CLASSIFICATION OF BRAIN TUMOUR IN MAGNETIC RESONANCE IMAGES USING HYBRID KERNEL BASED SUPPORT VECTOR MACHINE

Medical image segmentation is a knotty and challenging task. Predominantly, the brain has a complicated structure and its exact segmentation is very essential for identifying the tumours, edemas, and necrotic tissues in order to provide proper treatment. In this paper, we have proposed a novel brain tumour classification of MR images using texture features and hybrid kernel based SVM. Our proposed approach comprises the following major steps: i) preprocessing, ii) tumour region location iii) feature extraction and iv) final classification. In preprocessing steps, anisotropic filtering will be applied to diminish the noise and improve quality of the image for further processing. In the next steps to perform the skull stripping and tumour regions are identified using regionprops algorithm. In feature extraction some specific feature will be extracted using GLCM (Gray Level Co-occurrence Matrix). In the classification stage, the hybrid kernel will be designed and applied to training of support vector machine (SVM) to perform automatic detection of tumour in MRI images. For comparative analysis, our proposed approach is compared with the existing works. The accuracy level (93%) for our proposed approach is good at detecting the tumours in the brain MRI images.