AhmadIshaqai/brain-mri-plane-aware-vlm

Brain MRI Plane-Aware VLM Dataset

Dataset Description

This dataset is derived from the BRISC 2025 dataset and has been processed specifically for training Vision-Language Models (VLMs) with plane-aware understanding of brain MRI scans.

Source Dataset

Original dataset: BRISC 2025

The BRISC 2025 dataset contains:

• 6,000 T1-weighted MRI images
• Four tumor classes: Glioma, Meningioma, Pituitary Tumor, and No Tumor
• Pixel-wise segmentation masks validated by physicians and radiologists
• Three anatomical planes: Axial, Coronal, and Sagittal
• Clean, stratified splits: 5,000 training images and 1,000 testing images

Our Contribution

We have enhanced the original BRISC 2025 dataset by:

1. Combining tumor class labels (Glioma, Meningioma, Pituitary Tumor, No Tumor) with anatomical plane information (Axial, Coronal, Sagittal)
2. Generating descriptive text captions for each image to enable vision-language model training
3. Structuring the data for VLM tasks including image captioning, visual question answering, and plane-aware tumor classification

Dataset Structure

Data Fields

• image: Brain MRI scan image (T1-weighted)
• text: Generated text description/caption for VLM training
• tumor_type: Tumor classification (Glioma, Meningioma, Pituitary Tumor, or No Tumor)
• plane: MRI anatomical plane (Axial, Coronal, or Sagittal)
• sequence: MRI sequence information
• directory: Original file path reference

Data Splits

• Total examples: 6,000
• Size: 138 MB

Usage

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("AhmadIshaqai/brain-mri-plane-aware-vlm")

Use Cases

This dataset is suitable for:

• Training vision-language models on medical imaging
• Plane-aware brain tumor classification
• Medical image captioning
• Visual question answering for radiology
• Multi-modal medical AI research

Citation

If you use this dataset, please cite the original BRISC 2025 dataset:

License

This dataset inherits the license from the original BRISC 2025 dataset. Please refer to the original dataset for licensing terms.

Acknowledgments

We thank the creators of the BRISC 2025 dataset for providing high-quality, validated brain MRI data with segmentation masks.