Rapid Polyp Classification in Colonoscopy Using Textural and Convolutional Features.

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Colorectal cancer is a major global health concern.
• Untreated colon adenomatous polyps are a primary cause of colorectal cancer.
• Current polyp diagnosis relies on colonoscopy and biopsy, which can be time-consuming.

Purpose of the Study:

• To develop a quantitative polyp classification method using colonoscopy images.
• To provide gastroenterologists with a rapid and objective assessment tool for polyps.
• To differentiate between hyperplastic polyps and malignant adenomatous polyps and adenocarcinomas.

Main Methods:

• A database of 1991 colonoscopy images (1053 hyperplastic, 938 adenomatous/carcinomas) was collected.
• Extracted textural features were used in machine learning classifiers.
• Deep convolutional neural networks (DCNNs), including AlexNet, Inception-V3, ResNet-101, and DenseNet-201, were employed.
• AlexNet was trained from scratch for polyp classification.

Main Results:

• AlexNet trained from scratch achieved the highest accuracy of 96.4%.
• This performance surpassed methods using transfer learning and textural features alone.
• The DCNN models demonstrated effectiveness in quantitative polyp classification.

Conclusions:

• AI-powered image analysis can accurately classify colon polyps during colonoscopy.
• The developed models offer a rapid, objective tool for gastroenterologists.
• This technology can expedite treatment planning for colorectal cancer patients.