Application of Artificial Neural Networks for Accurate Non-Invasive In Vivo Glucose Determination with Microwave Sensor

Microwave sensors offer precise and non-invasive real-time glucose monitoring for diabetes management. This study explores the integration of microwave sensors and artificial neural networks (ANNs) for non-invasive glucose determination. The setup involves a Vector Network Analyser across a frequency range of 0.015 GHz to 8 GHz. Dielectric measurements for in vivo blood glucose levels were conducted with an open-ended coaxial sensor, capturing data from the pulse, thumb, and finger, and compared against invasive glucose meters. The Levenberg-Marquardt algorithm is employed to train ANNs, enhancing predictive accuracy. Results indicate the optimal neural network configuration varies per dataset, with the pulse dataset achieving the highest regression (R) value of 0.9706. Sensitivity analysis highlights the significance of 'Frequency' and 'Dielectric Constant' as key predictors. This study validates the feasibility of microwave sensors and ANNs for accurate, non-invasive in vivo glucose monitoring, advancing diabetes management.