Non-Invasive Blood Glucose Estimation System Based on a Neural Network with Dual-Wavelength Photoplethysmography and...

Area of Science:

• Biomedical Engineering
• Medical Devices
• Biosensing Technology

Background:

• Conventional blood glucose monitoring methods are invasive and can cause discomfort.
• There is a need for noninvasive, accurate, and reliable blood glucose estimation systems.

Purpose of the Study:

• To develop and validate a noninvasive blood glucose estimation system.
• To assess the system's accuracy and robustness using dual-wavelength photoplethysmography (PPG) and bioelectrical impedance analysis.

Main Methods:

• Utilized dual-wavelength photoplethysmography (PPG) signals and bioelectrical impedance measurements.
• Extracted physiological features from PPG (mean, variance, skewness, kurtosis, standard deviation, information entropy) and bioelectrical impedance data (real/imaginary parts, phase, amplitude across 11 frequencies).
• Employed principal component analysis for feature extraction and a back-propagation neural network (BPNN) for blood glucose estimation.

Main Results:

• The system demonstrated high accuracy with a coefficient of determination (R²) of 0.997.
• Performance metrics included a mean squared error of 40.736, root mean squared error of 6.3824, and mean absolute error of 5.0896.
• Results fall within clinically accurate region A of the Clarke error grid analysis, validated on data from 40 volunteers.

Conclusions:

• The proposed noninvasive system effectively estimates blood glucose levels.
• The combination of PPG and bioelectrical impedance offers a promising, comfortable, and accurate alternative to invasive monitoring.
• The system's robustness and clinical accuracy support its potential for widespread adoption.