Improving machine learning recognition of colorectal cancer using 3D GLCM applied to different color spaces

• Published in: Multimedia tools and applications 2022-03, Vol.81 (8), p.10839-10860
• Main Authors: Alqudah, Ali Mohammad, Alqudah, Amin
• Format: Article
• Language: English
• Subjects: Algorithms; Artificial neural networks; Cancer; Channels; Colon; Color; Colorectal cancer; Computer Communication Networks; Computer Science; Data Structures and Information Theory; Datasets; Decision analysis; Decision trees; Discriminant analysis; Feature extraction; Intestine; Learning theory; Machine learning; Multimedia Information Systems; Special Purpose and Application-Based Systems; Support vector machines; Texture; Training
• ISSN: 1380-7501; 1573-7721
• DOI: 10.1007/s11042-022-11946-9

Colorectal cancer (CRC) is one of the widely happening cancers among men and women. This cancer, which is also known as bowel cancer, affects the human large intestine, especially the rectum or colon. Therefore, providing new techniques with high accuracy to detect CRC cancer leads to providing an early and successful plan to treat it. In this research, we proposed a method to classify colorectal cancer using different machine learning algorithms. The method uses extracted features from 3D Gray Level Cooccurrence Matrix (GLCM) matrices of three different color spaces namely RGB, HSV, and L*A*B colors spaces. The 3D GLCM matrices of the used images were calculated and evaluated using a training dataset of 3504 images and a testing dataset of 1496 images. The five different widely used machine learning algorithms, which are Support Vector Machine (SVM), Artificial Neural Network (ANN), K-Nearest Neighbor (KNN), Quadratic Discriminant Analysis (QDA), and Classification Decision Tree (CDT). The results show that the proposed methodology can detect CRC with a high-performance rate. This higher rate is due to combining texture features from all color space channels. The best performance rate for the used machine learning models was greater than 97% for the training and testing sets using QDA using RG. The obtained results show that the proposed methodology can be used efficiently to detect CRC with high performance compared to all previous methods since texture features from the three color space channels. This research represents the first of its kind in the current research trend.