Artificial Intelligence Assisted Measurement of Glucose, Sodium, and Potassium Concentrations in Diluted Aqueous Solutions Using Microwaves

This letter proposes an artificial intelligence (AI)-driven noninvasive and contactless microwave sensor capable of determining the composition of aqueous solutions containing three different components: glucose, sodium (Na+), and potassium (K+). The sensing element is a one-port microstrip transmission line loaded with a pair of unequal three-turn complementary spiral resonators (CSRs) etched in the ground plane, as well as a capillary that drives the liquid under test (LUT) to the sensing region (i.e., the slots of the CSR). The two CSRs provide a rich frequency response (reflection coefficient), with many singularities (magnitude notches and phase jumps) over a broad frequency band that are necessary to selectively determine the concentration of the different solute components by virtue of their different dispersive behavior. The complete sensor includes the necessary mechanical accessories to implement an automated system with a liquid pump as well as temperature and fluid control. The system analyzes the renormalized S11 response to quantify the variations generated by the different components of the solution, demonstrating a high capacity of detecting their presence and reliably predicting their concentrations. A convolutional neural network (CNN) with a multilayer perceptron (MLP) maps the renormalized reflection coefficient spectra to solute concentrations. Validated on binary to quaternary mixtures, the method yields mean absolute errors of 8.2 mg/dL (glucose), 6.8 mg/dL (Na+), and 1.2 mg/dL (K+), enabling real-time quantification in complex solutions. This modular approach supports scalable dataset generation and adaptable AI training pipelines for other solutes and liquid matrices.