Real-time artificial intelligence-based histologic classification of colorectal polyps with augmented visualization.

Area of Science:

• Gastroenterology
• Artificial Intelligence
• Medical Imaging

Background:

• Artificial intelligence (AI)-based computer-aided diagnostic (CADx) algorithms show promise for real-time histology (RTH) of colonic polyps.
• Current methods lack transparency and interpretability in AI-driven RTH.

Purpose of the Study:

• To present a novel in situ CADx approach for colonic polyp histology.
• To enhance transparency and interpretability through augmented visualization of predicted histology.

Main Methods:

• Developed a deep learning model using semantic segmentation for polyp delineation.
• Classified polyp subregions and aggregated them into a surface histology map.
• Trained and validated the model on 740 images from 607 polyps.

Main Results:

• The model achieved high sensitivity (.96) and specificity (.84) in distinguishing neoplastic from non-neoplastic polyps.
• Achieved a negative predictive value (NPV) of .91 and high-confidence rate (HCR) of .88.
• Demonstrated strong performance on smaller polyps (≤5 mm) as well.

Conclusions:

• The CADx model accurately distinguishes polyp types and provides a spatial histology map.
• This approach improves interpretability and transparency of AI-based RTH.
• Offers intuitive, real-time guidance for clinical management and documentation of optical histology.