A Machine Learning-Driven Surface-Enhanced Raman Scattering Analysis Platform for the Label-Free Detection and Identification of Gastric Lesions.

Background: Gastric lesions pose significant clinical challenges due to their varying degrees of malignancy and difficulty in early diagnosis. Early and accurate detection of these lesions is crucial for effective treatment and improved patient outcomes. Methods: This paper proposed a label-free and highly sensitive classification method for serum of patients with different degrees of gastric lesions by combining surface-enhanced Raman scattering (SERS) and machine learning analysis. Specifically, we prepared Au lotus-shaped (AuLS) nanoarrays substrates using seed-mediated and liquid-liquid interface self-assembly method for measuring SERS spectra of serum, and then the collected spectra were processed by principal component analysis (PCA) - multi-local means based nearest neighbor (MLMNN) model to achieve differentiation. Results: By employing this pattern analysis, AuLS nanoarray substrates can achieve fast, sensitive, and label-free serum spectral detection. The classification accuracy can reach 97.5%, the sensitivity is higher than 96.7%, and the specificity is higher than 95.0%. Moreover, by analyzing the PCs loading plots, the most critical spectral features distinguishing different degrees of gastric lesions were successfully captured. Conclusion: This discovery lays the foundation for combining SERS with machine learning for real-time diagnosis and recognition of gastric lesions.