Adaptive Boosting Based Personalized Glucose Monitoring System (PGMS) for Non-Invasive Blood Glucose Prediction with...

Area of Science:

• Biomedical Engineering
• Medical Devices
• Machine Learning in Healthcare

Background:

• Accurate blood glucose monitoring is crucial for diabetes management.
• Current non-invasive methods often lack sufficient accuracy.
• Personalized calibration is needed to improve device performance.

Purpose of the Study:

• To develop and validate a personalized glucose monitoring system (PGMS).
• To enhance the accuracy of non-invasive glucose measurements.
• To reduce the error in predicted glucose values for improved patient outcomes.

Main Methods:

• Developed a PGMS integrating invasive and non-invasive sensors.
• Trained machine learning models using paired invasive and non-invasive data.
• Employed adaptive boosting (AdaBoost) for personalized error prediction models across glucose ranges.
• Calibrated models based on individual patient characteristics.

Main Results:

• Achieved a significant reduction in Mean Absolute Relative Difference (MARD) for predicted values (7.3% and 7.1%) compared to measured non-invasive values (25.4% and 18.4%).
• Clarke Error Grid Analysis (CEGA) showed 97-98% of predicted data in Zone A (clinically accurate).
• Demonstrated high accuracy and reliability across two independent datasets.

Conclusions:

• The developed PGMS effectively provides accurate, personalized, non-invasive glucose monitoring.
• The system's performance surpasses traditional non-invasive methods.
• PGMS holds potential for improving diabetes self-management and reducing complications.