Detection of Brain Tumor using Machine Learning Classifiers – A Comparative Study

Abstract

The machine leaning (ML) and deep learning (DL) classifiers are effectively used for brain tumor identification in its early stages which is a challenging task. In this work, a comparative study is done among the different types of ML classifiers to find an optimal classifier. The Magnetic Resonance (MR) images are first skull stripped to avoid the unnecessary computation. Then it undergoes the segmentation process to obtain the region we are interested to operate. The Gray Level Co-occurrence Matrix (GLCM) method is applied to get the features from the segmented regions. The extracted features are given to ML classifiers such as Decision Tree, Discriminant Model, K-Nearest Neighbor (kNN), Ensemble, Support Vector Machine (SVM), Naive Bayes, Neural Network (NN) and Kernel. Feature extraction is a time-consuming process. Hence a deep NN is used here to predict the severity of the disease. By using CNN layers, the image is predicted whether it is benign tumor or malignant tumor. The DL classifier extracts features automatically but gives less accuracy compared to ML classifiers. Hence, for combining the advantages of both, the hybrid DL classifiers are implemented to identify the brain tumor. After a comparison among ML classifiers, DL classifier and Hybrid classifiers, it is found that a Hybrid classifier with the combination of DL and Decision Tree gives the best results with accuracy of 98.8%, sensitivity of 100%, specificity of 9S.1%, precision of 9S.1%, NPV of100% and PPV of 9S.3%.