Endoscopic Artificial Intelligence for Image Analysis in Gastrointestinal Neoplasms

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Deep learning-based AI is increasingly used in medical fields, particularly in gastroenterology for gastrointestinal (GI) endoscopy.
• Computer-aided detection/diagnosis (CAD) systems assist endoscopists in identifying and differentiating GI neoplasms.
• AI systems for colorectal polyps are in clinical practice, and CAD systems for upper GI neoplasms are available in Asian countries.

Purpose of the Study:

• To review recent studies on endoscopic AI systems for GI neoplasms.
• To focus on esophageal squamous cell carcinoma (ESCC), esophageal adenocarcinoma (EAC), gastric cancer (GC), and colorectal polyps.
• To evaluate the performance and utility of AI in endoscopic diagnostics.

Main Methods:

• Review of recent articles on endoscopic AI systems.
• Analysis of meta-analyses and randomized controlled trials (RCTs) for CADe and CADx systems.
• Focus on diagnostic performance metrics such as sensitivity and specificity.

Main Results:

• CADe systems for ESCC and EAC show high sensitivities (91.2%, 93.1%) and specificities (80%, 86.9%).
• CADe systems for GC achieve high sensitivity and specificity (around 90%), with RCTs demonstrating reduced miss rates.
• CADx systems for GC show expert-level performance, and AI systems for colorectal polyps improve detection rates and differentiation accuracy.

Conclusions:

• Endoscopic AI systems generally perform better than nonexpert endoscopists and are comparable to expert endoscopists.
• AI tools can significantly reduce the risk of overlooking GI lesions.
• These systems hold promise for enhancing diagnostic accuracy in GI endoscopy.