S3129 Artificial Intelligence-Based Radiomics for Non-Invasive Detection of Lymph-Node Metastasis in Gastric Cancer: A Systematic Review and Meta-Analysis of Diagnostic Test Accuracy

Introduction: Accurate preoperative lymph node staging remains a critical challenge in gastric cancer management, as conventional imaging cannot reliably detect micrometastases. Radiomics, which extracts subvisual quantitative features from standard imaging modalities, utilizes deep learning and artificial intelligence (AI) to decode metastatic patterns undetectable by human observers. This meta-analysis evaluates its diagnostic accuracy for lymph node metastasis (LNM0 detection. Methods: A PRISMA-guided systematic review of PubMed, Embase, Scopus, and CENTRAL was conducted for original studies until April 2025. Eligibility criteria included studies that developed or validated radiomic models using AI/deep learning tools for non-invasive LNM detection in gastric cancer across various imaging modalities. Bivariate random-effects models pooled sensitivity, specificity, and area under the curve (AUC) with 95% confidence intervals (Cis). Heterogeneity was quantified using I2 statistics. PROSPERO ID: CRD420251064088. Results: Analysis of 28 studies (9,433 patients) revealed a pooled sensitivity of (0.80; 95% CI 0.76-0.84; I2 = 89.6%; P < 0.0001) and specificity of (0.83; 95% CI 0.80-0.85; I2 = 90.8%; P < 0.0001). For AUC evaluation across 39 studies (14,002 patients), the pooled AUC was (0.85; 95% CI 0.84-0.87; I2 = 97.3%; P < 0.0001). Substantial heterogeneity primarily reflected variation in segmentation methodology, though uniform lymph node size thresholds improved consistency. The Quality Assessment of Diagnostic Accuracy Studies, version 2 tool assessed methodological quality, revealing low bias risk in most studies. Conclusion: Radiomics demonstrates clinically meaningful accuracy for nodal metastasis detection, enabling diagnostic de-escalation through reduced reliance on invasive staging, personalized surgical planning via informed lymphadenectomy decisions, and neoadjuvant therapy optimization for high-probability metastasis cases. Standardization of feature extraction and multicenter prospective validation integrating endoscopic ultrasound remains essential before clinical implementation. This technology promises to transform gastric cancer staging from anatomical approximation to quantitative precision oncology.