Texture Feature Based Colonic Polyp Detection And Classification Using Machine Learning Techniques

Having a high mortality rate, colorectal cancer (CRC) is the second most prevalent disease in women and the third most prevalent cancer in males. The majority of CRCs are derived from intestinal or colonic polyps. Colorectal cancer, commonly referred to as bowel cancer, is a type of cancer that arises when cells in the colon, rectum, or appendix grow out of control. Early colonic polyp detection and excision is the most effective strategy to prevent CRC. In the proposed study, feature extraction is carried out using local feature descriptors. The feature selection is carried out using the correlation filter method- and wrapper-based techniques and then applied to machine learning algorithms that can be utilized to identify and classify polyps. The Kvasir [1] and CVC Clinic dataset is used in the proposed effort to categorize the polyps. Before feature selection, features are extracted using local feature descriptors such as the Gray Level Co-occurrence Matrix (GLCM) [2], Local Binary Pattern (LBP) [3], and Histogram of Oriented Gradients (HOG) [4] we get 1789 features. Then correlation-based feature selection (CFS) [5] is used to find the relevant features. After that apply a wrapper-based strategy to locate the most relevant feature by using the Naive Bayes classifier. The model performances are assessed using Bayes Network, Naive Bayes, Stochastic Gradient Descent, Support Vector Machine, Random Forest, and Logistic Model Tree. The three types of polyps can be classified in the proposed model. They are large polyps, normal polyps, and small polyps. It is evaluated by using the evaluation metrics such as Recall, Precision, Accuracy, Specificity, F-Score, Area under the Curve and Kappa score. The Bayes Network is the best classifier and shows an accuracy of 99.3%.