MR brain image tumor classification via kernel SVM with different preprocessing techniques

Abstract

Medical imaging interpretation and analysis requires automatic and exact classification. Several methods have been proposed in the last few years. This paper presents the effect of classification accuracy through different pre-processing techniques in the existing method tested on different Kernel SVM. Occurrence of irregular discontinuities causing bias field effect and intensity variations while capturing MR images requires the pre-processing of images. Three different preprocessing techniques such as Anisotropic diffusion, Homomorphic and Alphatrimmed filters are applied to brain MR images. We first enhance the attributes of the MRI image using these filters individually and then segments the tumor region. The relevant features are extracted from tumor regions and trained in a classifier. For feature extraction Wavelet transform is used, followed by feature reduction by using principle component analysis (PCA). The reduced features are trained with Kernel Support Vector Machine (KSVM) and classifies the tumor in MRI image as malignant and benign. We validate the performance of our approach on a dataset through multiple iterations to calculate the average classification accuracy subject to different preprocessing techniques.