AI-Assisted Endoscopy Improves Upper Gastrointestinal Cancer Detection Rates: Validated via Data from 50,000 Individuals in Large-Scale Screening Facility

A research group including Specially Appointed Associate Professor Ryo Nakagawa and Associate Professor Jun Kato at the Chiba University Graduate School of Medicine has demonstrated using real-world data from approximately 50,000 individuals that AI-assisted endoscopy significantly increases gastric cancer detection rates and improves biopsy positive predictive values. Previously, the efficacy of AI-assisted endoscopy was primarily validated through studies using recorded images. This study is significant for being the first in the world to demonstrate its effectiveness in actual screening environments using large-scale data. Future plans involve promoting the adoption of AI-assisted endoscopy and utilizing it as an educational tool to complement operator experience, thereby contributing to earlier and more certain detection of upper gastrointestinal cancers.

The findings were published in the international journal Digestive Endoscopy on May 1, 2026.

[Key Research Findings] This study analyzed 49,980 individuals who underwent upper gastrointestinal endoscopy at a large-scale screening facility between April 2021 and March 2024. A comparative analysis was conducted around April 2023, when the AI-assisted system was introduced. The pre-AI group (32,318 people) and post-AI group (17,662 people) were compared by extracting 17,662 pairs using propensity score matching. As a result, the detection rate of gastric cancer significantly increased by approximately three times after the introduction of the AI system, to 0.10% in the AI group compared to 0.03% in the non-AI group. Furthermore, the positive predictive value of biopsies (the proportion of cases diagnosed with cancer out of biopsies performed) also significantly increased to 4.84% in the AI group compared to 2.16% in the non-AI group for gastric and esophageal cancers combined, indicating that AI-assisted endoscopy leads to biopsies that are more diagnostic. Additionally, gastric cancers detected in the AI group were significantly smaller than those in the non-AI group, with a notably higher detection of small lesions 10mm or smaller. Additionally, examination time was 7.22 minutes in the AI group and 7.37 minutes in the non-AI group; thus, it was confirmed that AI introduction actually slightly shortened examination time and did not increase the burden on examination.

[Future Outlook] Based on these findings, future plans include expanding the use of AI-assisted endoscopy to community and workplace screenings, utilizing it as an educational tool to complement operator experience, and verifying its efficacy against other cancer types such as esophageal cancer, aiming to contribute to earlier and more certain detection of upper gastrointestinal cancers and improvement of patient prognosis.