A multimodal artificial intelligence method for accurate diagnosis of autoimmune gastritis

Autoimmune gastritis (AIG), a chronic inflammatory disease associated with various comorbidities and complications, is often subject to missed or delayed diagnoses.

The objective of this study was to develop an artificial intelligence (AI) system to assist in the diagnosis of AIG by integrating multimodal information, including endoscopic images, biopsy pathological reports, and laboratory test results.

Multimodal data were collected from 590 patients diagnosed with AIG, H. pylori positive, and H. pylori negative chronic atrophic gastritis at three medical centers. A multimodal AI system was developed to diagnose AIG and other types of chronic gastritis. The performance of the AI system was evaluated using both internal and external datasets. Six endoscopists were invited to perform three-category classification for comparison. The SHapley Additive exPlanations (SHAP) framework and EigenCAM were used to improve the interpretability of the AI system.

The proposed multimodal AI system achieved a sensitivity of 0.931, specificity of 0.963, and accuracy of 0.954, significantly outperforming unimodal models (all P<0.001) and the best-performing invited expert endoscopist under controlled conditions. SHAP analysis indicated that the three modalities provided complementary and synergistic information.

This multimodal AI system has the potential to enable practical AI-assisted diagnosis of AIG with exceptional accuracy.