High-sensitivity graphene-gold metasurface optical biosensor for early melanoma detection optimized with machine learning using a one-dimensional convolutional neural network and binary encoding

Abstract

This research presents an advanced terahertz metasurface sensor incorporating graphene and gold elements in a W-shaped resonator configuration for non-invasive melanoma detection. The sensor design, optimized through COMSOL Multiphysics simulations, operates by detecting minute variations in the refractive index of skin tissue that occur during early melanoma development. The optimized sensor achieves a sensitivity of 450 GHzRIU⁻¹ with a narrow spectral linewidth of 35 GHz in the terahertz regime. A one-dimensional convolutional neural network (1D-CNN) algorithm enhances the sensor's predictive capabilities, achieving R² values exceeding 0.95 across various operational parameters. The sensor demonstrates dual functionality through binary information encoding capability via chemical potential modulation. The proposed design shows significant advantages over conventional diagnostic methods, offering rapid, non-invasive detection with high accuracy. Moreover, our numerical findings reveals that the designed sensor provides some robustness performance against various geometric parameters and incident angles, which in turns make it promising for practical melanoma diagnosis applications.