Brain MRI Classification for Detection of Brain Tumors Using Hybrid Feature Extraction and SVM

Abstract

Brain tumor is one form of brain abnormalities that affects the brain tissues. Magnetic resonance imaging (MRI) is the most popular imaging modality that capture and preserve greatest quality brain image with rich information that provides anatomical structures and internal contents of the brain. It is very challenging in the part of the radiologist to detect the abnormal structures of human brain using clinical expertise and manual image identification methods. Computer-assisted diagnosis (CAD), on the other hand, helps in early identification of brain disorders much faster and easier. In this paper, we proposed and studied a support vector machine (SVM)-based classifier model that implemented a hybrid feature extraction and reduction technique to categorize an input MRI images as normal or tumorous. The model initially implements two-dimensional discrete wavelet transform (2D-DWT) to extract the features of given input image followed by principal component analysis (PCA) to minimize the dimensions of extracted features. The experiments were conducted with 400 MR images comprising of 160 normal and 240 abnormal MRIs with brain tumors. Classifications were performed with five traditional classifiers, namely SVM, random forest, logistic regression, backpropagation neural network (BPNN), and K-nearest neighbor (KNN). On evaluating the classifier models, the proposed SVM-based model achieved better performance compared to the rest of classifiers in terms of accuracy, specificity and sensitivity, and the area under curve (AUC) values.