A machine learning-enhanced gastric cancer diagnostic method based on shell-isolated nanoparticle-enhanced Raman spectroscopy

Gastric cancer (GC) remains one of the most prevalent and lethal malignancies worldwide, necessitating the development of efficient, non-invasive methods for early detection. In this study, a serum diagnostic approach based on shell-isolated nanoparticle-enhanced Raman spectroscopy (SHINERS) is proposed to address the limitations of conventional Raman spectroscopy, such as weak signal intensity, sample inhomogeneity, and nonspecific adsorption. Silver-coated silica (Ag@SiO₂) core-shell nanoparticles were synthesized <i>via</i> a chemical reduction method and employed as signal enhancement substrates. The SHINERS platform significantly improves spectral stability and specificity by enhancing Raman signals and mitigating the coffee-ring effect. A clinical cohort comprising 100 patients with GC and 100 healthy controls was established. Distinct molecular fingerprint spectra with high signal-to-noise ratios were obtained from 3 μL serum samples within a 10 minute detection window. Four classification models were developed using machine learning algorithms, including one-dimensional convolutional neural network (1D-CNN), random forest (RF), support vector machine (SVM), and <i>k</i>-nearest neighbors (kNN). Among these, the SVM model demonstrated the highest classification performance with an area under the receiver operating characteristic (ROC) curve of 0.9000, significantly outperforming other algorithms (<i>p</i> < 0.01). These results confirm the feasibility of combining SHINERS with machine learning for reliable, rapid, and minimally invasive screening of gastric cancer, and underscore its potential application in clinical diagnostics.