Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Non-Invasive Glucose Prediction System Enhanced by Mixed Linear Models and Meta-Forests for Domain Generalization

Yuyang Sun, Panagiotis Kosmas

TL;DR

This work tackles the challenge of non-invasively predicting blood glucose by integrating NIR spectroscopy and mm-wave sensing. It introduces a Mixed Linear Model for principled feature selection and leverages Meta-forests for domain-generalization to handle inter-subject variability, achieving MAE = 17.47 mg/dL, RMSE = 31.83 mg/dL, and MAPE = 10.88% on unseen subjects. Key contributions include balanced data augmentation via Mix-up, statistically guided feature selection, and a DG-enabled predictive framework that improves generalization to new individuals. The results indicate a promising path toward personalized, non-invasive glucose monitoring with potential clinical impact, though broader validation and regulatory validation are needed.

Abstract

In this study, we present a non-invasive glucose prediction system that integrates Near-Infrared (NIR) spectroscopy and millimeter-wave (mm-wave) sensing. We employ a Mixed Linear Model (MixedLM) to analyze the association between mm-wave frequency S_21 parameters and blood glucose levels within a heterogeneous dataset. The MixedLM method considers inter-subject variability and integrates multiple predictors, offering a more comprehensive analysis than traditional correlation analysis. Additionally, we incorporate a Domain Generalization (DG) model, Meta-forests, to effectively handle domain variance in the dataset, enhancing the model's adaptability to individual differences. Our results demonstrate promising accuracy in glucose prediction for unseen subjects, with a mean absolute error (MAE) of 17.47 mg/dL, a root mean square error (RMSE) of 31.83 mg/dL, and a mean absolute percentage error (MAPE) of 10.88%, highlighting its potential for clinical application. This study marks a significant step towards developing accurate, personalized, and non-invasive glucose monitoring systems, contributing to improved diabetes management.

Non-Invasive Glucose Prediction System Enhanced by Mixed Linear Models and Meta-Forests for Domain Generalization

TL;DR

This work tackles the challenge of non-invasively predicting blood glucose by integrating NIR spectroscopy and mm-wave sensing. It introduces a Mixed Linear Model for principled feature selection and leverages Meta-forests for domain-generalization to handle inter-subject variability, achieving MAE = 17.47 mg/dL, RMSE = 31.83 mg/dL, and MAPE = 10.88% on unseen subjects. Key contributions include balanced data augmentation via Mix-up, statistically guided feature selection, and a DG-enabled predictive framework that improves generalization to new individuals. The results indicate a promising path toward personalized, non-invasive glucose monitoring with potential clinical impact, though broader validation and regulatory validation are needed.

Abstract

In this study, we present a non-invasive glucose prediction system that integrates Near-Infrared (NIR) spectroscopy and millimeter-wave (mm-wave) sensing. We employ a Mixed Linear Model (MixedLM) to analyze the association between mm-wave frequency S_21 parameters and blood glucose levels within a heterogeneous dataset. The MixedLM method considers inter-subject variability and integrates multiple predictors, offering a more comprehensive analysis than traditional correlation analysis. Additionally, we incorporate a Domain Generalization (DG) model, Meta-forests, to effectively handle domain variance in the dataset, enhancing the model's adaptability to individual differences. Our results demonstrate promising accuracy in glucose prediction for unseen subjects, with a mean absolute error (MAE) of 17.47 mg/dL, a root mean square error (RMSE) of 31.83 mg/dL, and a mean absolute percentage error (MAPE) of 10.88%, highlighting its potential for clinical application. This study marks a significant step towards developing accurate, personalized, and non-invasive glucose monitoring systems, contributing to improved diabetes management.
Paper Structure (22 sections, 6 equations, 4 figures, 7 tables)

This paper contains 22 sections, 6 equations, 4 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Mixed Linear Model in Statistical Analysis
  4. Fixed Effects
  5. Random Effects
  6. Meta-forests
  7. Methodology
  8. Data Preparation
  9. Data Processing
  10. Mix-up Augmentation
  11. Mixed Linear Model for Feature Selection
  12. Models Prediction
  13. Experiments and Results
  14. Experiment Setup
  15. Generalized Model Experiments Results
  16. ...and 7 more sections

Figures (4)

  • Figure 1: System Overview of the Prediction Model Workflow, illustrating stages from raw data input through '$S_{21}$ mm-wave' and 'Transmittance NIR' to feature selection via Mixed Linear Model Analysis, model prediction with Random Forests and Meta-forests, and final evaluation using MAE, RMSE, and MAPE metrics.
  • Figure 2: Estimated Coefficients of $S_{21}$ Parameters for mm-wave Frequencies from Mixed Linear Model Analysis. The lower and upper error bars delineate the 95% confidence intervals (2.5%, 97.5%) of estimated coefficients. The coefficient for each frequency reflects the estimated change in the response variable (normalized glucose level) for a one-unit increase in the predictor variable ($S_{21}$ parameter) while holding other variables constant. Red bars indicate frequencies with statistically significant associations ($p < 0.05$), selected for further study, while blue bars represent non-significant frequencies ($p \geq 0.05$), which were removed from subsequent analysis.
  • Figure 3: Pearson's Correlation Coefficients Between $S_{21}$ Parameters and Blood Glucose Values Across $N_{i}$ ($N_{i}, i = 1, 2, ..., 5$) Subjects.
  • Figure :