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

Analysis of colonoscopy images is an important diagnostic procedure in the identification of colorectal cancer. It has been observed that owing to advancements in technology, numerous machine-learning models now excel in the analysis of colorectal polyps classification. This work focused on developing a framework that can classify polyps using images during colonoscopy.

First, the images were corrected by removing their spectral reflection. Second, feature pools were obtained by applying Radon transform (θ=45, 90, 135, and 180). From the Radon transform, fractal dimension was calculated as a feature vector combined with Zernike moment obtained from the Zernike features. Finally, Extreme Gradient Boosting (XGBoost) algorithm was applied for the classification and to compare it with state-of-the-art methods.

The experimental results obtained with the proposed framework have been reported, cross-validated, and discussed. The proposed method gives a classification accuracy of 93% for light XGBoost and 92% for XGBoost.

This study shows that by applying scale invariant features over a small dataset, XGBoost outperforms state-of-the-art methods when it comes to polyp classification.