Machine-learning-assisted validation of a graphene-based terahertz metasurface for simulated hemoglobin sensing

This study presents a graphene-based metasurface sensor designed for haemoglobin detection within the terahertz (THz) frequency regime. Numerical simulations were performed using COMSOL Multiphysics, which employs the finite element method (FEM) to resolve the electromagnetic field distributions and quantify the sensor's spectral characteristics. Under these optimized conditions, the proposed sensor exhibits a global sensitivity of 225 GHz/RIU (R² = 0.90), a quality factor of 14.358, a figure of merit of 8.019 and a detection limit of 0.131 RIU for haemoglobin concentrations ranging from 10 to 40 g/L, corresponding to refractive indices between 1.34 and 1.43. The numerical analysis reveals a strong linear correlation between the resonance frequency and analyte concentration across the entire tested range. Furthermore, the K-nearest neighbour (KNN) regression model demonstrates cross-validated prediction accuracies between 92% and 98% across the simulated dataset.