Brain tumor disease diagnostic using vision transformer

Abstract

Brain tumors pose significant diagnostic challenges due to their complexity and life-threatening nature. MRI plays a central role in non-invasive diagno- sis by providing detailed brain imagery. However, manual MRI interpretation is time-consuming and prone to human error, highlighting the need for auto- mated solutions. This study explores the use of Vision Transformers (ViTs), an emerging deep learning model that leverages self-attention to capture global image features more effectively than traditional Convolutional Neural Networks (CNNs). Despite their potential, ViTs face high computational demands. To address this, the study proposes an optimized ViT for multi-class brain tumor classification using a public dataset of 7,023 MRI images across four tumor types. Through hyperparameter tuning and comparison of optimizers (Adam, Adamax, SGD), the model achieves 95.8% accuracy, over 95% precision and sensitivity, and a 99.15% AUC. These results underscore the model’s reliability and its promise for integration into clinical diagnostic workflows