Precise cancer detection via the combination of functionalized SERS surfaces and convolutional neural network with independent inputs
Abstract
Combining the advanced approaches of surface functionalization and chemistry, plasmonics, surface enhanced Raman spectroscopy (SERS), and machine learning, we propose the advanced route for express and precise recognition of normal and cancer cells. Our interdisciplinary approach uses plasmonic coupling between the specific nanoparticles and underlying periodical plasmonic surface and achieves high SERS enhancement factor.
The surface of gold multibranched nanoparticles (AuMs) was functionalized with different chemical groups to achieve partially selective entrapping of biomolecules from cells cultivation media and generate information-rich inputs for machine learning methods and SERS-based cells recognition. Evaluation of convolutional neural networks (CNN) training results, performed with ad hoc feature selection method, suggests that the grafted functional groups provide specificity to proteins, nucleic acids and lipids, responsible for cancer line identification.
The dataset of SERS control spectra of normal and cancer cell's metabolites were classified by the trained CNN and perfectly distinguished with 100 % prediction accuracy.