MRI-Based Brain Tumor Detection and Classification System Using Vision Transformers and Transfer Learning

Brain tumors are critical health conditions which adversely affect thinking, emotions, and movement of patients. Detecting and diagnosing brain tumors at an early stage is essential. However, it can be hard to detect them because initial signs of brain tumor can be very mild. This paper presents an innovative method to identify and classify brain tumors successfully. Our system, formulated for MRI scans, uses techniques such as Convolutional Neural Networks, Vision Transformer, Transfer Learning, and hybrid model to detect and identify three specific types of tumors: pituitary, glioma, and meningioma. We trained our deep learning models using well-organized sets of tumor and non-tumor MRI images. The accuracy of Vision Transformer is 99.08%, while the Hybrid CNN-ViT got 98.05% The Transfer Learning with VGG16 showed an accuracy of 97.08% Also, when we put together the EfficientNet, ResNet, and VGG models, they achieved accuracy rates of 94.11%, 94.75%, and 94.96%, respectively These effective models help find tumors quickly and accurately, making it easier to start appropriate treatment.