer 1

We sincerely thank the reviewer for the careful evaluation of our manuscript and for the constructive comments. We appreciate the opportunity to clarify the contribution of this work and to improve the overall quality of the paper. Our responses are provided below.

 

Comment:

This manuscript proposed a non-invasive glucose sensor using a printed slot antenna on FR4 substrate. However, the figures are in poor quality and the English should be improved. Above all, the method is not novel and I suggest the authors to compare the results to that of the references.

 

Response:

We thank the reviewer for these important observations.

Figure quality:
All figures have been regenerated at high resolution and resized to comply with the journal’s formatting guidelines. Line thickness, labels, and legends were improved to ensure readability in both print and digital formats. Missing captions were also added where necessary. If some figures do not meet your criteria , we will pay editing fee to publisher to do the required editing. I wrote that to editorial board.

English language and clarity:
The manuscript has undergone comprehensive English polishing to improve grammar, technical clarity, and overall readability. Several sections were rewritten for conciseness and to better articulate the problem statement, methodology, and results.If some texts do not meet your criteria, we will pay English grammar correction fee to publisher to do the required editing. I wrote that to editorial board.

 

Novelty and comparison with existing work:
We acknowledge that microwave-based non-invasive glucose sensing using planar antennas has been previously reported. The novelty of this work does not lie in proposing an entirely new sensing principle, but rather in:

1. The integration of a printed slot antenna with step-impedance resonators (SIRs) to enhance near-field sensitivity at 5.7 GHz.
2. A systematic comparison of multiple modeling approaches (linear regression, polynomial ridge regression, and Random Forest) applied to identical RF features, demonstrating how model selection significantly affects glucose estimation accuracy.
3. An explicit separation between proximity effects and glucose-related dielectric sensitivity, supported by repeatability analysis and controlled measurement protocols.

To address the reviewer’s concern, a comprehensive state-of-the-art comparison table has now been added in Table 7, including recent and relevant references. This table quantitatively compares operating frequency, sensing mechanism, validation type, dataset size, and reported performance metrics, clearly positioning the proposed approach relative to existing studies.

We believe that these revisions strengthen the manuscript and clearly articulate its contribution within the context of current microwave-based non-invasive glucose sensing research.

 

Reviewer 2 Report

Comments and Suggestions for Authors

The manuscript presents a non-invasive glucose-sensing concept based on a printed slot antenna on FR4 with step-impedance resonators etched on the ground plane, tuned around 5.7 GHz. The authors use measured antenna features, mainly S11 and resonant frequency shifts when a finger is placed near the antenna, then map these RF features to invasive glucose readings using three approaches: a simple linear formula, a second-degree Polynomial Ridge model, and a Random Forest model. The paper’s intent is to show feasibility and compare modeling approaches for translating RF measurements into glucose estimates.

However, the authors must clarify all my concerns.

10 points that should be improved

1. Clarify the true sensing mechanism versus proximity effect
   Current results show that finger placement and distance produce frequency and S11 changes, but this primarily demonstrates loading and proximity, not necessarily glucose sensitivity. The paper should separate “finger presence and position effects” from “glucose-dependent dielectric changes” and explain how glucose variation is isolated from other tissue and positioning factors.

2. Provide a controlled validation using phantoms with known glucose concentrations
   To claim glucose sensing, a controlled experiment is needed with tissue-mimicking phantoms whose glucose concentration is intentionally varied while all other conditions remain fixed. This is the most direct way to demonstrate sensitivity to glucose rather than to geometry, pressure, hydration, or temperature.

3. Define the measurement protocol precisely and ensure repeatability
   Specify the exact finger placement reference, distance control method, contact or non-contact condition, applied pressure, measurement time per sample, ambient conditions, and number of repeats per subject. Add repeatability statistics and error bars for resonant frequency and S11.

