Robust classification of Multi Class brain Tumor in MRI images using Hybrid Descriptor and Pair of RBF Kernel - SVM

摘要

In medical imaging, detecting and classifying the brain tumors in Magnetic Resonance Image (MRI) is a demanding and critical task. MRI gives anatomical structure's information, and the potential abnormal tissues' information. Thus, this paper proposes a new system for MRI brain tumor segmentation and classification. This work includes the following stages: preprocessing, segmentation, extraction of feature, selection of a feature, and classifying the images. Removing of Speckle and white Gaussian noise in the given MRI images is done in the preprocessing stage, by using the Distribution based Adaptive Filtering (DAF) technique. It smoothens the image by removing the noise and enhancing the intensity of the image. In segmentation stage, the clustering and label formation processes are performed to predict the tumor part. Here, the Neighboring Cellular Automata (NCA) model is proposed for clustering. Then, the labels such as Back Ground (BG), border area, Gray Matter (GM) and White Matter (WM) are formed for the clustered image. Hence, the features of the segmented image are extracted by using the Differential Binary Pattern (DBP) technique. After extracting the feature vectors, the firefly optimization technique is employed to select the best features. After selecting the set of features, the Pointing Kernel Classifier (PKC) is employed to classify both the abnormal and normal brain images and the type of brain tumors. The performance of the proposed method is evaluated using sensitivity, specificity, accuracy, correction rate, positive likelihood and negative likelihood.