Vision Transformers for Brain Tumor Classification: A Novel Deep Learning Approach

Brain tumors are a highly aggressive and prevalent form of the central nervous system (CNS), comprising approximately 85-90% of cases. With around 11,700 new diagnoses annually, the prognosis remains grim, with a low survival rate of 34% for men and 36% for women. Brain tumors encompass various subtypes, including pituitary tumors, meningiomas, and gliomas, each with distinct characteristics, treatments, and outcomes. Magnetic Resonance Imaging (MRI) is widely regarded as the most accurate and reliable imaging modality for the diagnosis of brain tumors, earning it the distinction of being the gold standard in this field. However, manual review by radiologists can be error-prone due to the complexity of these tumors. Automated systems that leverage machine learning can significantly enhance the diagnostic process. Rapid progress in deep learning (DL) has facilitated the creation of sophisticated, fine-tuned models, including convolutional neural networks (CNNs) such as VGG16, ResNet, and DenseNet, as well as Vision Transformers (ViTs) such as ViT-B16, ViT-B/32, ViT-L/16, and ViT-L/32, which have shown remarkable performance in various computer vision tasks. These models have shown promise in the identification and classification of brain tumors from MRI images. Building on the success of ViT models, which offer higher accuracy with lower resource demands, our proposed model aims to further improve accuracy and tumor type classification. Using the strengths of transformer-based architectures, our objective is to create a highly accurate and effective diagnostic tool for the classification of brain tumors.