AI-Assisted Endoscopy Reduces Blind Spots, Not Missed Gastric Neoplasia : Gastroenterology | June 2026

Introduction:

Gastric Cancer remains a major global cause of cancer mortality, particularly in East Asia, where early detection through upper gastrointestinal endoscopy is central to improving outcomes. Artificial intelligence (AI)-assisted endoscopy has emerged as a promising strategy to enhance lesion recognition and reduce missed neoplasia during routine gastroscopy.

Problem Statement:

Although multiple retrospective and single-center studies have suggested that AI may improve upper gastrointestinal lesion detection, robust evidence from large multicenter randomized controlled trials evaluating its real-world clinical effectiveness has been lacking. Whether AI meaningfully improves true gastric neoplasm detection beyond enhancing procedural quality metrics remains uncertain.

Summary:

This large multicenter randomized controlled trial evaluated the real-world impact of AI-assisted esophagogastroduodenoscopy on gastric neoplasm detection across 24 hospitals in China.

A total of 29,514 patients were randomized to either AI-assisted or conventional upper endoscopy, making this one of the largest prospective studies assessing AI integration into gastrointestinal endoscopy practice.

The primary endpoint was pathologically confirmed gastric neoplasm detection after expert pathological review.

Importantly, AI assistance did not significantly improve the final detection rate of gastric neoplasms after pathological confirmation, with rates of 1.42% in the AI group versus 1.25% in the conventional group.

However, AI did improve lesion detection before central pathological review, suggesting that AI may enhance initial lesion recognition during live procedures.

One of the most important procedural findings was the marked reduction in endoscopic blind spots with AI assistance, decreasing from 2.52 to 1.07 blind spots per examination.

This suggests that AI significantly improves procedural completeness and mucosal visualization quality during upper endoscopy.

AI-assisted procedures were associated with longer inspection and procedural times, indicating that enhanced scrutiny may partially explain improvements in lesion recognition.

Subgroup analyses provided clinically relevant insights, demonstrating potential benefit among less experienced endoscopists and during fatigue-associated procedural periods.

These findings support the concept that AI may function most effectively as a quality-support tool rather than a replacement for expert endoscopic interpretation.

Importantly, AI demonstrated excellent sensitivity for advanced gastric lesions, correctly identifying 100% of gastric adenocarcinomas and over 90% of high-grade intraepithelial neoplasia cases confirmed on pathology.

Performance was less robust for low-grade intraepithelial neoplasia, highlighting ongoing limitations in detecting subtle early lesions.

The study is highly relevant because it challenges the widespread assumption that AI universally improves clinically meaningful gastric cancer detection rates.

Instead, the findings suggest that the current generation of AI systems may primarily enhance procedural standardization, inspection quality, and operator vigilance rather than independently increasing definitive neoplasm detection.

Clinically, the data indicate that AI may have greatest utility in community settings, lower-volume centers, or among less experienced endoscopists where variability in inspection quality is more pronounced.

The reduction in blind spots is particularly important because missed lesions remain a major contributor to post-endoscopy upper gastrointestinal cancers.

The trial also highlights a critical issue in AI research: improvements in surrogate procedural metrics do not always translate into improved clinically validated outcomes.

Future development will likely require more sophisticated multimodal AI platforms capable of integrating lesion morphology, mucosal pattern recognition, and real-time histologic prediction.

Further real-world validation studies are needed to determine whether AI-assisted systems can improve long-term gastric cancer outcomes, interval cancer rates, and cost-effectiveness in routine practice.

Overall, this landmark randomized trial demonstrates that AI-assisted gastroscopy improves procedural quality and reduces blind spots but does not yet significantly increase pathologically confirmed gastric neoplasm detection, underscoring both the promise and current limitations of AI-enabled endoscopy.