A novel artificial intelligence-based system for quality monitoring during esophagogastroduodenoscopy: a multicenter randomized controlled study

Esophagogastroduodenoscopy (EGD) is the pivotal procedure for diagnosis of upper gastrointestinal (UGI) lesions. However, significant variation in EGD performance among endoscopists impacts detection rates of UGI cancers and precursor lesions. We developed a novel EGD quality monitoring system and evaluated its effectiveness in a randomized controlled study.The endoscopy quality control assistant (EQCA) was developed using deep convolutional neural networks and long short-term memory. Patients (≥18 years) undergoing EGD in seven hospitals were consecutively enrolled and randomly assigned to the EQCA-assisted group or control group. The primary outcome was the detection rate for cancer-related lesions (low and high grade intraepithelial neoplasia and cancer) and cancer (early and advanced cancer) in the UGI tract.After randomization and exclusions, 16 005 patients in the control group and 16 012 in the EQCA group were analyzed. Detection rates for UGI cancer-related lesions and cancer were significantly higher in the EQCA group than in the control group (8.00% vs. 5.55%; 1.93% vs. 1.21%; both P < 0.001). The EQCA group had a higher operation score, reflecting examination quality, and longer inspection time than the control group. The detection rate for UGI cancer-related lesions was positively correlated with operation score (r = 0.9217, P < 0.001) and inspection time (r = 0.8943, P < 0.001) for each hospital.The use of EQCA during EGD was associated with increased detection of UGI cancer and precancerous lesions. Our novel EQCA system can be an effective tool for monitoring real-time EGD quality.