A Vision Transformer Deep Learning Model with CLAHE Technique of Image Enhancement for Brain Tumor Classification

This paper explores the deep learning-based classification of Brain Tumors using MRI (Magnetic Resonance Imaging) images. Brain tumors are characterized by abnormal and fast-growing tissues in the brain leading to severe health risks and require accurate and timely diagnosis for effective treatment. MRI is non-invasive radiation free technique commonly used to capture high-resolution images of body parts. Manual interpretation of these images is slow process and dependent on radiologists’ expertise. Since Deep learning in medical imaging have shown the promising results to automate the interpretation process, we have used in our work a fine-tuned Vision Transformer Model (ViT_B_16) for classification task. We have also modified its MLP head to customize classification into four categories. This model was trained on a publicly available brain MRI dataset from Kaggle [22]. In testing phase, ViT_B_16 achieved accuracy of 97.94% with overall precision, recall and F1-Score 97.91%, 97.76% and 97.79% respectively. The comparison of ViT_B_16 with CNN based models shows the performance superiority of the ViT_B_16 model. This better performance is attributed to vision transformer self-attention mechanism, which captures global image features, unlike CNNs that focus on local spatial features. This highlights the potential of transformer-based architectures for more accurate brain tumor classification.