Advanced Optimization of THz Biosensor for Breast Cancer Detection Through Machine Learning

Abstract This work presents a highly sensitive and reconfigurable terahertz (THz) biosensor for the detection of breast cancer biomarkers. The device consists of a hybrid graphene-gold THz biosensor patterned on a dielectric substrate, providing strong plasmonic field confinement and tunable resonance characteristics. To optimize its optical response, a Support Vector Machine (SVM) regression model was employed to capture the nonlinear interaction among graphene’s chemical potential, relaxation time, and temperature. The optimized structure achieves excellent sensing performance, with sensitivities of 137.2 GHz/RIU for normal epithelium (NE), 193.7 GHz/RIU for CA15-3, 241.6 GHz/RIU for CA27-29, and 311.5 GHz/RIU for HER2, along with a quality factor (Q-Factor) of 6.66 and a figure of merit (FOM) of 636.7 RIU⁻¹. Combining tunability, compactness, and machine-learning-based optimization, the proposed hybrid graphene-gold THz biosensor demonstrates great promise for real-time, label-free, and precise breast cancer diagnostics.