SERS-AI-LUA-Driven Salivary Diagnosis of Head and Neck Cancer Using Graphene-Assisted Plasmonic Nanocorals

Abstract

The early detection of head and neck cancer (HNC) remains an important challenge owing to the lack of reliable noninvasive biomarkers. This study introduces a graphene-assisted plasmonic nanocoral platform coupled with an artificial intelligence-linear unmixing algorithm for diagnosing HNC from saliva and identifying associated metabolic biomarkers. The nanocoral structures, formed via a spontaneous gold growth mechanism on graphene templates, exhibit strong plasmonic enhancement and selective adsorption of volatile metabolites. Raman signals acquired from the saliva of HNC patients and healthy individuals are analyzed using a logistic regression model, achieving 98% classification accuracy. To identify potential metabolic biomarkers, candidate metabolites are initially selected based on spectral similarity using the Pearson correlation coefficient. Subsequently, the nonnegative least squares method is applied to refine this selection and extract the final set of biomarker candidates. This approach identifies 15 potential metabolic biomarkers, and their clinical relevance is corroborated through comparison with the findings of previous clinical studies. This study not only introduces a highly sensitive, noninvasive diagnostic platform for HNC but also establishes a robust framework for Raman-based biomarker discovery, with potential applicability that warrants evaluation in other biofluid-based disease models in future studies.