Computer-aided Diagnosis of Polyp Classification Using Scale Invariant Features and Extreme Gradient Boosting.

Area of Science:

• Medical Imaging
• Machine Learning
• Computational Pathology

Background:

• Colorectal cancer diagnosis relies on accurate colonoscopy image analysis.
• Machine learning models have shown promise in classifying colorectal polyps.
• Advancements in technology enable sophisticated analysis of endoscopic imagery.

Purpose of the Study:

• To develop a framework for classifying polyps from colonoscopy images.
• To enhance the diagnostic accuracy of colorectal cancer identification.
• To evaluate the performance of a novel polyp classification approach.

Main Methods:

• Colonoscopy images were preprocessed to remove spectral reflections.
• Scale-invariant features were extracted using Radon transform and Zernike moments.
• Extreme Gradient Boosting (XGBoost) algorithm was employed for polyp classification.

Main Results:

• The proposed framework achieved a classification accuracy of 93% with light XGBoost and 92% with XGBoost.
• Cross-validation confirmed the robustness of the experimental results.
• The method demonstrated superior performance compared to existing state-of-the-art techniques.

Conclusions:

• XGBoost, combined with scale-invariant features, effectively classifies polyps from colonoscopy images.
• The developed framework offers a promising tool for improving colorectal cancer diagnosis.
• This approach shows potential for application in clinical settings with small datasets.