BDNeuro-MRI: A Bangladeshi Clinical Brain Tumor MRI Dataset for Four-Class Deep Learning Classification

This dataset contains 5,941 pre-processed, leakage-free T1-weighted contrast-enhanced MRI brain images, categorized into four classes: Glioma, Meningioma, Pituitary, and No Tumor. Images were sourced from Epic & CSCR Hospital, Bangladesh, and rebuilt from an earlier release of this dataset through a rigorous cleaning pipeline: removal of exact and near-duplicate images, followed by a stratified 70% training / 15% validation / 15% testing split, with every class preserved in the same proportion across all three splits. A programmatic integrity check confirms zero image overlap between the training, validation, and test sets — eliminating a common source of inflated benchmark results in brain tumor MRI datasets. Images are organized into subfolders by class and split. The dataset is suitable for machine learning and deep learning research in brain tumor classification, tumor detection, and computer-aided diagnosis. **🧪 Applications:** --------------------- - Brain tumor classification - Deep learning (CNN, ViT, transfer learning) - Computer-aided diagnosis (CAD) - Radiology research and teaching - Benchmarking dataset-integrity / leakage-detection methodology The full preprocessing, deduplication, splitting, and leakage-verification pipeline — along with a 7-model baseline benchmark (CNN, ViT, Hybrid CNN–ViT, ResNet50, and DenseNet121, from-scratch/frozen/fine-tuned) — is publicly available and fully reproducible on GitHub: 🔗 **Code**: https://github.com/irfanulkabirhira/A-Bangladeshi-Clinical-Brain-Tumor-MRI-Dataset-for-Four-Class-Deep-Learning-Classification