Related Experiment Video
Updated: May 12, 2026
Comparison of Predictive Performance of Three Lymph Node Staging Systems in Colorectal Signet Ring Cell Carcinoma Based on Machine Learning Model
Published on: April 18, 2025
Artificial intelligence models in predicting lymph node metastasis in early gastric cancer: a systematic review and
Xiaopeng Chen1, Zhengguo Yi1, Jinping Ye1
1Department of General Surgery, Shangrao Municipal Hospital, Jiangxi, Shangrao, China.
Introduction:
Accurate preoperative assessment of lymph node metastasis (LNM) is a key determinant of treatment selection in early gastric cancer (EGC), particularly when choosing between endoscopic resection and minimally-invasive gastrectomy with lymphadenectomy. Although artificial intelligence (AI)-based models have been increasingly developed for LNM prediction, their overall diagnostic performance and clinical relevance to minimally-invasive treatment decision-making remain unclear.
Aim:
This systematic review and meta-analysis aimed to evaluate the diagnostic accuracy of AI-based models for predicting LNM in EGC, and to clarify their potential role in guiding minimally-invasive and endoscopic treatment strategies.
Materials And Methods:
A comprehensive literature search of PubMed, Embase, and Web of Science databases was conducted through August 2025. Studies applying machine learning or deep learning algorithms to predict LNM in EGC were included. Study quality was assessed using the Quality Assessment of Diagnostic Accuracy Studies-2 tool. Pooled sensitivity, specificity, and area under the curve (AUC) were calculated using a bivariate random-effects model. Subgroup analyses, meta-regression, and publication bias assessment were performed.
Results:
A total of 18 studies involving 41 505 patients were included. In the internal validation cohorts, AI-based models demonstrated a pooled sensitivity of 0.81 and specificity of 0.82, with the AUC of 0.88. Comparable performance was observed in the external validation cohorts (sensitivity, 0.81; specificity, 0.84; AUC, 0.9), indicating good generalizability. In the studies directly comparing AI with clinician assessment, AI models consistently achieved higher sensitivity and overall diagnostic accuracy.
Conclusions:
AI-based models show robust performance for predicting LNM in EGC and outperform clinician assessment. Importantly, these models have the potential to serve as clinically meaningful decision-supporting tools for minimally-invasive and endoscopic management by assisting in the selection between endoscopic resection and gastrectomy with lymphadenectomy, thereby optimizing surgical extent while preserving oncological safety.
