A colonial serrated polyp classification model using white-light ordinary endoscopy images with an artificial...

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate classification of colonic polyps is crucial for preventing colorectal cancer.
• Distinguishing between traditional adenoma (TA), sessile serrated adenoma (SSA), and hyperplastic polyp (HP) can be challenging.
• AI offers potential for improving diagnostic accuracy in endoscopy.

Purpose of the Study:

• To develop and evaluate an AI model for classifying colonic polyps.
• To assess the effectiveness of data augmentation in improving AI model performance.
• To aid physicians in differentiating between TA, SSA, and HP.

Main Methods:

• Collected endoscopy images under white light and NBI.
• Utilized a Convolutional Neural Network (CNN) model, specifically Inception V4.
• Implemented data augmentation techniques for model training.
• Trained the final AI model using only white light images.

Main Results:

• The AI classification model achieved 94% prediction accuracy for colon polyp types.
• The model demonstrated 98% discriminability (area under the curve).
• The final model was constructed using white light images and data augmentation.

Conclusions:

• The developed AI model can assist physicians in classifying colonic polyps.
• The model shows high accuracy in distinguishing between TA, SSA, and HP.
• This AI tool has the potential to improve the identification of precancerous lesions like TA and SSA.