Machine learning enabled smart non-invasive glucose monitoring system using RF sensors

초록

This study presents a non-invasive blood glucose monitoring system that integrates an RF-based biosensor with machine learning techniques for enhanced accuracy and reliability. The sensor utilizes an Octagonal- Shaped Complementary Split Ring Resonator (OSCSRR) arranged in a honeycomb structure, operating at a resonance frequency of 2.45 GHz. Blood glucose levels ranging from 0 to 300 mg/dL were analyzed using the Cole-Cole model to determine variations in the reflection coefficient (S11). The sensor was fabricated on a Rogers RO4003C substrate and validated using a Keysight N5227A Microwave Network Analyzer, with a measured return loss deviation of 5 dB from simulated results. Machine learning algorithms, including Singular Value Decomposition (SVD), were employed to process dielectric characterization data and enhance the sensitivity in detecting glucose concentration variations. The trained model effectively distinguished glucose levels by identifying subtle changes in electromagnetic responses. The proposed approach demonstrates significant potential for non-invasive glucose monitoring, reducing patient discomfort while ensuring high sensitivity and accuracy.

키워드