Usefulness of Artificial Intelligence for Transnasal Esophagogastroduodenoscopy in Clinical Practice

Objectives We evaluated the diagnostic performance of an artificial intelligence (AI) diagnostic system for biopsy-considered suspicious lesions during transnasal esophagogastroduodenoscopy (EGD), focusing on its negative predictive value (NPV) to demonstrate whether a negative AI result may help endoscopists make a more conservative biopsy decision in selected low-risk lesions. Methods, patients, and materials We conducted a prospective single-center observational study. Endoscopists observed the entire stomach using transnasal EGD and the AI diagnostic system (gastroAI™ model-G, AI Medical Service Inc, Tokyo, Japan). Two endoscopic images of the suspected lesion were recorded using an AI diagnostic system; biopsies were collected for histological diagnosis, regardless of the AI diagnosis. The primary endpoint was the NPV of the gastroAI™ model-G. The reference standard used for calculating the diagnostic performance of the gastroAI™ model-G was histopathological diagnosis. Results Overall, 392 patients were enrolled, and 98 specimens from 87 patients were analyzed. A total of 11 endoscopists (five experts and six non-experts) performed the examinations. Five specimens (5%) were diagnosed as adenocarcinoma. As all specimens diagnosed as “negative” using the gastroAI™ model-G were pathologically benign, the NPV of the system was 100% (95% confidence interval (CI): 84%-100%). Its sensitivity, specificity, positive predictive value, and accuracy were 100% (95% CI: 36%-100%), 34% (95% CI: 25%-45%), 8% (95% CI: 3%-17%), and 38% (95% CI: 28%-48%), respectively. Specificity and accuracy were significantly higher for endoscopists than for the model (96% vs. 34% and 94% vs. 38%, respectively; p < 0.01), whereas the model had a higher sensitivity than did endoscopists (100% vs. 60%), although the difference was not statistically significant. Conclusions This study reveals the potential clinical utility of the AI diagnostic system for transnasal EGD, particularly indicating its high NPV.