Deep learning-assisted graphene-based THz metasurface for label-free and high-performance MCF-7 breast cancer cell sensing

A triple-band graphene-based terahertz metasurface, integrated with a DL framework, is proposed for high-efficiency polarization conversion and label-free MCF-7 breast cancer cell detection. The unit cell, comprising concentric graphene elements and silver strips on an \(\mathrm {Al_2O_3}\) substrate with a silver ground plane, facilitates reflective-mode operation with polarization conversion ratios (PCR) exceeding 90% across three distinct operational bands. The finite-element method (FEM), finite-difference time-domain (FDTD) technique, and multiple-interference theory rigorously validate the physical mechanisms and electromagnetic responses, showing strong agreement. For biosensing applications, the metasurface leverages the distinct dielectric signatures of MCF-7 cells, inducing measurable resonance shifts at 1.48, 2.85, and 4.12 THz. The platform achieves a peak sensitivity of 0.866 THz/RIU and a figure of merit of 4.04 RIU \(^{-1}\) . To improve spectral analysis, we develop a one-dimensional convolutional neural network to predict PCR responses directly from THz spectral data. This model shows exceptional robustness, achieving an \(R^2\) of 0.972 even under 5% additive white Gaussian noise. By accurately resolving minute refractive-index variations associated with early-stage cellular-level oncogenic alterations, the proposed deep-learning-assisted biosensor offers a highly sensitive, non-contact, and non-destructive platform with significant potential for early-stage breast cancer screening.