Artificial Intelligence-Assisted Endoscopy Reduces Gastric Neoplasm Miss Rate

Investigators assessed the effects of an artificial intelligence system on improving detection of gastric neoplasms and reducing the miss rate.

The gastric neoplasm miss rate during upper gastrointestinal endoscopy was significantly reduced by the assistance of an artificial intelligence (AI) system, according to a study published in The Lancet Gastroenterology and Hepatology.

In a single-center trial conducted at Renmin Hospital of Wuhan University, China, patients were randomly assigned to receive either AI-assisted (via the ENDOANGEL-LD system) or routine white light endoscopy, followed immediately by the other procedure. Only the patients and pathologists were blinded to the procedure and targeted biopsies for all detected lesions were taken after the second procedure. The primary endpoint was the miss rate of gastric neoplasms.

Overall, 907 patients were included in the AI-first group and 905 in the routine endoscopy-first group. The number of detected neoplasms across both groups after both examinations was 93, with 49 gastric neoplasms diagnosed in 47 patients from the AI-first group and 44 neoplasms in 43 patients from the routine-first group.

The miss rate was significantly lower in the AI-first group (relative risk, 0.224; 95% CI, 0.068–0.744; P =.015). In the AI-first group, 3 of 47 patients with neoplasms (6.4%) were missed while 11 of 43 (25.6%) in the routine-first group were missed (P =.024). The difference in detection rate was not statistically significant, though numerically higher for the AI-first group.  However, no superficial depressed-type or superficial elevated- and depressed-type neoplasms were missed in the AI-first group, while 57.1% (95% CI, 20.2–88.2) of superficial depressed-type neoplasms and 40.0% (95% CI, 7.3–83.0) of superficial elevated- and depressed-type neoplasms were missed in the routine-first group. The only adverse event reported was bleeding after biopsy from a targeted lesion in the routine-first group.

The study had several limitations. Some lesions missed in the first exam may have also been missed in the second. Additionally, there was potential for observer bias, as the same endoscopists performed both examinations.

Based on these results researchers concluded the ENDOANGEL-LD system, “could effectively reduce the miss rate of gastric neoplasms and minimise unnecessary biopsies without adding inspection time.” They recommend lager, multi-center studies to determine scalability and generalizability.


Wu L, Shang R, Sharma P, et al. Effect of a deep learning-based system on the miss rate of gastric neoplasms during upper gastrointestinal endoscopy: a single-centre, tandem, randomised controlled trial. Lancet Gastroenterol Hepatol. Published online July 20, 2021. doi:10.1016/S2468-1253(21)00216-8