Simulation-driven dual-band Graphene–Silver terahertz metasurface biosensor integrated with machine learning for mode-resolved hemoglobin detection

Accurate hemoglobin measurement is essential for diagnosing hematological disorders and monitoring cardiovascular health. This study introduces a terahertz metasurface biosensor combined with machine learning algorithms for rapid and non-invasive hemoglobin detection in clinical settings. The metasurface exhibits dual-band resonance in two terahertz regions, achieving a sensitivity of 450 GHz/RIU within a refractive index range of 1.34–1.43 RIU, with corresponding frequency shifts of 40 GHz and 30 GHz. Machine learning models, including Random Forest, Support Vector Machines, and Neural Networks, enhance the sensor’s analytical capability. Across four clinical categories—normal, mild anemia, moderate anemia, and severe anemia—the models attain 96.5 percent classification accuracy, with recall and precision scores above 0.94. Ensemble learning reduces the root mean square error to 0.28 g/dL, while denoising methods increase the signal-to-noise ratio by 16.3 dB. The biosensor supports real-time analysis, requires minimal sample volume, and eliminates complex preparation, making it suitable for continuous hemoglobin monitoring and cardiovascular health management. • Dual-band THz metasurface achieves 450 GHz/RIU sensitivity for hemoglobin detection. • ML integration enables 96.5% classification accuracy across anemia severity categories. • Ensemble learning lowers RMSE to 0.28 g/dL for precise hemoglobin quantification. • ML-driven denoising enhances spectral SNR by 16.3 dB for robust signal analysis. • Real-time, low-volume, preparation-free sensing supports proactive cardiovascular care.