High sensitivity terahertz metasurface biosensor integrating plasmonic coupling and machine learning for wearable diabetes monitoring

This study presents a novel terahertz (THz) metasurface biosensor for highly sensitive and non-invasive glucose detection. Comprehensive finite element method (FEM) simulations performed using COMSOL Multiphysics demonstrate exceptional sensing performance, achieving a maximum sensitivity of 2000 GHz/RIU over a refractive index range of 1.335–1.347, corresponding to physiological glucose concentration variations. The proposed sensor exhibits a high figure of merit (FOM) of 44.444 RIU −1 , a quality factor (Q) of 6.244 and stable operation within the 0.27–0.31 THz frequency band with a tunability of 50 GHz. The sensing mechanism is driven by strong plasmonic coupling and enhanced electromagnetic field confinement arising from the hybrid integration of graphene, silver and gold resonators. To further validate the robustness of the sensing response, a machine learning–based predictive model is employed, yielding high prediction accuracy with R 2 values ranging from 84% to 100% across incident angles from 0° to 80°. Compared to conventional invasive glucose monitoring techniques, the proposed label-free optical biosensor offers superior sensitivity, real-time response and improved patient comfort. These results highlight the strong potential of the proposed THz metasurface platform for future wearable glucose monitoring and point-of-care diabetes management systems.