Sohaaam/Brain_Tumour_Curated

Brain Tumour Curated Dataset

A curated collection of publicly available brain MRI images assembled from multiple source datasets for research, education, and machine learning experimentation.

Dataset Summary

This release contains 40,416 unique brain MRI images across four classes:

Class	Images
Glioma	21,581
Meningioma	12,039
No Tumor	4,089
Pituitary	2,707
Total	40,416

The dataset was constructed by collecting images from multiple publicly available datasets, standardizing their class labels, and performing exact duplicate detection using SHA-256 hashes.

Classes

• glioma
• meningioma
• pituitary
• no_tumor

The glioma category combines several glioma-related tumor labels present in the source datasets, including labels such as glioma, glioblastoma, astrocytoma, oligodendroglioma, and ependymoma.

Quality Control

The following processing was performed:

• Exact duplicate detection using SHA-256 hashing.
• Duplicate image copies were removed from the final collection.
• Cross-class label conflicts were checked.
• Segmentation masks were excluded from the classification dataset.
• All final images were checked for readability.
• 0 corrupt/unreadable images were found.
• 4,793 segmentation masks were excluded.
• No train/validation/test split is provided in this release.
• No class balancing or resampling has been performed.

Image Characteristics

The dataset contains mixed image formats and dimensions inherited from the source datasets.

Image formats include:

• JPEG
• PNG
• TIFF

Image modes include:

• RGB
• Grayscale
• RGBA
• Palette

The most common image dimensions are 512×512 and 224×224, but other dimensions are also present.

Users should perform appropriate preprocessing, such as resizing and conversion to a consistent number of channels, before model training.

Metadata

The metadata/ directory contains:

• final_metadata.csv — metadata for all 40,416 final images.
• source_occurrences.csv — source occurrence information.
• final_class_counts.csv — final class counts.
• duplicate_summary.csv — duplicate detection summary.
• excluded_segmentation_masks.csv — segmentation-mask exclusion information.

Source Datasets

This collection was assembled from multiple publicly available datasets, including datasets published on Kaggle and the BRISC 2025 dataset.

The original sources include:

1. Brain Tumor Classification MRI — sartajbhuvaji
2. Brain Tumor Detection / BR35H — ahmedhamada0
3. Brain Tumor Dataset — orvile
4. BRISC 2025 — briscdataset
5. Brain Tumor MRI Dataset — masoudnickparvar
6. Brain Tumor 12K MRI Images with Masks, Metadata and Bounding Boxes — fernando2rad
7. Brain Tumor Classification 2D — shubhamgajjar
8. Brain Tumor MRI Dataset for Deep Learning — alamshihab075
9. MRI Brain Tumor Dataset 4-Class 7023 Images — mohamadabouali1

Users should consult the original source pages and their associated terms before using or redistributing individual images.

Licensing and Attribution

This collection does not have a single uniform license.

It is a compilation of images originating from multiple datasets, and the applicable license or terms may differ between sources.

The curator does not claim ownership of the original MRI images.

Before using, redistributing, or commercially exploiting this collection, users should review the licensing and usage terms of each original source dataset and comply with any applicable attribution or other requirements.

The metadata files are provided to help identify the provenance of the images.

Intended Use

This dataset is intended for:

• Machine learning research
• Deep learning experimentation
• Image classification
• Computer vision experiments
• Educational projects
• Benchmarking and exploratory analysis

This dataset is not clinically validated and should not be used for medical diagnosis, treatment decisions, or other clinical decision-making.

Limitations

Because this dataset combines multiple independently collected datasets:

• Image acquisition protocols may differ.
• Image dimensions and formats vary.
• Class distributions are uneven.
• Dataset-source bias may be present.
• Label definitions may differ between original datasets.
• The dataset may contain demographic and institutional biases inherited from the source datasets.

No attempt was made in this release to eliminate these source-level differences.

Recommended Preprocessing

For model training, users may consider:

1. Converting images to RGB.
2. Resizing images to a consistent resolution.
3. Normalizing pixel values.
4. Applying training-only augmentation where appropriate.
5. Creating their own train/validation/test split while avoiding duplicate leakage.

Citation

If you use this curated collection, please cite this Hugging Face repository and, where possible, also cite the original source datasets used in the collection.

Disclaimer

This dataset is provided for research and educational purposes. The curator makes no claim that the images are suitable for clinical use.

Users are responsible for verifying the provenance, licensing, and suitability of the data for their intended application.