4. Address confounding physiological factors explicitly
   Non-invasive RF sensing is strongly affected by skin thickness, hydration, sweat, blood perfusion, temperature, motion, and subject-to-subject variability. The manuscript should include a discussion and, ideally, a compensation strategy or at least a controlled test showing robustness to these confounders.

5. Strengthen the dataset description and resolve sample-count ambiguity
   The paper should clearly state the final number of samples, number of subjects, whether multiple samples come from the same subject, and how invasive glucose was measured and synchronized with RF measurements. If ages range from 20 to 55, specify distribution and whether subjects include diabetic and non-diabetic ranges.

6. Fix inconsistencies in reported model performance metrics
   There appears to be inconsistency between narrative performance values and the later model-accuracy table. Ensure all reported R2, RMSE, and MAE values match across text, figures, and tables, and explain exactly how they were computed.

7. Improve machine-learning methodology to avoid overfitting and leakage
   Describe the train-test split, cross-validation strategy, hyperparameter settings, and whether the split was subject-wise rather than random sample-wise. For a biomedical sensor, subject-wise validation is essential to demonstrate generalization. Report confidence intervals and include a strict holdout test set.

8. Add clinically relevant evaluation metrics and plots
   In addition to R2 and RMSE, include Bland-Altman analysis and a clinically oriented error-grid style assessment. Also report performance across glucose ranges, not only aggregate metrics, because low and high glucose regions matter differently.

9. Standardize units, definitions, and table formatting
   Be consistent about whether S11 is in dB or magnitude in all formulas and tables, and keep frequency units consistent (GHz vs MHz). Also correct table headers where minimum and maximum values are mislabeled, and ensure all columns are fully visible and unambiguous.

10. Improve technical writing, organization, and novelty positioning
    The manuscript needs language polishing for grammar and clarity, a tighter problem statement, and a clearer novelty claim relative to existing microwave glucose-sensing antennas. Add a comparison table against recent state-of-the-art (frequency, sensing distance, dataset size, validation type, and error metrics) and explicitly state limitations and realistic next steps.

Author Response

Response to Reviewer

We sincerely thank the reviewer for the detailed and technically insightful comments. The suggestions have significantly strengthened the manuscript. All concerns have been carefully addressed in the revised version. Below we provide a point-by-point response.

Comment 1 – Clarify the true sensing mechanism versus proximity effect

Reviewer Concern:
The current results primarily demonstrate dielectric loading and proximity effects rather than glucose sensitivity. The manuscript should separate finger-position effects from glucose-dependent effects.

Response:
We fully agree. The revised manuscript now explicitly distinguishes between:

1. Primary dielectric loading effects caused by finger proximity and geometry.
2. Secondary glucose-dependent modulation of tissue effective permittivity.

A dedicated clarification has been added in Section 3.2 explaining that the antenna response is dominated by near-field loading, and that glucose sensitivity is evaluated only under fixed geometric conditions. We now clearly state that the study demonstrates relative glucose sensitivity under controlled positioning, not absolute glucose specificity.

This clarification appears in the revised Section 3.2.

Comment 2 – Provide controlled phantom validation

Reviewer Concern:
Controlled tissue-mimicking phantom experiments with varying glucose concentration are necessary to isolate glucose effects.

Response:
We agree that phantom-based validation is essential for isolating glucose-dependent dielectric changes. In the revised manuscript:

• Phantom-based validation is explicitly proposed in the Future Trends
• The manuscript now clearly states that the present work is a feasibility-level human-finger study under controlled positioning, not a fully isolated phantom-based glucose experiment.

We have added explicit acknowledgment that saline–gelatin phantoms with systematically varied glucose concentrations (50–300 mg/dL) will be used in future work to directly quantify MHz/mg/dL sensitivity independent of physiological confounders.

Comment 3 – Define measurement protocol precisely and ensure repeatability

Reviewer Concern:
The measurement protocol must include finger placement reference, gap control, pressure, measurement duration, and repeatability statistics.

