OBJECTIVES: Artificial intelligence (AI)-assisted endoscopy has been developed for the early detection of upper gastrointestinal cancer; however, its clinical effectiveness remains insufficiently evaluated. This study assessed its effectiveness in a health screening facility. METHODS: This retrospective cohort study compared AI-assisted and non-AI-assisted upper gastrointestinal endoscopy at Omiya City Clinic, Japan (April 2021-March 2024). Participants who underwent endoscopy between April 2021 and March 2023 were classified as the non-AI group, while those examined between April 2023 and March 2024 comprised the AI group. The AI-assisted system was introduced in April 2023. The primary outcome was cancer detection rate (CDR), with secondary outcomes including biopsy rate and positive predictive value (PPV). Propensity score matching (PSM) was performed for age, sex, alcohol consumption, smoking, Helicobacter pylori infection history, endoscopist experience, and prior-year endoscopy to minimize bias. RESULTS: In total, 17,662 were included in the AI group and 32,318 in the non-AI group. PSM created 17,662 matched pairs. In the AI group, the CDR for gastric cancer (GC) was significantly higher compared to the non-AI group (0.10% vs. 0.03%, p < 0.05). The biopsy rate was slightly higher in the AI group, with no significant difference, whereas the PPV of biopsy for gastric cancer and esophageal cancer was significantly increased (4.84% vs. 2.16%, p < 0.05). CONCLUSIONS: In a clinical screening setting, AI-assisted endoscopy significantly improved the CDR of GC and enhanced the PPV of biopsies. These findings highlight AI-assisted endoscopy as a valuable tool for early GC diagnosis in screening environments.
Goto et al. (Fri,) studied this question.