S3117 Accuracy of AI-Based Raman Spectroscopy in the Diagnosis of Gastric Cancer: A Systematic Review and Meta-Analysis

Introduction: Gastric cancer (GC) is a major global health burden, largely due to late-stage diagnosis. While endoscopy is the gold standard, it is invasive and may miss early lesions. Raman spectroscopy (RS), a technique detecting molecular vibrations, combined with artificial intelligence (AI), offers a rapid, minimally invasive alternative. This systematic review and meta-analysis assessed the diagnostic performance of AI-assisted RS for GC detection. Methods: Following PRISMA-DTA guidelines, 5 databases (PubMed, Web of Science, Scopus, VHL, and Google Scholar) were searched until February 2025. Eligible studies assessed AI-based RS for GC diagnosis and provided data for 2 × 2 contingency tables. Study quality was assessed with Quality Assessment of Diagnostic Accuracy Studies, version 2. Pooled sensitivity, specificity, likelihood ratios, diagnostic odds ratio, and area under the curve (AUC) were calculated using R. Heterogeneity, publication bias, and subgroup analyses (sample type, RS technique, AI model) were evaluated. Results: Nineteen studies (1,932 patients; 18,987 spectra) were included. Study quality was moderate-to-high, though risks of bias were noted in patient selection and index tests. Per-spectrum analysis (12 studies) showed pooled sensitivity of 94.6% (95% confidence interval [CI] 87.9-97.6%), specificity of 96.7% (95% CI 89.6-99.0%), and AUC of 0.967. Per-patient analysis (7 studies) reported sensitivity of 92.9% (95% CI 85.7-96.6%), specificity of 93.7% (95% CI 90.6-95.9%), and AUC of 0.964. Conventional RS performed best (P < 0.05). No significant differences were found by sample type (P = 0.1581) or AI model (P > 0.05). Egger’s test indicated possible publication bias (P < 0.05). Conclusion: AI-assisted RS shows high diagnostic accuracy for GC across sample types and models. Conventional RS performed best. These findings support its potential as a noninvasive diagnostic aid. Future research with standardized protocols and larger cohorts is needed to confirm clinical utility and reduce heterogeneity.