MR Brain Tumor Classification and Segmentation Via Wavelets

Abstract

Timely, accurate detection of magnetic resonance (MR) images of brain is most important in the medical analysis. Many methods have already explained about the tumor classification in the literature. This paper explains the method of classifying MR brain images into normal or abnormal (affected by tumor), abnormality segments present in the image. This paper proposes DWT-discrete wavelet transform in first step to extract the image features from the given input image. To reduce the dimensions of the feature image principle component Analysis (PCA) is employed. Reduced extracted feature image is given to kernel support vector machine (KSVM) for processing. The data set has 90 brain MR images (both normal and abnormal) with seven common diseases. These images are used in KSVM process. Gaussian Radial Basis (GRB) kernel is used for the classification method proposed and yields maximum accuracy of 98% compared to linear kernel (LIN). From the analysis, compared with the existing methods GRB kernel method was effective. If this classification finds abnormal MR image with tumor then the corresponding part is separated and segmented by thresholding technique.