Endoscopic three-categorical diagnosis of Helicobacter pylori infection using linked color imaging and deep learning: a single-center prospective study (with video)

Abstract

Background: Helicobacter pylori (H. pylori) eradication is required to reduce incidence related to gastric cancer. Recently, it was found that even after the successful eradication of H. pylori, an increased, i.e., moderate, risk of gastric cancer persists in patients with advanced mucosal atrophy and/or intestinal metaplasia. This study aimed to develop a computer-aided diagnosis (CAD) system to classify the status of H. pylori infection of patients into three categories: uninfected (with no history of H. pylori infection), currently infected, and post-eradication. Methods: The CAD system was based on linked color imaging (LCI) combined with deep learning (DL). First, a validation dataset was formed for the CAD systems by recording endoscopic movies of 120 subjects. Next, a training dataset of 395 subjects was prepared to enable DL. All endoscopic examinations were recorded using both LCI and white-light imaging (WLI). These endoscopic data were used to develop two different CAD systems, one for LCI (LCI-CAD) and one for WLI (WLI-CAD) images. Results: The diagnostic accuracy of the LCI-CAD system was 84.2% for uninfected, 82.5% for currently infected, and 79.2% for post-eradication status. Comparisons revealed superior accuracy of diagnoses based on LCI-CAD data relative based on WLI-CAD for uninfected, currently infected, and post-eradication cases. Furthermore, the LCI-CAD system demonstrated comparable diagnostic accuracy to that of experienced endoscopists with the validation data set of LCI. Conclusions: The results of this study suggest the feasibility of an innovative gastric cancer screening program to determine cancer risk in individual subjects based on LCI-CAD.