Features extraction for classification of brain tumor MRI images using support vector machine

Abstract

This paper proposes to obtain the optimal quantifiable number of GLCM (gray level co-occurrence matrix) texture features for the medical MRI image data set to classify brain tumors as normal or abnormal. Initially, all the GLCM texture features from MRI images are extracted than the best features are extracted from MRI images, which in further given as input to a machine learning model, support vector machine (SVM) for the classification. Within this experiment, all of the known 19 GLCM texture features are initially extracted from the image, then the high-performance accuracy is computed by choosing a subset of the various number of features based on heat map. The experiment was carried out with 244 images consisting of 90 normal and 154 abnormal MRI images from the Kaggle dataset. It is observed that the SVM classifier accuracy increases with an increase in the number of features at 6-7 features, the testing accuracy of nearly 92 % is achieved than with the increase in number of features, then the accuracy stagnates.