Diagnosing Chronic Atrophic Gastritis By Gastroscopy Using Artificial Intelligence

Objective:Gastric cancer is the fifth most common cancer in the world and the third leading cause of cancer-related death.Gastric mucosal atrophy is a key stage in the progression of gastric cancer.Early diagnosis of chronic atrophic gastritis can provide lifestyle or drug intervention for patients,which is of great significance to prevent the occurrence and development of gastric cancer.However,the sensitivity of endoscopic diagnosis of chronic atrophic gastritis is only 42%.The clinical diagnosis of chronic atrophic gastritis is based on pathological examination,which leads to a high rate of missed diagnosis of chronic atrophic gastritis in clinical work.In order to improve the endoscopic diagnosis rate of chronic atrophic gastritis,we aim to apply artificial intelligence technology to the study of endoscopic images of chronic atrophic gastritis,and establish the artificial intelligence technology assisted endoscopic diagnosis model of chronic atrophic gastritis(CNN-CAG).Methods:We collected 5470 antral images of 1699 patients in Shanxi Provincial People’s Hospital from April 2018 to April 2019,and marked them according to the results of histopathological examination.Among them,3042 images were chronic atrophic gastritis and 2428 images were non atrophic gastritis.According to the proportion of 70%,15% and 15%,they were divided into the training set,validation set and test set of CNN-CAG model.The accuracy,sensitivity and specificity of CNN-CAG model in the diagnosis of atrophic gastritis under endoscopic images were trained and tested.The atrophic degree of chronic atrophic gastritis according to pathological results was divided into mild,moderate and severe.The CNNCAG model was trained to observe the accuracy of diagnosis of CAG with different degrees of gastric mucosal atrophy.The CNN-CAG model was used to diagnose CAG with three experts and two novices at the same time,and the difference of diagnosis accuracy between CNN-CAG model and endoscopists was compared.Results:The accuracy,sensitivity and specificity of artificial intelligence technology assisted endoscopic diagnosis model in the diagnosis of chronic atrophic gastritis were 0.942,0.945 and 0.940 respectively,which were slightly better than those of experts,obviously better than novices;the accuracy of CNN-CAG model in the diagnosis of mild,moderate and severe atrophic gastritis were 93%,95% and 99% respectively.Conclusion:The application of artificial intelligence technology to the endoscopic diagnosis model of chronic atrophic gastritis can accurately diagnose chronic atrophic gastritis and evaluate the degree of gastric mucosa atrophy under endoscopy.This can greatly simplify the diagnosis procedure,reduce the burden of endoscopic doctors,and save the energy of endoscopists and the medical expenses of patients.