Real-Time Artificial Intelligence-Based Histologic Classifications of Colorectal Polyps Using Narrow-Band Imaging

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Gastroenterology

Background:

• Advancements in artificial intelligence (AI) enable real-time computer-aided detection (CAD) in clinical settings.
• The study focuses on developing an AI-based CAD system (CAD-N) optimized for narrow-band imaging (NBI).

Purpose of the Study:

• To develop and validate an AI-based CAD-N model for the real-time histologic classification of colorectal polyps.
• To optimize the diagnostic performance of the AI model using NBI images.

Main Methods:

• A CAD-N model was developed using ResNeSt architecture with NBI images.
• The model was trained to classify colorectal polyps into four types based on histopathology.
• 116 polyp videos were collected for real-time accuracy validation.

Main Results:

• The AI model analyzed 10,573 images from 478 patients, achieving high sensitivity, specificity, and accuracy for all polyp types.
• Overall accuracy for histologic classification was 93%.
• For polyps ≤5 mm, accuracy was 95.59%; video validation showed 85.34% accuracy.

Conclusions:

• A real-time AI-based system (CAD-N) for histologic classification of colorectal polyps using NBI images has been successfully developed.
• The system demonstrates good accuracy, potentially assisting in clinical management and documentation of optical histology findings.