Development of a Novel Approach for Classification of MRI Brain Images Using DWT by Integrating PCA, KSVM and GRB Kernel

Abstract

In today’s world, automated computation is crucial in medical field and precise classification of magnetic resonance (MR) brain images is equally imperative in medical analysis and its interpretation by medical practitioners. In this context, various techniques have already been proposed. This research article focuses on the wavelet transform technique which is first applied on MR images for feature extraction followed by principal component analysis (PCA) technique which simplifies the datasets as it is a dimensionality reduction technique for features which in turn increases the discriminative power. The reduced features are further applied to kernel support vector machine (KSVM) for classification of image as normal/abnormal. The experimentation has been applied to four different kernels namely linear kernel (LIN), homogeneous polynomial (HPOL), inhomogeneous polynomial (IPOL, and Gaussian radial basis (GRB). The overall results show that the accuracy of GRB kernel is best and the processing time has also been reduced. The proposed technique has been compared with the other techniques available in literature and found that use of DWT, PCA and KSVM along with GRB kernel obtained the best results. Considerable amount of datasets have been used in the experimentation and it can be concluded from the results that by using the proposed methodology, the time taken for classification of segmented image is reduced drastically which could be the turning point in medical field for diagnosis of tumor.