Artificial intelligence-based diagnosis of abnormalities in small-bowel capsule endoscopy

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Gastroenterology

Background:

• Deep learning-based artificial intelligence (AI) is crucial for advancing automated diagnosis in medical imaging.
• Small-bowel capsule endoscopy (SBCE) requires improved diagnostic tools for identifying multiple abnormalities.
• Current AI models need further development for accurate and simultaneous diagnosis of diverse SBCE findings.

Purpose of the Study:

• To develop an AI model for automatic diagnosis of multiple abnormalities in SBCE videos.
• To compare the AI model's diagnostic performance against physicians with varying experience levels.
• To evaluate the AI model's efficacy as an auxiliary tool for endoscopists.

Main Methods:

• Trained an AI model using 280,426 images from 2,565 patients.
• Validated the AI model's diagnostic performance on 240 SBCE videos.
• Compared AI-assisted diagnosis accuracy with unassisted diagnosis by junior doctors and expert endoscopists.

Main Results:

• The AI model demonstrated high sensitivity (94.7%-100%) and specificity (67.7%-97.5%) across various abnormalities.
• AI model achieved high accuracy rates for detecting red spots, inflammation, blood content, and lesions.
• AI assistance improved junior doctors' accuracy from 85.5% to 97.9%, comparable to expert performance (96.6%).

Conclusions:

• A well-trained AI model can automatically diagnose multiple small-bowel abnormalities from SBCE videos.
• The AI model shows significant potential as an auxiliary diagnostic system for endoscopists.
• AI-powered tools can enhance diagnostic accuracy and support less-experienced clinicians in SBCE interpretation.