Journal of gastroenterology and hepatology · 2026
Artificial Intelligence for Single-Image Diagnosis of Autoimmune Gastritis Using Gastric Corpus Atrophy Images: A Multicenter Retrospective Diagnostic Study.
СтатьяМногоцентровое исследованиеНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Song Yiming, Zhang Xunbing, Shi Liwei, Lu Junyu, Li Zeyu, Wang Ruilan, Chen Jinnan, Huang Yu, Zhou Yujie, Li Zhao, Lin Yansheng, Huang Jian, Tang Zhaorong, Shang Cheng, Xu Wenhui, Yang Shiying, Li Meixuan, Chen Huimin, Lu Hong, Liang Xiao, Zhang Qingwei, Li Xiaobo
Аннотация
BACKGROUND AND AIMS: Endoscopic differentiation between autoimmune gastritis (AIG) and Helicobacter pylori-associated atrophic gastritis (HpAG) remains clinically challenging, often leading to misdiagnosis and inappropriate treatment. We aimed to develop and validate an AI system for AIG diagnosis using single white-light endoscopic images of gastric corpus atrophy.
METHODS: This multicenter retrospective study included 2137 patients (1064 AIG and 1073 HpAG) from five tertiary hospitals in China. Data from two centers were used for training and internal validation, and three centers for external validation. We developed CorpusNet-AIG using the RegNet architecture and compared its diagnostic performance with that of 15 endoscopists (5 experts, 5 seniors, and 5 novices) using 100 cases. Endoscopists performed diagnoses under two conditions: single gastric corpus atrophy images and complete endoscopic image sets. Metrics included accuracy, sensitivity, specificity, PPV, NPV, and AUC.
RESULTS: CorpusNet-AIG achieved excellent performance in internal validation (accuracy 94.51%, sensitivity 93.23%, specificity 95.71%, AUC 0.990) and external validation (accuracy 92.94%, sensitivity 93.18%, specificity 92.68%, AUC 0.972). Using single images, the AI system (93.00%) significantly outperformed novice (73.80%, p < 0.001), and senior endoscopists (81.80%, p = 0.009) showed comparable performance to expert endoscopists (89.20%, p = 0.334) and maintained superior performance over novices (74.60%, p < 0.001) and seniors (84.20%, p = 0.033) even when they used complete image sets, while achieving equivalent performance to experts using complete images (93.00%, p > 0.999).
CONCLUSIONS: We developed the first AI system to achieve expert-level AIG diagnosis using single gastric corpus atrophy images. Prospective multicenter validation is needed to support clinical implementation.
METHODS: This multicenter retrospective study included 2137 patients (1064 AIG and 1073 HpAG) from five tertiary hospitals in China. Data from two centers were used for training and internal validation, and three centers for external validation. We developed CorpusNet-AIG using the RegNet architecture and compared its diagnostic performance with that of 15 endoscopists (5 experts, 5 seniors, and 5 novices) using 100 cases. Endoscopists performed diagnoses under two conditions: single gastric corpus atrophy images and complete endoscopic image sets. Metrics included accuracy, sensitivity, specificity, PPV, NPV, and AUC.
RESULTS: CorpusNet-AIG achieved excellent performance in internal validation (accuracy 94.51%, sensitivity 93.23%, specificity 95.71%, AUC 0.990) and external validation (accuracy 92.94%, sensitivity 93.18%, specificity 92.68%, AUC 0.972). Using single images, the AI system (93.00%) significantly outperformed novice (73.80%, p < 0.001), and senior endoscopists (81.80%, p = 0.009) showed comparable performance to expert endoscopists (89.20%, p = 0.334) and maintained superior performance over novices (74.60%, p < 0.001) and seniors (84.20%, p = 0.033) even when they used complete image sets, while achieving equivalent performance to experts using complete images (93.00%, p > 0.999).
CONCLUSIONS: We developed the first AI system to achieve expert-level AIG diagnosis using single gastric corpus atrophy images. Prospective multicenter validation is needed to support clinical implementation.