Open dataset · CT

BHSD — Brain CT Intracranial Hemorrhage

Reconstructed BHSD head CT volumes: all labeled/unlabeled source files included; axial images preserved while AP slice drift and z spacing were repaired.

No download, no account needed for the demo case — it opens read-only in the MedSeg editor. Sign up (free) to browse all 2,174 cases, run AI models, and segment.

CT 2,174 cases MIT Reference masks Pathology
BHSD — Brain CT Intracranial Hemorrhage — example case rendered in the MedSeg editor
Example case with its reference segmentation — straight from the catalog, rendered by MedSeg.

Browse cases

Demo cases open instantly in the browser-based editor — scroll the slices, inspect the reference masks, window the image. Everything else is one free account away.

About this dataset

The Brain Hemorrhage Segmentation Dataset (BHSD) ships 192 pixel-labeled intracranial hemorrhage CT volumes and 1980 unlabeled volumes. The public release is NIfTI-only, and many files carry NIfTI geometry that makes sagittal/coronal views look distorted even though the axial slices themselves look correct.

For this MedSeg import we reconstructed the dataset using the best interpretation of the source geometry: axial pixel planes are preserved, while two inter-slice errors are repaired. First, slice-to-slice anterior/posterior drift is removed by translating each axial slice in-plane. Second, implausible through-plane spacing is replaced with a plausible z spacing inferred from the source affine, slice count, and visual QA. The AP translation is scaled together with the z-spacing repair, which handles files whose NIfTI slice vector appears to have been half-scaled or tenfold-scaled during creation. No full gantry-angle or cover-grid resampling is applied: the original axial slice appearance and voxel matrix are kept. All labeled and unlabeled source files are included; 4D source files are split into separate 3D volumes.

FactValue
Cases2,174
Series2,174
Size17.6 GB
ModalityCT
Reference masksYes
LicenseMIT
PublisherRSNA / BHSD authors
Version1.0-reconstructed-ap-z
DOI10.1007/978-3-031-45673-2_15

What you can do with it in MedSeg

Copied cases behave like normal project series — the public image bytes are linked, not duplicated, so copies are instant and take no extra storage.

  1. Copy cases into a project.
    Filter, multi-select, copy — reference masks come along if you want them.
  2. Run AI segmentation.
    TotalSegmentator, MRSegmentator, nnInteractive 3D clicks/scribbles, or text-prompted VoxTell.
  3. Edit and measure.
    Brush, lasso, fill, oblique planes, volumes in ml — in the browser.
  4. Train your own nnU-Net.
    Correct masks on a handful of cases and train a custom model on hosted GPUs.

Citation

If you use this dataset in your research, cite the source per its license terms.

Wu, B., Xiao, Q., Liu, S., Yin, L., Pechenizkiy, M., Mocanu, D. C., Van Keulen, M., & Mocanu, E. (2023). BHSD: A 3D Multi-Class Brain Hemorrhage Segmentation Dataset. In MLMI@MICCAI 2023.

License: MIT

Open data still carries obligations — attribution at minimum. Check the license terms and the source publication before publishing work built on this dataset. MedSeg is a research tool, not a medical device.

More public datasets