Objective To investigate the application of artificial intelligence (AI)-assisted digestive endoscopy in the diagnosis of early gastric cancer and precancerous lesions.

Methods Patients who underwent painless gastroscopy at Endoscopy Center in Inner Mongolia People’s Hospital from August to December 2024 and met the inclusion and exclusion criteria were randomly assigned into the AI-assisted gastroscopy group and traditional gastroscopy group. The detection rate of endoscopic lesions and pathological characteristics of biopsies were compared between two groups.

Results The detection rate of elevated erosion/mucosal roughness and gastric polyps in the AI-assisted gastroscopy group under endoscopy was better than that in the traditional gastroscopy group (41.7% vs. 33.6%, 30.6% vs. 23.0%, both P ＜ 0.05), while also achieving shorter median time for gastroscopy examination and gastric observation (5.00 min vs. 5.80 min, 3.15 min vs. 4.00 min, both P ＜ 0.05). In the absence of significant differences in biopsy rate (48.4% vs. 46.2%, P = 0.440) and the median number of biopsies 0 (0, 1) vs. 0 (0, 1), P = 0.472, the detection rate of high-risk lesions in the AI-assisted gastroscopy group was higher than that in the traditional gastroscopy group (59.6% vs. 51.8%, P = 0.028), and the detection rate of low-grade intraepithelial neoplasia in the AI-assisted gastroscopy group was significantly higher than that in the traditional gastroscopy group (33.9% vs. 24.2%, P = 0.027). In the AI-assisted gastroscopy group, the detection rate of precancerous lesions (25.2% vs. 17.7%, P = 0.024) and precancerous condition (27.5% vs. 20.1%, P = 0.032) was higher than that in the traditional gastroscopy group.

Conclusions AI-assisted digestive endoscopy significantly improves the detection rate of precancerous lesions and precancerous condition, and no adverse events caused by the system are observed throughout the study. It is safe and controllable, and has significant clinical application value.