Atrophy App: deep learning-based detection of atrophic gastritis

Conventional white-light endoscopy has a relatively low sensitivity, low specificity and high inter-observer variability for the diagnosis of atrophic gastritis. Thus, it is thought to be insufficient to reliably diagnose it. Our goal is to improve endoscopic accuracy for the diagnosis of atrophic gastritis using deep learning.

Our convolutional neural network was trained to analyze endoscopic images from the proximal stomach, achieving an accuracy of 92.9% and performing significantly better than the combined results of 6 independent endoscopists. With this web tool we want to provide wide access to this software. It allows you to upload your own endoscopic images for analysis, and provides an indicative diagnosis.