Tumor ViT-GRU-XAI: Advanced Brain Tumor Diagnosis Framework: Vision Transformer and GRU Integration for Improved MRI Analysis: A Case Study of Egypt

Brain tumors, marked by unchecked cell proliferation within the central nervous system, present substantial challenges for accurate diagnosis and effective treatment. Early and precise detection is crucial for successful intervention. This study presents an innovative framework aimed at enhancing the detection and classification of brain tumor characteristics in Magnetic Resonance Imaging (MRI) scans. Due to the complexity of tumor interpretation in medical imaging—an issue even experienced radiologists face—our approach integrates a Vision Transformer (ViT) with a Gated Recurrent Unit (GRU) to significantly improve both reliability and accuracy. The Vision Transformer excels at extracting vital features from MRI images, while the Gated Recurrent Unit effectively manages the relationships among these features, addressing the prevalent class imbalance in MRI datasets. This combined model outperforms current diagnostic methods. Our framework was applied to primary MRI data sourced from the National Cancer Institute in Cairo, Egypt, and Brast2013/2015. After processing the dataset and training our model, we achieved notable performance metrics: a precision of 98.8%, recall of 98.4%, and F1-score of 98.3%. We evaluated the model’s performance using various optimizers, obtaining peak accuracies of 98.27% with Adam and 99.346% with AdamW. For the dataset Brast2013, and Brast2015, the ViT-GRU model achieved accuracy (99.429%, and 99.808%), precision (99.1%, and 98.8%), recall (98.9%, and 98.8%), and F1-score (98.7%, and 98.5%) respectively. Extensive 10-fold cross-validation and comparisons with existing Transfer Learning (TL) models further substantiated our results. Additionally, we applied Explainable Artificial Intelligence (XAI) techniques to clarify the model’s decision-making process, enhancing transparency and fostering clinician trust, which is essential for practical clinical use. Our results, benchmarked against state-of-the-art models, underscore the effectiveness of our proposed framework in advancing the diagnosis of brain tumors.