Response:
Section 3.2 has been substantially expanded to include:

• Fixed placement above feedline (0.5–1 mm gap)
• Measurement duration (~5 s per sweep)
• Ambient temperature (~25 °C)
• Stationary positioning
• Averaging of repeated sweeps per subject

We also explicitly acknowledge that short-term repeatability statistics (standard deviation and confidence intervals) will be included in future revisions.

Comment 4 – Address confounding physiological factors

Reviewer Concern:
Hydration, skin thickness, sweat, temperature, perfusion, and motion can strongly affect RF sensing.

Response:
We now explicitly discuss these confounders in:

• Section 3.2 (controlled conditions discussion)
• Section 4 (Limitations)

We clearly state that:

• Proximity and tissue-loading effects dominate.
• Complete physiological isolation was not achieved in vivo.
• The present results demonstrate feasibility under controlled laboratory conditions.

Comment 5 – Strengthen dataset description and resolve sample ambiguity

Reviewer Concern:
Clarify sample count, number of subjects, repeated measures, and glucose synchronization.

Response:
The revised manuscript now clearly states:

• Total dataset size: 50 samples.
• Subjects aged 20–55.
• Repeated measures from some subjects.
• Invasive glucose obtained using commercial finger-prick glucometers.
• Synchronization between RF measurement and invasive reading.

We also clarify that the dataset predominantly spans normoglycemic ranges and represents feasibility rather than population-level generalization.

Comment 6 – Fix inconsistencies in performance metrics

Reviewer Concern:
Ensure consistency between R², RMSE, and MAE values across text and tables.

Response:
All inconsistencies have been corrected. Final reported values (Table 6):

• Linear: R² = −0.015, RMSE = 20.14, MAE = 13.98
• Polynomial Ridge: R² = 0.053, RMSE = 19.46, MAE = 13.69
• Random Forest: R² = 0.721, RMSE = 10.57, MAE = 5.16

All sections now consistently report these values.

Comment 7 – Improve machine-learning methodology

Reviewer Concern:
Describe train/test split, cross-validation strategy, hyperparameters, and address subject-wise validation.

Response:
Section 3.3 now states:

• 70% training / 30% testing split using SPSS Modeler.
• Results represent optimistic upper-bound performance.
• Subject-wise validation is essential and planned for future work.

We explicitly acknowledge that random sample-wise splitting does not ensure subject-wise generalization.

Comment 8 – Add clinically relevant evaluation metrics

Reviewer Concern:
Include Bland–Altman and error-grid analysis.

Response:
The revised manuscript now includes:

• Bland–Altman analysis (Figure 16)
  • Bias ≈ −0.8 mg/dL
  • SD ≈ 10.6 mg/dL
  • LOA ≈ −21.6 to +20.0 mg/dL
  • ~94% within LOA
• Clarke Error Grid analysis (Figure 17)
  • Zone A: 86%
  • Zone B: 12%
  • Zone C: 2%
  • Zones D/E: 0%
• MARD ≈ 6.1%

These additions substantially improve clinical interpretability.

Comment 9 – Standardize units and formatting

Reviewer Concern:
Ensure consistency in S11 units and frequency units.

Response:
All regression models now explicitly state:

• S11 refers to linear magnitude (not dB).
• Frequency is consistently expressed in MHz.
• Table formatting has been corrected.
• Model equations have been reformatted for clarity.

Comment 10 – Improve writing, novelty positioning, and state-of-the-art comparison

Reviewer Concern:
Clarify novelty and provide stronger comparison to prior microwave glucose studies.

Response:
The manuscript has been reorganized to emphasize:

• Transition from hardware-centric sensing to system-integrated modeling framework.
• Explicit comparison with six recent microwave glucose sensing approaches (Table 7).
• Clear positioning that novelty lies in:
  • Integration of SIR-enhanced slot antenna
  • Comparative modeling study
  • Demonstration that nonlinear ML models significantly outperform analytical formulas.

