Convolutional Neural Networks for Automated Classification of Brain Tumors

Abstract

Image classification is a fundamental task in computer vision that has seen tremendous advances with the advent of deep learning. In the medical imaging domain, image classification using deep neural networks has shown potential for automated analysis and diagnosis. In this work, we investigate convolutional neural networks (CNNs) for brain tumor image classification using magnetic resonance imaging (MRI) scans. The accurate classification of brain tumors from MRI scans enables non-invasive screening and diagnosis, thus improving clinical workflows and patient outcomes. We provide a review of CNN architectures for image classification, and discuss relevant datasets for brain tumor MRI analysis. We identify limitations of prior arts and propose a methodology leveraging CNNs for enhanced multi-class brain tumor classification performance. Our contributions include comparative evaluations of CNN architectures, optimization techniques, and preprocessing strategies. Extensive experiments demonstrate that our proposed approach achieves state-of-the-art classification accuracy on benchmark datasets. We employ data augmentation, batch normalization, and dropout regularization to prevent overfitting and improve generalization. The ReLU activation and Adam optimization enable efficient training. Qualitative and quantitative results validate the efficacy of our methodology, while ablation studies provide insights into model components. We conclude by summarizing our findings, and discussing potential extensions as well as limitations of our approach to guide future works. Our work aims to advance research on employing deep CNNs for computer-aided diagnosis using MRI scans.