Sensitivity Evaluation of a Dual-Finger Metamaterial Biosensor for Non-Invasive Glycemia Tracking on Multiple Substrates.

Accurate, non-invasive glucose monitoring remains a major challenge in biomedical sensing. We present a high-sensitivity planar microwave biosensor that progresses from a 2-cell hexagonal array to an 8-cell hexagonal array, and finally to a 16-cell double-honeycomb (DHC-CSRR) architecture to enhance field confinement and resonance strength. Full-wave simulations using Debye-modeled glucose phantoms demonstrate that the optimized 16-cell array on a Rogers RO3210 substrate substantially increases the electric field intensity and transmission response |S21| sensitivity compared with FR-4 and previous multi-CSRR designs. In vitro measurements using pharmacy-grade glucose solutions (5-25%) and saline mixtures with added glucose, delivered through an acrylic channel aligned to the sensing region, confirm the simulated trends. In vivo, vector network analyzer (VNA) tests were conducted on four human subjects (60-150 mg/dL), comparing single- and dual-finger placements. The FR-4 substrate (εr = 4.4) provided higher frequency sensitivity (2.005 MHz/(mg/dL)), whereas the Rogers RO3210 substrate (εr = 10.2) achieved greater amplitude sensitivity (9.35 × 10-2 dB/(mg/dL)); dual-finger contact outperformed single-finger placement for both substrates. Repeated intra-day VNA measurements yielded narrow 95% confidence intervals on |S21|, with an overall uncertainty of approximately ±0.5 dB across the tested glucose levels. Motivated by the larger |S21| response on Rogers, we adopted amplitude resolution as the primary metric and built a compact prototype using the AD8302-EVALZ with a custom 3D-printed enclosure to enhance measurement precision. In a cohort of 31 participants, capillary blood glucose was obtained using a commercial glucometer, after which two fingers were placed on the sensing region; quadratic voltage-to-glucose calibration yielded R2 = 0.980, root-mean-square error (RMSE) = 2.316 mg/dL, overall accuracy = 97.833%, and local sensitivity = 1.099 mg/dL per mV, with anthropometric variables (weight, height, age) showing no meaningful correlation. Clarke Error Grid Analysis placed 100% of paired measurements in Zone A, indicating clinically acceptable agreement with the reference meter. Benchmarking against commercial continuous glucose monitoring systems highlights substrate selection as a dominant lever for amplitude sensitivity and positions the proposed fully non-invasive, consumable-free architecture as a promising route toward portable RF-based glucose monitors, while underscoring the need for larger cohorts, implementation on flexible biocompatible substrates, and future regulatory pathways.