Usefulness of artificial intelligence in diagnosing early gastric cancer using magnifying endoscopy with narrow-band imaging.

BACKGROUND AND AIMS: In this study, we developed an artificial intelligence (AI) system to support the diagnosis of early gastric cancer using magnifying endoscopy with narrow-band imaging (NBI), with the goal of improving the diagnostic accuracy of upper gastrointestinal endoscopy. This study aimed to evaluate the diagnostic performance of the AI-based system compared with endoscopists.

METHODS

A total of 500 cases, including early gastric cancer and noncancer cases, were used to develop the diagnostic support AI system. We evaluated the AI system with models created using k-fold cross-validation. We determined the diagnostic performance of the AI system and 10 endoscopists (5 experts and 5 nonexperts) for detecting early gastric cancer and compared the diagnostic performance between the AI system and endoscopists. Histopathologic diagnosis was used as the reference standard for cancerous lesions, while magnifying endoscopic diagnosis with high confidence served as the reference standard for noncancerous lesions.

RESULTS

The median (interquartile range [IQR]) sensitivity for the AI system and all endoscopists was 100% (100%-100%) and 76.8% (65.5%-78.9%), respectively. The sensitivity of the AI system was significantly higher than that of the endoscopists (P = .002). The median (IQR) specificity for the AI system and all endoscopists was 100% (99.0%-100%) and 86.0% (72.0%-91.8%), respectively. The specificity of the AI system was also significantly higher (P = .01).

CONCLUSIONS

This study demonstrates the feasibility of applying the diagnostic support AI system during magnified observation with NBI and suggests that it could enhance the diagnostic accuracy of upper gastrointestinal endoscopy for early gastric cancer.