Brain MRI Images (Tumor and No Tumor)
Description
This dataset comprises primary clinical MRI scans acquired directly from the Department of Radiology at Dhaka Medical College Hospital (DMCH). It encompasses four diagnostic categories: Glioma, Meningioma, Pituitary, and No Tumor, representing real-world imaging scenarios with authentic scanner noise and contrast variations. All samples have undergone strict anonymization (PHI removal) and rigorous pre-processing, including cropping, color grading, image enhancement and resize. This collection is uniquely suited for evaluating Deep Learning model performance on raw, single-source clinical data distinct from standard public benchmarks.
Files
Steps to reproduce
This hybrid dataset comprises a comprehensive collection of T1-weighted Magnetic Resonance Imaging (MRI) scans designed for the multi-class classification of brain tumors, specifically targeting Glioma, Meningioma, Pituitary tumors, and No Tumor (healthy controls). The primary subset consists of raw clinical samples physically acquired from the Department of Radiology at Dhaka Medical College Hospital (DMCH), Bangladesh, providing real-world diagnostic complexity with authentic scanner noise and contrast variations. These privacy-sensitive images were rigorously de-identified in compliance with ethical protocols and subsequently integrated with high-quality samples from open-access benchmarks to address class imbalance and domain shift challenges. All images have undergone a standardized pre-processing pipeline, including automated brain contour cropping to remove non-informative skull artifacts and Contrast Limited Adaptive Histogram Equalization (CLAHE) to enhance tumor boundary visibility, making this dataset uniquely optimized for training robust Deep Learning models capable of generalizing across diverse imaging environments.
Institutions
- Daffodil International University