New Novel Segmentation and Classification of MR Brain Images using DTCWT and KSVM

Brain tumor is considered as the most critical diseases in the medical era. The formation of abnormal cells within the brain results in brain tumors. It is a mass of tissues that results in hormonal changes results in mortality. There are several works evolved for brain image segmentation and classifications. We propose a novel classification technique is proposed to classify as normal and abnormal for the given set of brain MR images. At first, the pre-processing is performed by median filtering, and segmentation is by employing the morphological technique. Then the dual-tree complex wavelet transform (DTCWT) is applied to extract the features, which are then given to a kernel support vector machine (KSVM) for classification. To improve the simplification of KSVM, K-fold cross-validation is used. Four common brain tumor diseases such as Alzheimer’s disease, glioma, meningioma, metastatic, and sarcoma are taken as abnormal images. 200 brain MR images are collected from the Harvard Medical School website, in which 20 images are normal, and 180 images are abnormal. The classification of KSVM is performed with various kernels such as Gaussian Radial Basis (GRB), Homogeneous Polynomial (HPOL), and Inhomogeneous Polynomial (IPOL), in which the GRB attains the highest accuracy of 99.42%. The other two kernels HPOL and IPOL, attain 96.7% and 98.05, respectively. In which the proposed DTDWT+GRB provides better accuracy than the conventional methods.