A machine-learning model for staging autoimmune gastritis based on endoscopic and serological features

Autoimmune gastritis (AIG) lacks a unified, practical staging system. The AIG-atrophic stage (AIG-AS) scheme, based on the proportional area of remnant oxyntic mucosa (ROM), is promising but observer-dependent. We developed machine-learning models to stage AIG using endoscopic and serological features.

This single-center cross-sectional study enrolled 203 patients with AIG confirmed by integrated endoscopic, histological, and serological criteria between December 2023 and December 2025. White-light endoscopy (WLE), magnifying endoscopy with narrow-band imaging (ME-NBI), and a serum panel were collected. Patients were classified under AIG-AS as Stage 1 (50%

Stages 1, 2, and 3 included 87 (42.9%), 36 (17.7%), and 80 (39.4%) patients. The combined endoscopy+serology random forest achieved the highest three-class macro-AUC (0.878 ± 0.051; accuracy 74.9%), surpassing WLE stacking (0.853 ± 0.056) and serology-only models (≤0.693). For binary staging, the WLE+ME-NBI random forest reached an AUC of 0.913 ± 0.046. Loss of gastric folds, gastrin-17, pepsinogen I, crypt-opening grade, and cast-off skin appearance were the dominant predictors. Barrett esophagus prevalence was 66.8%, with long-segment Barrett esophagus 3.3-fold more common in advanced AIG than in earlier disease.

Multi-modal machine learning enables accurate, interpretable AIG staging. Endoscopic features outperform serology, and combined models offer the best three-class performance. The high prevalence of Barrett esophagus supports routine esophageal surveillance in AIG.