A Hybrid ViT-GRU Approach for Brain Tumor Detection and Classification in MRI Images with Explainable Artificial Intelligence

Brain tumors present substantial obstacles in both diagnosis and treatment. It’s imperative to accurately and promptly detect brain tumors to ensure that diagnosis and treatment are effective. This work proposes an innovative hybrid ViT-GRU model utilizing a Vision Transformer (ViT) and Gated Recurrent Unit (GRU) for enhanced brain tumor detection and classification in MRI scans. The ViT extracts key features, while the GRU captures temporal dependencies, addressing class imbalance common in MRI datasets. We used primary MRI data collected from Bangabandhu Sheikh Mujib Medical College Hospital (BSMMCH) in Faridpur, Bangladesh. Our model surpasses existing methods, achieving impressive results with metrics like precision, recall, and F1-score 96%, 95.97%, and 95.86% respectively. We evaluated the model with various optimizers SGD, Adam, and AdamW attaining peak accuracies of 82.55%, 97.55%, and 97.98%, respectively. Additionally, Explainable Artificial Intelligence (XAI) techniques were applied to enhance model interpretability, building trust in AI-driven diagnostics by clarifying the reasoning behind tumor classification. By incorporating methods like LIME, SHAP, and Attention maps, our model achieved essential transparency, reinforcing its reliability for clinical use. This comprehensive analysis with diverse optimizers and XAI integration positions our work as a strong contender for accurate and interpretable brain tumor detection in MRI.