Metabolic SERS Profiling under Triple-Coupled Hotspots Aided by Machine Learning for Precise Identification of Gastrointestinal Malignancies.

Serum metabolic analysis based on label-free SERS fingerprints represents a highly promising technique for the detection of tumors. However, the insufficient hotspot density of SERS substrates often results in incomplete molecular Raman bands, hindering the accurate discrimination of tumors. Herein, we propose a clean, hydrophobic monolayer gold nanofilm/potassium iodide-encapsulated silver-coupled (Ag@KI/HMGF) SERS substrate. This triple-coupled SERS substrate achieves exceptional signal amplification and enables the acquisition of abundant SERS bands for multicomponent systems. Robust machine learning classifiers were employed to identify and analyze the Raman spectral features of serum metabolites, and this method exhibited high clinical accuracy, sensitivity, and specificity in distinguishing samples from patients with different types of gastrointestinal malignancies and healthy individuals. The serum metabolic analysis based on the coupled SERS substrate constructed in this study provides a highly promising solution to address the clinical challenge of gastrointestinal malignant tumor discrimination and can be further extended to clinical cancer detection based on other biological fluids.