The vertices number determined SERS activity of polyhedra and the application in oral cancer detection based on deep learning.

Due to the inherent specificity and high sensitivity, biomedical detections based on Surface-Enhanced Raman Scattering (SERS) technology have garnered increasing attention. In the SERS detection process, fabricating a highly sensitive SERS substrate is the most critical. Although various methods, such as self-assembly and nanocavities, are used to enhance the local electric field intensity and thus improve SERS activity, the foundation still lies in the preparation of individual noble metal nanoparticles with high SERS activity. The paper models spherical, tetrahedral, cubic, octahedral, and dodecahedral shapes and use the Finite-Difference Time-Domain (FDTD) simulation to study the impact of the number of vertices in polyhedra on the SERS activity of nanoparticles, finding that fewer vertices in the polarization direction of the local electric field can achieve the maximum SERS activity. Based on this result, we fabricated gold nano-tetrahedron SERS substrates and used Rhodamine 6G (R6G) as a probe molecule, measuring a SERS enhancement factor (EF) of 1.1 × 10 at 611 cm, with the limit of detection (LOD) of 1 × 10 M and the linear detection range from 2.48 nM to 1000 nM. Additionally, we used these nanoparticles to prepare a SERS substrate for the detection of saliva from oral cancer patients and combined it with the deep learning neural network to achieve intelligent differentiation between different stages oral cancer patients. This study indicates that the combination of SERS technology and deep learning neural network technology has tremendous potential in clinical SERS detection.