GlucoNet: A Hybrid Learning Methodology for Non-Invasive Blood Glucose Estimation from PPG

Area of Science:

• Biomedical Engineering
• Medical Devices
• Artificial Intelligence in Healthcare

Background:

• Rising global diabetes prevalence necessitates frequent blood glucose level (BGL) monitoring.
• Conventional BGL monitoring methods are invasive, painful, and unsuitable for continuous tracking.
• Non-invasive techniques, like photoplethysmography (PPG), are crucial for improved diabetes management.

Purpose of the Study:

• To develop and validate a non-invasive deep learning model for estimating blood glucose levels using PPG signals.
• To assess the model's accuracy and clinical applicability compared to existing methods.
• To provide a painless and continuous BGL monitoring solution.

Main Methods:

• Introduced GlucoNet, a deep learning architecture combining Convolutional Neural Network (CNN) and Bidirectional Long Short-Term Memory (BiLSTM) modules.
• Trained and validated the model on both public online and real-world datasets.
• Utilized PPG signals for spatial and temporal feature extraction to estimate BGL.

Main Results:

• Achieved high accuracy with a Mean Absolute Error (MAE) of 2.15 mg/dL and Root Mean Squared Error (RMSE) of 3.28 mg/dL.
• Demonstrated excellent performance with an R² value of 0.99.
• 100% of predictions fell within the clinically acceptable Clarke Error Grid zones A and B.

Conclusions:

• The proposed GlucoNet model offers a robust, non-invasive method for real-time BGL estimation using PPG signals.
• This technology enables painless, continuous monitoring, supporting early diabetes detection and management.
• GlucoNet represents a significant advancement in wearable health technology for diabetes care.