Detection and classification of glucose solution concentration and blood sugar levels at the fingertip using a novel planar microwave sensor and deep learning techniques

This research introduces an omega-shaped resonator coupled split-ring resonator (OSR-SRR) microwave sensor designed for non-invasive glucose monitoring in solutions and fingertip samples. Enhanced with machine learning, the sensor detects glucose levels for non-diabetic and diabetic individuals. Fabricated via PCB processes, the sensor's performance was characterized by reflection coefficient (S11) and resonance frequency (Fr). Simulations analyzed the effects of skin and blood layer properties on the sensor's response. Linear regression assessed the impact of fingertip properties, while nonlinear regression predicted glucose levels from measurements. Experimental validation, using a vector network analyzer, included in-vitro and in-vivo methods with microfluidic channels, cylindrical tubes, and fingertip measurements for glucose concentrations between 0-200 mg/dL. The sensor achieved sensitivities of 282.20 x 10-3 dB/(mg/dL) for S11 and 0.56 MHz/(mg/dL) for Fr. Glucose level classification was performed using supervised learning with a multilayer perceptron (MLP) and unsupervised K-means clustering. Compact, cost-effective, and highly sensitive, the proposed sensor reliably measures glucose in both solutions and fingertip samples.