Hybrid CNN-SVM Model for Brain Tumor Classification utilizing Different Datasets

Brain tumor is a pre-eminent deadly disease of human survival. Identifying the brain tumor at primordial stage is important for rapid remedy. CAD system classification of the brain tumor proves magnificent outcome though it has some limitations. To classify the brain tumor utilizing MRI, this paper proposes a custom CNN-SVM based hybrid classifier model and uses three different datasets for experiments. Hybrid framework is used to get superior prediction rather than utilizing only CNN. Firstly, MR images are preprocessed by image denoising, skull stripping, cropping and then resized 130 by 130 pixels in order to diminish computation. Secondly, augmentation is performed for all MR images which is advantageous to reduce overfitting problem in deep learning and increase the image number. Thirdly, feature extraction is accomplished by custom CNN model and classification is performed by SVM model. Proposed paper also discusses the variation of results of augmented data and without augmented data. Comparison of the proposed model and other related works are demonstrated. Obtained test results of three different datasets are 98.55% 94%, 96.85% sequentially. The effectiveness of this model is indicated by this improved test result which is advantageous for medical diagnosis process.