Artificial Intelligence-Based Colorectal Polyp Histology Prediction by Using Narrow-Band Image-Magnifying Colonoscopy

Area of Science:

• Gastroenterology
• Artificial Intelligence in Medicine
• Medical Imaging Analysis

Background:

• Colorectal polyps require accurate histological classification to guide treatment.
• Narrow Band Imaging (NBI) magnifying colonoscopy is a key tool for in-vivo polyp assessment.
• Distinguishing between hyperplastic and neoplastic polyps is crucial for patient management.

Purpose of the Study:

• To develop and evaluate an artificial intelligence based polyp histology prediction (AIPHP) method for NBI colonoscopy images.
• To compare the diagnostic accuracy of AIPHP with the Narrow-band Imaging International Colorectal Endoscopic (NICE) classification.
• To assess the performance of AIPHP and NICE classification in predicting polyp histology.

Main Methods:

• A study involving 373 colorectal polyps from 279 patients undergoing polypectomy.
• Analysis of NBI still images using both the developed AIPHP software and the NICE classification.
• AIPHP software utilizes machine learning to measure five geometrical and color features from endoscopic images.

Main Results:

• The overall accuracy of AIPHP was 86.6%.
• AIPHP showed higher accuracy in non-diminutive polyps (92.2%) compared to diminutive polyps (82.1%).
• The NICE classification achieved significantly higher accuracy in predicting hyperplastic histology than AIPHP for both diminutive and all polyps.

Conclusions:

• The developed AIPHP software demonstrated high accuracy in predicting polyp histology, particularly in larger polyps.
• The NICE classification outperformed AIPHP in predicting hyperplastic polyp histology.
• Further refinement of AI algorithms may be needed to improve accuracy across all polyp sizes and types.