🧠 BRISC 2025

Published: 23 April 2026| Version 1 | DOI: 10.17632/c7dj4pwhcv.1
Contributor:
BRISC Dataset

Description

BRISC 2025 β€” Brain Tumor MRI Dataset BRISC (BRain tumor Image Segmentation & Classification) β€” a curated, expert-annotated T1 MRI dataset for multi-class brain tumor classification and pixel-wise segmentation. Published in: Scientific Data (Nature Portfolio) DOI: https://doi.org/10.1038/s41597-026-06753-y πŸš€ Overview BRISC is designed to address common shortcomings in existing public brain MRI collections (e.g., class imbalance, limited tumor types, and annotation inconsistency). It provides high-quality, physician-validated pixel-level masks and a balanced multi-class classification split, suitable for benchmarking segmentation and classification algorithms as well as multi-task learning research. Highlights - 6,000 T1-weighted MRI slices (5,000 train / 1,000 test) - Four classes: Glioma, Meningioma, Pituitary Tumor, No Tumor - Pixel-wise segmentation masks reviewed by radiologists - Slices from three anatomical planes: Axial, Coronal, Sagittal - Clean, stratified train/test splits and aligned image–mask filenames πŸ“¦ Dataset structure brisc2025/ β”œβ”€ classification_task/ β”‚ β”œβ”€ train/ β”‚ β”‚ β”œβ”€ glioma/ β”‚ β”‚ β”‚ β”œβ”€ brisc2025_train_00001_gl_ax_t1.jpg β”‚ β”‚ β”‚ └─ ... β”‚ β”‚ β”œβ”€ meningioma/ β”‚ β”‚ β”œβ”€ pituitary/ β”‚ β”‚ └─ no_tumor/ β”‚ └─ test/ β”‚ β”œβ”€ glioma/ β”‚ β”‚ β”œβ”€ brisc2025_test_00001_gl_ax_t1.jpg β”‚ β”‚ └─ ... β”‚ β”œβ”€ meningioma/ β”‚ β”œβ”€ pituitary/ β”‚ └─ no_tumor/ β”œβ”€ segmentation_task/ β”‚ β”œβ”€ train/ β”‚ β”‚ β”œβ”€ images/ β”‚ β”‚ β”‚ β”œβ”€ brisc2025_train_00001_gl_ax_t1.jpg β”‚ β”‚ β”‚ └─ ... β”‚ β”‚ └─ masks/ β”‚ β”‚ β”œβ”€ brisc2025_train_00001_gl_ax_t1.png β”‚ β”‚ └─ ... β”‚ └─ test/ β”‚ β”œβ”€ images/ β”‚ β”‚ β”œβ”€ brisc2025_test_00001_gl_ax_t1.jpg β”‚ β”‚ └─ ... β”‚ └─ masks/ β”‚ β”œβ”€ brisc2025_test_00001_gl_ax_t1.png β”‚ └─ ... β”œβ”€ manifest.json β”œβ”€ manifest.csv β”œβ”€ manifest.json.sha256 β”œβ”€ manifest.csv.sha256 └─ README.md Notes: - Classification folders contain image-level labels suitable for standard image classification pipelines. - Segmentation folders contain paired MRI images/ and corresponding binary masks/. - Image and mask filenames are identical except for file extension (images: .jpg, masks: .png). - All images are T1-weighted slices. πŸ“Š Dataset statistics - Total samples: 6,000 (5,000 train / 1,000 test) - Classes: 4 (balanced distribution across train/test) - Planes: Axial / Coronal / Sagittal (balanced representation) - Imaging modality: T1-weighted MRI - Annotation quality: Reviewed and corrected by medical experts πŸ“„ Citation If you use BRISC in your work, please cite: @article{fateh2025brisc, title={Brisc: Annotated dataset for brain tumor segmentation and classification with swin-hafnet}, author={Fateh, Amirreza and Rezvani, Yasin and Moayedi, Sara and Rezvani, Sadjad and Fateh, Fatemeh and Fateh, Mansoor and Abolghasemi, Vahid}, journal={arXiv preprint arXiv:2506.14318}, year={2025} } πŸ”— References & inspirations This dataset drew design and organizational inspiration from widely used brain tumor imaging datasets (e.g., BraTS, Figshare datasets, Kaggle collections).

Files

Steps to reproduce

Sources: We compiled this dataset by surveying multiple publicly available medical imaging repositories and selecting high‑resolution chest and abdominal scans suitable for both classification and segmentation tasks. After an initial review of existing collections, we identified and extracted only those images that met our quality and modality criteria. All images were then cross‑checked with collaborating radiology departments to ensure clinical relevance and de‑identified in accordance with institutional privacy standards. Collection Methodology: We compiled the dataset by reviewing multiple existing medical‑imaging collections and selecting only high‑quality scans suitable for both classification and segmentation. In collaboration with board‑certified radiologists, each image was annotated at both the image and pixel levels. All cleaned and labeled images were then aggregated into a unified repository and subjected to a final audit to eliminate duplicates and ensure consistent label integrityβ€”resulting in a precise, reliable dataset ready for research use.

Categories

Image Processing, Image Segmentation, Brain Tumor, Brain Imaging, Image Classification, Medical Image Processing, Deep Learning, Magnetic Resonance Imaging of Brain Tumor

Licence