Machine Learning Optimized Optical Surface Plasmon Resonance Biosensor Using Locally Weighted Linear Regression for Rapid and Accurate Detection of Tuberculosis Biomarkers

Detection and quantification of Mycobacterium tuberculosis (MTB) remain critical challenges in global public health, particularly in resource-limited settings where tuberculosis (TB) drives significant morbidity and mortality. This study presents the development and characterization of a THz-based optical biosensor platform for ultra-sensitive MTB detection. The biosensor architecture incorporates a hybrid graphene-gold metasurfaces fabricated on silicon dioxide, featuring an optimized array of L-shaped, plus-shaped, and square-shaped resonators. Computational electromagnetic simulations performed via COMSOL Multiphysics demonstrated exceptional sensitivity to refractive index modulations associated with MTB-specific biomarkers, achieving a maximum sensitivity of 2000 GHzRIU-1. Systematic parametric analyses were conducted to evaluate sensor performance across varying graphene chemical potentials, resonator geometries, and electromagnetic field incident angles. Implementation of a locally weighted linear regression (LOWESS) model enabled accurate prediction of sensor response characteristics at intermediate frequencies, yielding coefficient of determination (R2) values exceeding 85% across all investigated parameters. This label-free biosensing platform demonstrates promising potential for rapid, highly specific MTB detection, addressing a critical need for improved tuberculosis diagnostics in clinical settings.