Angelou0516/brats2023-gli-dataset

BraTS2023 GLI Dataset

Dataset Description

The BraTS2023 Glioma (GLI) dataset for brain tumor segmentation. This dataset contains multi-modal MRI scans with dense segmentation annotations.

Multi-Modal MRI

Each patient case includes 4 MRI modalities:

• T1n: Native T1-weighted MRI
• T1c: Post-contrast T1-weighted MRI
• T2w: T2-weighted MRI
• T2f: T2-FLAIR MRI

All 4 modalities share the same segmentation mask.

Dataset Structure

Each sample in the JSONL file contains:

{
  "modalities": {
    "t1n": "path/to/t1n.nii.gz",
    "t1c": "path/to/t1c.nii.gz",
    "t2w": "path/to/t2w.nii.gz",
    "t2f": "path/to/t2f.nii.gz"
  },
  "mask": "path/to/seg.nii.gz",
  "label": ["necrotic brain tumor core", "brain edema", "enhancing brain tumor", "brain tumor"],
  "dataset": "BraTS2023_GLI",
  "official_split": "train",
  "patient_id": "BraTS-GLI-XXXXX-XXX"
}

Segmentation Labels

The dataset includes segmentation masks for brain tumor regions:

• Necrotic Tumor Core (NCR): Label 1
• Peritumoral Edema (ED): Label 2
• Enhancing Tumor (ET): Label 3

Common evaluation metrics:

• Whole Tumor (WT): NCR + ED + ET
• Tumor Core (TC): NCR + ET
• Enhancing Tumor (ET): ET only

Data Format

• Modality: MRI (T1n, T1c, T2w, T2-FLAIR)
• Images & Masks: NIfTI format (.nii.gz)
• Resolution: Resampled to 1mm³ isotropic
• Size: 240 × 240 × 155

Usage

from datasets import load_dataset
import nibabel as nib

# Load metadata
ds = load_dataset("Angelou0516/brats2023-gli-dataset")

# Get a patient case
patient = ds['train'][0]
print(patient['patient_id'])
print(patient['modalities'])  # Dict with 4 modalities

# Download the full dataset to load actual images
from huggingface_hub import snapshot_download
local_path = snapshot_download(repo_id="Angelou0516/brats2023-gli-dataset", repo_type="dataset")

# Load NIfTI files
t1n = nib.load(patient['modalities']['t1n'])
t1c = nib.load(patient['modalities']['t1c'])
t2w = nib.load(patient['modalities']['t2w'])
t2f = nib.load(patient['modalities']['t2f'])
seg = nib.load(patient['mask'])

Citation

@article{brats2023,
  title={The BraTS 2023 Challenge on Brain Tumor Segmentation},
  author={BraTS Organizers},
  journal={ArXiv},
  year={2023}
}

License

CC-BY-4.0

Dataset Homepage

https://www.synapse.org/#!Synapse:syn51156910