Noncontact and Noninvasive Detection of Glucose Concentration Using a Single-Port Microwave Sensor Coupled With Artificial Neural Networks

In this study, we introduce a novel approach for measuring glucose concentration using an innovative single-port microwave sensor coupled with artificial neural networks (ANNs) and unsupervised classification based on K-means clustering. The sensor is fabricated on a Dicadd 880 substrate, which has a relative permittivity of 2.2, a thickness of 1.6 mm, and a loss tangent of 0.0009, employing printed circuit board (PCB) lithography techniques. The reflection coefficient ( $S_{11}$ ) and the resonance frequency ( $F_{r}$ ) are recorded and serve as key features for data analysis and classification of glucose concentration using ANN. The proposed sensor operates based on a noncontact and noninvasive mechanism for detecting glucose concentrations, with a measurable range from 0 to 200 mg/dL. The operating frequency spans from 2 to 3 GHz. Experimental results indicate a clear correlation between $S_{11}$ , $F_{r}$ , and glucose concentration. The sensor’s sensitivity is measured at 0.064 dB/mg/dL and 0.133 MHz/mg/dL, respectively. To enhance classification performance, ANN was employed due to its capability to model complex and nonlinear relationships between $S_{11}$ , $F_{r}$ , and glucose concentration. Additionally, principal component analysis (PCA) was utilized for dimensionality reduction, and K-means clustering was applied for unsupervised classification. The analysis of glucose concentration using ANN combined with PCA and K-means clustering achieved the accuracies of 98.89% and 100%, respectively. This study establishes a foundation for future research by demonstrating that a noncontact, single-port microwave sensor can effectively differentiate glucose concentrations, offering a potential pathway for developing practical biomedical sensing applications.