Skip to main content

Performance of an Artificial Intelligence-Based Diagnostic Support Tool for Early Gastric Cancers: Retrospective Study

Authors
Toshiaki Hirasawa1, Mitsuaki Ishioka1, Hiroyuki Osawa2, Tomohiro Tada3,4,5

Affiliations

  1. Departments of Gastroenterology, Cancer Institute Hospital of the Japanese Foundation for Cancer Research.
  2. Division of Gastroenterology, Department of Medicine, and 7Department of Pathology, Jichi Medical University
  3. AI Medical Service I
  4. Department of Surgical Oncology, Graduate School of Medicine, The University of Tokyo
  5. Tada Tomohiro Institute of Gastroenterology and Proctology

Introduction
The ability of endoscopists to diagnose early gastric cancers (EGCs) varies significantly, especially between specialists and nonspecialists. To address this issue, we developed "Tango," an artificial intelligence (AI)-based diagnostic support tool designed to effectively differentiate EGCs. This study evaluates the performance of Tango compared to that of endoscopists, aiming to validate its effectiveness.

Methods
This research compared the diagnostic performances of Tango and endoscopists (34 specialists and 42 nonspecialists) using still images of 150 neoplastic lesions (including EGCs and adenomas) and 165 non-neoplastic lesions. The primary objective was to demonstrate the noninferiority of Tango based on sensitivity compared to specialists. Secondary objectives included establishing the noninferiority of Tango based on accuracy against specialists and its superiority in terms of sensitivity and accuracy over nonspecialists.

Results
Tango demonstrated superior sensitivity (84.7%) compared to specialists (65.8%), with a statistically significant difference (18.9%, 95% CI 12.3-25.3%). Tango also showed noninferior accuracy (70.8% compared to 67.4% for specialists). Against nonspecialists, Tango exhibited superiority in both sensitivity (84.7% vs. 51.0%) and accuracy (70.8% vs. 58.4%).

Conclusion
In diagnosing EGCs, Tango outperformed endoscopists, indicating its potential to reduce misdiagnoses, especially among nonspecialists. These findings suggest that AI-based tools can play a crucial role in the diagnosis of EGCs, warranting further validation in future clinical trials.