A novel endoscopic artificial intelligence system to assist in the diagnosis of autoimmune gastritis: a multicenter study

Abstract

Background: Autoimmune gastritis (AIG), distinct from Helicobacter pylori-associated atrophic gastritis (HpAG), is underdiagnosed due to limited awareness. This multicenter study aimed to develop a novel endoscopic artificial intelligence (AI) system for assisting in AIG diagnosis.

Methods: Patients diagnosed with AIG, HpAG, or nonatrophic gastritis (NAG), were retrospectively enrolled from six centers. Endoscopic images with relevant demographic and medical data were collected for development of the AI-assisted system based on a multi-site feature fusion model. The diagnostic performance of the AI model was evaluated in internal and external datasets. Endoscopists' performance with and without AI support was tested and compared using Mann-Whitney U test. Heatmap analysis was performed to interpret AI model outputs.

Results: 18 828 endoscopy images from 1070 patients (294 AIG, 386 HpAG, 390 NAG) were collected. On testing datasets, AI identified AIG with 96.9 % sensitivity, 92.2 % specificity, and area under the receiver operating characteristic curve (AUROC) of 0.990 (internal), and 90.3 % sensitivity, 93.1 % specificity, and AUROC of 0.973 (external). The performance of AI (sensitivity 91.3 %) was comparable to that of experts (87.3 %) and significantly outperformed nonexperts (70.0 %; P = 0.01). With AI support, the overall performance of endoscopists was improved (sensitivity 90.3 % [95 %CI 86.0 %-93.2 %] vs. 78.7 % [95 %CI 73.6 %-83.2 %]; P = 0.008). Heatmap analysis revealed consistent focus of AI on atrophic areas.

Conclusions: This novel AI system demonstrated expert-level performance in identifying AIG and enhanced the diagnostic ability of endoscopists. Its application could be useful in guiding biopsy sampling and improving early detection of AIG.