Hybrid model-based Brain Tumor Detection and Classification using Deep CNN-SVM

Brain tumors are the major and life-threatening diseases. Early detection and classification of brain tumors are essential steps for optimal medication or chemotherapy. However, manually detecting and classifying the various forms of brain tumors is time-consuming and may be inaccurate. The traditional classification methods need manual feature extraction, which requires the experimenter's experience and professional knowledge. As a result, the categorization rate of the shallow diagnostic model is low. To solve the problem, this research proposes an effective feature selection technique using a lightweight deep convolution neural network (CNN). The main idea of this proposed lightweight deep CNN is to achieve the highest accuracy with fewer parameters. The established method is separated into three sections: pre-processing, feature extraction step, and classification. In the pre-processing section data augmentation is utilized to enhance the number of images of the dataset and K-fold cross-validation is employed for the robust classification. In the feature extraction section, effective features are extracted using modified CNN and finally, SVM is employed to prominently classify the various forms of brain tumors from MRI scans. The proposed model CNN-SVM can effectively detect and classify the various forms of brain tumors and provides the best accuracy of 96.7%.