Gastric cancer AI tested for nodal spread

Xiong and colleagues developed a machine learning model to estimate lymph node metastasis risk in early gastric cancer. They assessed it in external and prospectively enrolled cohorts. The European Journal of Surgical Oncology published the study.

Study cohorts

The researchers used four cohorts from three tertiary centers. They retrospectively enrolled patients with early gastric cancer who underwent gastrectomy and lymph node dissection for training and external validation. The training cohort included 3,412 patients; 467, or 13.7%, had lymph node metastasis. That count excludes the validation cohorts.

A separate cohort enrolled prospectively from September 2023 to December 2024 provided temporal validation. The researchers also analyzed a preoperative pathology validation cohort from January 2025 to January 2026.

Model performance

The team compared six algorithms and selected XGBoost. Its area under the receiver operating characteristic curve, or AUC, was 0.8583 in external validation and 0.8766 in prospective validation. The preoperative pathology cohort had an AUC of 0.8653. AUC measures discrimination, with 0.5 indicating chance performance and 1.0 perfect separation.

Important predictors included lymphovascular or neural invasion, invasion depth, tumor location, differentiation and size. The team used SHAP to explain the contribution of individual inputs and developed a web calculator.

Limits of the evidence

Nodal risk matters when considering endoscopic submucosal dissection, or ESD, versus surgery. The authors propose the model as decision support. They did not test whether AI-guided treatment choices reduce unnecessary surgery, missed metastases or adverse patient outcomes.

The abstract does not report validation cohort sizes or sensitivity and specificity at a clinical decision threshold. It also does not give enough detail to establish how every predictor was obtained before surgery.

Study notes

This multicenter retrospective development study includes external and prospective validation. Reporting here relies on the public abstract. Funding, conflicts of interest and Japanese regulatory status could not be confirmed.

Sources

1. [1]
   論文：Xiong et al. Development and validation of an explainable model for predicting lymph node metastasis in early gastric cancer
2. [2]
   データベース：PubMed 公開要旨

Images: generated with GPT Image 2.5 (illustrative).