Machine learning-optimized terahertz biophotonic biosensor for label-free non-invasive glucose monitoring using hybrid two-dimensional materials

Wearable diabetes management requires glucose sensing technologies that combine high sensitivity, rapid analysis, and seamless integration within Internet of Medical Things (IoMT) ecosystems. To address these challenges, this work presents an artificial intelligence (AI)-driven terahertz (THz) biophotonic biosensor based on a miniaturized spiral-resonator architecture incorporating a hybrid graphene–black phosphorus (BP)–vanadium dioxide ( \(\textrm{VO}_2\) ) functional interface. The complementary electrical characteristics of these active materials are jointly exploited to reinforce electromagnetic energy localization and maximize the interaction between the confined THz field and glucose-induced dielectric perturbations. The electromagnetic behavior of the sensing platform is investigated using the Wave Concept Iterative Process (WCIP), whose accuracy is verified through analytical validation, while computational benchmarking against the finite-difference time-domain (FDTD) technique demonstrates a substantial reduction in simulation time. The proposed biosensor exhibits a continuous resonance displacement from 2.10 THz in air to 2.45 THz at a glucose concentration of 500 mg/dL. Over the investigated concentration range (80–500 mg/dL), the sensitivity increases from 0.294 to 0.921 THz/RIU with air considered as the reference medium. To accelerate device optimization, a deep neural network (DNN) surrogate is trained using WCIP-generated data to establish the nonlinear relationship between the material parameters and the sensing performance. The developed model delivers excellent predictive capability, achieving an MSE of \(2.3\times 10^{-4}\) , an MAE of 0.004, and an \(R^2\) value of 0.999. Guided by the trained DNN, the optimized biosensor reaches a maximum sensitivity of 1.70 THz/RIU. These results demonstrate that the proposed sensing platform combines high predictive accuracy, computational efficiency, and material reconfigurability, making it a promising candidate for next-generation wearable glucose monitoring and IoMT-enabled biomedical diagnostic systems.