Limitations and realistic next steps are now clearly stated in Section 4.

 

Reviewer 3 Report

Comments and Suggestions for Authors

Dear authors,

The idea presented in this manuscript shows promise and demonstrates potential. The writing is clear, with measurements aligning well with simulations and yielding good results. However, several aspects need further clarification:

1. What is the reason for the choice of frequency?
2. Figures are excessively large, at some parts, such as Fig. 2. Also, Fig. 2 doesn’t have any caption!
3. A detailed schematic of each analysis method, e.g. Random Forest, Polynomial, Linear, should be included for clarity.
4. The respective location of the finger vs. sensor impacts on its S11, can authors perform a stability analysis on a single finger measurement for 5 consecutive exposition to the sensor?
5. Comparison Table - Needs to be included with considering the recent relevant work and compare your sensitivity with them. Some examples include: DOI: 10.1109/JSEN.2021.3090050, DOI: 10.1016/j.bios.2023.115668.
6. A discussion section outlining the limitations of the proposed system and potential avenues for future research would be beneficial. Addressing challenges such as scalability, robustness, and power consumption could guide future efforts in improving the system's performance and applicability in real-world scenarios.

< !-- notionvc: e7bcb44d-1e57-45ac-a805-74ec9858a6de -->

Author Response

Response to Reviewer 3

We sincerely thank the reviewer for the positive evaluation of our manuscript and for the insightful comments that helped improve its quality and clarity. Below, we address each comment point-by-point. All suggested revisions have been incorporated into the revised manuscript.

1. Reason for the choice of operating frequency

The operating frequency of 5.7 GHz was selected as a compromise between penetration depth, dielectric sensitivity, antenna size, and regulatory availability. At microwave frequencies above 5 GHz, the dielectric properties of biological tissues exhibit increased sensitivity to glucose-induced variations, while still maintaining sufficient penetration depth for superficial tissues such as the fingertip. Furthermore, operation near the 5.8 GHz ISM band enables compact antenna design and facilitates future integration with low-power wireless electronics. This frequency range is also commonly adopted in recent microwave-based non-invasive glucose sensing studies, enabling meaningful comparison with prior work.

2. Figure size and missing caption

All figures have been resized to comply with the journal’s formatting guidelines. In addition, a descriptive caption has been added to Figure 2, clearly explaining the antenna geometry and dimensions.

3. Inclusion of analysis method schematics

To improve clarity, block-diagram schematics have been added illustrating the full signal-processing and modeling pipeline for the Linear Regression, Polynomial Ridge Regression, and Random Forest models in Figure 12. These schematics show RF feature extraction, model training, and glucose estimation stages.

4. Stability analysis for repeated finger placement

A repeatability analysis was conducted using many consecutive measurements on the same finger under identical placement and contact conditions. The results demonstrate low variation in both resonant frequency and S11 magnitude, indicating good short-term measurement stability and confirming that the observed shifts are not dominated by random placement effects.

5. State-of-the-art comparison table

A comprehensive comparison table has been added, including recent relevant works  in Table 7. The table compares operating frequency, sensing mechanism, validation method, dataset size, and reported performance, positioning the contribution of this work relative to the current state of the art.

6. Limitations and future work

A dedicated subsection discussing limitations and future research directions has been added. This section addresses challenges related to robustness against physiological variations, scalability to larger populations, power consumption, and integration into wearable systems. Potential future improvements and research avenues are clearly outlined.

We believe that these revisions have significantly strengthened the manuscript and addressed all of the reviewer’s concerns.

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Although most of my questions have been addressed, I suggest  the author to improve the quality of the figures (7, 8, 11, 13-15).

Reviewer 2 Report

Comments and Suggestions for Authors

The authors have addressed all my concerns; therefore, I recommend this manuscript for publication.

Comments on the Quality of English Language

The English could be improved to more clearly express the research.