Artificial Intelligence Model for Detecting Duodenal Endoscopic Changes on Images of Functional Dyspepsia

Introduction and Aims Recently, it has been suggested that the duodenum may be the pathological locus of functional dyspepsia (FD), but there are few reports on the presence of duodenal changes on imaging in FD. Methods Duodenal images acquired at our hospital with the presence or absence of the term “functional dyspepsia” on the electronic medical record, and H. pylori (HP) infection status on the Japan Endoscopy Database were obtained. The Google Cloud Platform AutoML Vision (single-label classification) was used to classify FD/HP current infection, versus FD/HP uninfected patients. We constructed an AI model to distinguish four groups, which included FD/HP current infection, FD/HP uninfected, non-FD/current infection, and non-FD/HP uninfected, and calculated the sensitivity, specificity, and AUC. Patient images with other organic diseases such as gastrointestinal cancer, peptic ulcer, postoperative abdominal organs, and gastroesophageal reflux disease were excluded. Narrow band imaging and dye-spread images were also excluded. Results The overall AUC of the four groups was 0.47 (FD/HP current infection 0.20, FD/HP uninfected 0.35, non-FD/current infection 0.46, non-FD/HP uninfected 0.74). Next, using the same images, we constructed a model to determine the presence or absence of FD in HP infected patients only. The sensitivity, specificity, positive predictive value, negative predictive value, and AUC were 71.4%, 66.7%, 50.0%, 83.3%, and 0.84, respectively. However, when we constructed a model to determine the presence of FD only in uninfected HP patients, the sensitivity, specificity, positive predictive value, negative predictive value, and AUC were 58.3%, 100%, 100%, 77.3%, and 0.85, respectively. Conclusion The results suggest that the duodenal imaging AL model may be able to determine the presence or absence of FD to a certain degree in HP-infected or uninfected patients.