Brain Tumor Classification with MRI Brain Images Using 2-Level GLCM Features and Sparse Representation based Segmentation

Abstract

Recently, detection of brain tumor using imaging techniques is a trending research area in the field of computer vision. Computer vision technology is popularly being used for tumor classification since the accuracy of diagnosis is very high. This research work proposes a novel scheme called Sparse Representation Segmentation with 2-Level GLCM and SVM classification (S2LGSVM) system for the segmentation and classification of brain. The MRI images are initially preprocessed using filtration and enhancement techniques. The enhanced MRI data are then segmented using a novel Sparse Representation Based Segmentation (SRBS) algorithm. The segmented data is used for feature extraction. A novel 2-Level Gray Level Co-Occurrence Matrix (2L-GLCM) features are proposed for representing the segmented data. The 2L-GLCM features are obtained by first finding GLCM matrix for the MRI data. The same algorithm is then applied at the second level to extract the 2L-GLCM matrix. Finally, support vector machine (SVM) classifier classifies the extracted features. The images are classified into malignant and benign categories. Performance analysis is done using publicly available OASIS dataset. The acceptability of the proposed framework is evaluated and compared with other techniques used in the literature. The proposed SRBS segmentation achieved a maximum Jaccard Coefficient of 0.8167. The proposed S2LGSVMframework achieved an accuracy of 96.92%, specificity of 97.93% and precision of 95.82%.