A Systemic Approach for Brain Tumor Classification

Abstract

In recent years, research into human brain tumors has greatly benefited from newlinemedical image-based analysis employing Computer-Aided Diagnosis (CAD). The earlier newlinea brain tumor is detected, the more likely it is to be treated successfully and save lives. newlineBrain cancers should be identified using precise, efficient segmentation and classification newlineprocedures availing Magnetic Resonance Imaging (MRI) help. The classification of newlineabnormal and normal brain MR images has been the focus of recent studies. This study newlineoffers segmentation and subsequent classification of brain MRI images for more newlineeffective and precise classification. The proposed methodology categorizes tumor to newlinedetermine whether it is normal or pathological. Further, to find out how each model newlineperforms, we study various strategies, such as interactive segmentation, hybrid newlinesegmentation, machine learning, deep learning, and transfer learning classification newlinetechniques. Three alternative methods are put forth to classify brain tumors. newlineThe first method suggests a framework, including pre-processing, segmentation, newlinefeature extraction, and classification. Using magnetic resonance imaging, this framework newlinehas been utilised to detect cancerous development in the cerebrum. In order to preprocess newlinethe images, methods including thresholding, gaussian, and median filters are newlineused. Then, a hybridised segmentation technique is used to precisely segment the region newlineof interest, including improved K-Means and fuzzy C-Means. Valuable image features newlineare retrieved using techniques like the GLCM and LBP operator. The features are finally newlinesubjected to an SVM classifier to determine whether the tumor is benign or malignant. newlineThis method produces better results for dataset BRATS, but it does not work as well for dataset Figshare. newlineThe second method suggests deep learning (DL) approaches. An effective CNN newlinedesign has been employed for classifying the tumor. This traditional CNN design is newlinereorganized with five convolutional layers, four pooling layers, and one fully connected newlineneural network