MXene–graphene terahertz metasurface sensor pairs sharp resonances with AI to detect glucose without blood

Researchers have unveiled a terahertz metasurface biosensor that combines an MXene–graphene hybrid nanocoating with deep-learning-assisted spectral analysis, aiming at a goal long sought in diabetes care: measuring blood glucose without drawing blood. The design, described in the Journal of Materials Science, was evaluated entirely through full-wave electromagnetic simulations in COMSOL Multiphysics, and the authors present it as a proof of concept for next-generation non-invasive glucose diagnostics rather than a clinically validated device.

The sensor architecture is built around a K-shaped resonator functionalized with Ti3C2Tx MXene, embedded within concentrically arranged gold and silver ring antennas on a silicon dioxide substrate. According to the study, this geometrically optimized configuration promotes intense electromagnetic near-field confinement and supports spectrally sharp, tunable surface plasmon resonances that respond strongly to perturbations in the local dielectric environment caused by an analyte such as glucose.

Performance metrics reported from the simulations are striking. The sensor achieved a refractive index sensitivity of 1000 GHz per RIU across a physiologically relevant index range of 1.335 to 1.347 RIU, a quality factor exceeding 14, and a figure of merit of 20 per RIU. The authors state that these metrics collectively position the design among the leading terahertz metasurface sensors reported to date, a comparison that reflects the competitive landscape of the field rather than an independent benchmark.

When glucose concentration was varied, the resonance frequency shifted systematically and monotonically, spanning from 0.719 to 0.714 THz. The design provides a usable tuning bandwidth of 30 GHz across the clinically significant concentration range, meaning the resonance moves in a predictable direction as glucose levels change, which is the behavior a sensing readout would rely on.

A distinctive feature of the work is the integration of machine learning beyond simple spectral readout. One-dimensional convolutional neural networks (1D-CNNs) were trained on the simulated transmission spectra to classify glucose concentrations, achieving an accuracy of 93%. The authors argue that this demonstrates machine learning integration substantially enhances analytical discrimination, allowing the platform to move beyond manual peak tracking toward automated quantification.

The choice of materials is central to the design.

The multiresonator arrangement—concentric gold and silver rings surrounding the functionalized K-shaped resonator—creates multiple resonant features in the terahertz spectrum. Multiple resonances can provide redundant or complementary readout channels, and the study reports that the resulting configuration supports sharp, tunable surface plasmon resonances highly sensitive to analyte-induced dielectric perturbations, the physical basis for label-free detection.

Because the platform relies on shifts in resonance frequency rather than chemical labels, reagents, or blood extraction, the authors describe it as ultrasensitive, label-free, and reagent-free, capable in principle of real-time analyte quantification. They also note that the design framework could extend beyond glucose to broader continuous metabolite monitoring applications, since any analyte that alters the local dielectric environment would in principle shift the resonance.

Important caveats remain. All reported figures—sensitivity, quality factor, figure of merit, resonance shifts, and the 93% classification accuracy—derive from simulations of transmission spectra, not from measurements on human subjects or physical device prototypes described in the available abstract. Translating such a design into practice would require fabrication of a physical device and experimental validation of the simulated performance, since all reported figures derive from simulations rather than measurements. The authors themselves frame the work as establishing a compelling proof of concept and a scalable design framework, not as a validated diagnostic tool.

If subsequent experimental work confirms the simulated performance, such platforms could contribute to painless, real-time diabetes management tools, an area of growing clinical demand that the authors identify as the motivation for the design.

Subject of Research: A terahertz metasurface biosensor using an MXene–graphene hybrid coating and 1D-CNN spectral analysis for non-invasive glucose detection.

Article Title: Multi-resonator MXene–graphene metasurface biosensor for non-invasive glucose detection with 1D-CNN optimization

Article References: Ochen, W., Ahmed, A. M., Elsayed, H. A., Abdelkawy, M. A., Mehaney, A., Rajakannu, A., Wekalao, J., & Muheki, J. (2026). Multi-resonator MXene–graphene metasurface biosensor for non-invasive glucose detection with 1D-CNN optimization. Journal of Materials Science. https://doi.org/10.1007/s10853-026-13807-4

Image Credits: AI Generated

DOI: 10.1007/s10853-026-13807-4

Keywords: terahertz, metasurface, biosensor, MXene, graphene, glucose monitoring, non-invasive, surface plasmon resonance, 1D-CNN, machine learning, COMSOL simulation, refractive index sensing