Improving Multi-Sensor Non-Invasive Glucose Detection through AI: A Domain Generalization Approach.

Area of Science:

• Biomedical Engineering
• Data Science
• Medical Diagnostics

Background:

• Accurate glucose monitoring is vital for diabetes management and preventing complications.
• Inter-patient variability presents a significant challenge for non-invasive glucose monitoring systems.
• Existing methods struggle to generalize across diverse patient populations.

Purpose of the Study:

• To develop and evaluate a novel domain generalization approach, meta-forests, for accurate non-invasive glucose monitoring.
• To address the challenge of inter-patient heterogeneity in glucose level prediction.
• To enhance the interpretability of non-invasive glucose detection models.

Main Methods:

• Utilized a dataset of 54,280 data points from five subjects over 10 days.
• Employed a non-invasive system integrating near-infrared (NIR) spectroscopy, millimeter-wave (mm-wave) sensing, and temperature measurements.
• Applied meta-forests, an ensemble-based domain generalization technique, and incorporated Shapley additive explanations (SHAP) for model interpretability.

Main Results:

• Achieved an average root mean square error (RMSE) of 1.10 mmol/L.
• Attained a mean absolute percentage error (MAPE) of 10.32% in subject-specific experiments.
• Demonstrated accuracy comparable to state-of-the-art non-invasive glucose detection methods.

Conclusions:

• Meta-forests effectively address inter-patient heterogeneity in non-invasive glucose monitoring.
• The developed system offers a promising approach for accurate and reliable glucose level detection.
• Enhanced model interpretability through SHAP analysis provides valuable insights into the prediction process.