Brain Tumor Detection using VIT

This study introduces a Vision Transformer-based model, ViT-BT, aimed at improving the classification of brain tumors using MRI scans. Leveraging transfer learning, the proposed approach utilizes Vision Transformers to address the limitations of conventional Convolutional Neural Networks (CNNs), which often fall short in capturing the broader contextual information in images. The self-attention mechanism in Vision Transformers enables effective modelling of long-range dependencies, making them well-suited for medical imaging tasks. In this work, the ViT-BT model is initially pre-trained and subsequently fine-tuned on dedicated brain tumor MRI datasets to enhance its classification performance across different tumor types. Evaluated using the BraTS 2023 dataset—which includes T1, T2, T1CE, and Flair modalities—the model achieved outstanding performance with precision of 97%, recall of 99%, F1-score of 99.41%, and accuracy of 98.17%. These results suggest that ViT-BT not only surpasses traditional CNN-based approaches but also holds significant promise for enhancing diagnostic precision in clinical environments. The findings highlight the effectiveness of Vision Transformer-based transfer learning in medical imaging and its potential for broader applications in healthcare.