The dataset viewer is not available because its heuristics could not detect any supported data files. You can try uploading some data files, or configuring the data files location manually.
Multi-Rater Brain Hemorrhage Segmentation Dataset (MR-BHSD)
Our dataset is hosting its latest competition at MICCAI! Stay tuned for our contests and win great prizes!
Description
The Multi-Rater Brain Hemorrhage Segmentation Dataset (MR-BHSD) is a 3D multi-class segmentation dataset for intracranial hemorrhage (ICH). Intracranial hemorrhage is a pathological condition characterized by bleeding within the skull or brain, which can arise from various factors. Accurately identifying, localizing, and quantifying ICH is crucial for clinical diagnosis and treatment. Our dataset comprises 192 volumes with voxel-level annotations from four annotators and 1980 volumes with case-level annotations across five ICH categories. The released validation set include 48 cases labelled by three annotators. All data are saved in NIFTI format that are reconstructed from DICOM images.
Data Contents
This dataset includes the training set for models and the validation set, including:
- MBH_Train_2025_voxel-label.zip: Contains 192 volumes with voxel-level annotations (from four different annotators). You can directly download it:
wget https://huggingface.co/datasets/mbhseg/mbhseg25/resolve/main/MBH_Train_2025_voxel-label.zip?download=true`
- MBH_Train_2025_case-label.zip: Contains 1980 volumes with case-level annotations (577 hemorrhage-related cases are shown in
case-wise_annotation.csv, the rest are hemorrhage-free). You can directly download it:
wget https://huggingface.co/datasets/mbhseg/mbhseg25/resolve/main/MBH_Train_2025_case-label.zip?download=true`
- NEW_RELEASE MBH_Val_2025_voxel-label.zip: Contains 48 volumes with voxel-level annotations from three annotators (
annot_1andannot_3are the same annotators from the training dataset, and an extra annotatorannot_5is included for testing generalizability). You can directly download it:
wget https://huggingface.co/datasets/mbhseg/mbhseg25/resolve/valset/MBH_Val_2025_voxel-label.zip?download=true
Applications
This dataset is primarily intended to support the use of deep learning techniques in medical image segmentation tasks, particularly for multi-class segmentation of intracranial hemorrhages. It can be used for supervised and semi-supervised ICH segmentation tasks, and we provide experimental results with state-of-the-art models as reference benchmarks.
IMPORTANT for MBH-Seg25 Participants
The provided validation et is intended primarily for evaluating trained models using the official evaluation script (can be found at Official Link).
You MAY include the validation data in your model fine-tuning for the final testing and submission.
Due to ethical requirements, the final test cases will NOT be released. Instead, participants are expected to have a maximum of 3 submissions of their final models (running scripts with trained weights) and final method description document (MICCAI template, up to 8 pages, plus up to 2 pages of reference).
Access and Usage
The dataset is now publicly available for download and is being used in the latest MICCAI competition. Researchers interested in using the dataset can download it and must comply with the relevant terms of use and copyright information.
Copyright and License
This dataset is protected under copyright law. Please ensure compliance with the relevant license terms before use.
Contact Information
For more information or to apply for access to the dataset, please contact us via:
- Email: mbhseg24@gmail.com
- Downloads last month
- 89