Machine Learning-Enhanced Non-Invasive Glucose Monitoring via a Compact Microwave CSRR Resonator Sensor

This paper presents a new, compact shape of a planar microwave resonator sensor for non-invasive blood glucose monitoring, enabling effective diabetes management. The proposed sensor is based on four cells of compact as irregularly curved Coupled Split Ring Resonators (CSRRs) printed on an affordable FR4 dielectric substrate. It functions by leveraging the interaction between biological tissues and microwave signals to detect glucose levels accurately. By transmitting microwave signals through the body and analyzing the S-parameters, the sensor identifies changes in the dielectric properties of blood, enabling the measurement of glucose concentrations. The magnitude and phase of S-parameters play a crucial role in providing a comprehensive analysis of glucose-induced variations. To validate the sensor’s performance, simulations use a phantom finger model, with laboratory data employed to verify these results, accounting for potential experimental variability. Sensitivity testing is conducted using glucose solutions with concentrations from 0 mg/dL to 350 mg/dL. Additionally, machine learning (ML) techniques are integrated to analyze sensor data, enhance glucose detection accuracy. The sensor can adapt to individual differences and environmental factors, improving the overall accuracy and reliability of the glucose monitoring system. Applying classification algorithms and signal processing methods, such as Discrete Wavelet Transform (DWT) decomposition, is used to classify microwave signals according to glucose levels more accurately. The integration of DWT significantly improves classification accuracy, achieving an 80% success rate across different glucose concentrations. The CSRR sensor design demonstrates a strong interaction between glucose samples and electromagnetic fields, resulting in a sensitivity of 10 MHz/(mg/dL) and 3dB/(mg/dL), alongside excellent detection capabilities.