Using Machine Learning Classifier and Statistical Technique To Detect and Classify The Human Brain Tumor

Abstract

Medical imaging, specifically Magnetic Resonance Imaging (MRI), has evolved into a fundamental aspect of brain tumor diagnosis and treatment. The integration of deep learning technologies, such as convolutional neural networks (CNNs), has revolutionized the automated analysis of MRI images, facilitating accurate detection and classification of brain tumors. This thesis investigates the efficacy of three prominent CNN architectures, namely U-Net, VGG-16, and ReSeg, in the context of brain tumor classification and detection from MRI scans. The study is motivated by the pressing need for accurate and efficient methods to assist radiologists in early detection and characterization of brain tumors, thus improving patient outcomes. The research methodology encompasses several key phases. Firstly, a thorough examination of the available literature is undertaken to establish a theoretical foundation and pinpoint areas where current research on brain tumor detection and classification utilizing deep learning methods may be lacking. Subsequently, a diverse dataset of MRI scans comprising both benign and malignant brain tumors is curated from multiple clinical sources, ensuring variability in tumor types, sizes, and imaging modalities. Preprocessing techniques, including image normalization, augmentation, and segmentation, are employed to enhance the quality and consistency of the dataset, mitigating potential biases and confounding factors. The core of the research involves the implementation and evaluation of the U-Net, VGG-16, and ReSeg models for brain tumor classification and detection tasks. Each CNN architecture is meticulously configured and trained using the curated dataset, employing appropriate hyperparameters and optimization algorithms to maximize performance metrics such as accuracy, sensitivity, specificity, and computational efficiency. The training process is conducted on high-performance computing platforms to leverage parallel processing capabilities and expedite model convergenceFollowing model training,