Artificial Intelligence in Gastric Cancer: Diagnostic, Prognostic, and Predictive Developments, Evidence Maturity, and Translational Challenges.

BACKGROUND

Gastric cancer (GC) remains a major global health burden characterized by substantial heterogeneity in diagnosis, prognosis, and treatment response. Artificial intelligence (AI), including machine learning (ML) and deep learning (DL), has been increasingly investigated for the analysis of endoscopic images, radiological scans, whole-slide pathology images, molecular profiles, liquid-biopsy data, and multimodal clinical datasets. This structured review summarizes recent AI-based developments in GC diagnosis, staging, prognostic assessment, and treatment-response prediction, with particular emphasis on the translational maturity of the available evidence. PubMed/MEDLINE, Web of Science, Embase, Scopus, and Google Scholar were searched for relevant English-language publications from database inception through January 2026. Eligible studies were evaluated according to clinical task, data modality, model type, cohort size, study design, validation strategy, interpretability, clinical workflow readiness, and reported limitations.

RECENT FINDINGS: Recent AI models have demonstrated promising performance in selected research settings, particularly in endoscopic image analysis, computational pathology, CT-based radiomics, and multimodal prognostic modeling. However, most studies remain retrospective, single-center, and internally validated. Calibration, robustness, prospective clinical utility, and generalizability across populations and institutions remain insufficiently assessed. Major barriers to translation include data heterogeneity, annotation variability, overfitting, data leakage, external validation failure, algorithmic bias, regulatory requirements, reimbursement, clinician-AI interaction, medicolegal responsibility, post-deployment monitoring, and model drift.

CONCLUSION

AI has considerable potential to support GC diagnosis, prognosis, and treatment planning, but most current models remain insufficiently validated for routine clinical use. Future progress will depend on transparent reporting, standardized validation, multicenter prospective trials, clinically meaningful endpoints, and the integration of explainable, trustworthy AI systems into routine clinical workflows.