OpenMind: head-and-neck 3D MRI pooled from OpenNeuro, one study at a time

pl.load("hf_openmind") loads this dataset. 114k MRI volumes from 800 OpenNeuro studies, pooled by the nnU-Net group (MIC-DKFZ) for self-supervised pre-training, each with a deface mask and an anatomy mask. The repo holds 266 721 files: listing it whole takes 15 minutes and is rate-limited for anonymous clients, so this manifest scopes the listing to one study (source.paths). pl.load('hf://MIC-DKFZ/OpenMind/OpenMind/ds000003') picks another study; openneuro_metadata.csv at the repo root indexes them all.

import pardesline as pl

ds = pl.load("hf_openmind")
print(ds.describe())  # roles, frames, units, labels, licence, citation
s = ds[0]
obj = s.t1w  # fetched now, cached under ~/.cache/pardesline/datasets/hf_openmind

Provenance and licence

Source	Hugging Face Hub repo MIC-DKFZ/OpenMind, paths OpenMind/ds000001
Scheme	hf
Licence	CC-BY-4.0
Commercial use	allowed by the licence
Redistribution	allowed
Licence note	CC BY 4.0 on the Hub card (checked 2026-10-13); each OpenNeuro study keeps its own CC0 / CC BY terms upstream, TO CONFIRM per study before redistribution.
Terms	https://huggingface.co/datasets/MIC-DKFZ/OpenMind
Homepage	https://huggingface.co/datasets/MIC-DKFZ/OpenMind
Cite	Wald T., Ulrich C., Suprijadi J., et al. An OpenMind for 3D medical vision self-supervised learning. arXiv:2412.17041, 2024.
Samples grouped by	subject

Roles (what each sample contains)

Role	Type	Frame	Units	Labels / layout	Pattern or file
t1w	volume (MR)	from the file header	from the file header		OpenMind/{study}/{subject}/anat/*_T1w.nii.gz T1-weighted MRI (NIfTI; LPS mm come from the header once loaded)
t2w	volume (MR)	from the file header	from the file header		OpenMind/{study}/{subject}/anat/*T2*.nii.gz T2-weighted MRI when the study has one (inplane T2 in ds000001)
deface_mask	labels	from the file header	from the file header	1=defaced	OpenMind/{study}/{subject}/anat/*_T1w__Data/deface_mask.nii.gz Voxels removed by defacing on the T1w image
anatomy_mask	labels	from the file header	from the file header	1=anatomy	OpenMind/{study}/{subject}/anat/*_T1w__Data/fb_mask.nii.gz Voxels that hold anatomy on the T1w image (fb_mask)

Standardized layout

Files are fetched on first access into the dataset cache and stay in their original format; ds.export(folder) writes every sample as standard files (PLY for points and meshes, MetaImage for volumes and label maps) with a .pardesline.json sidecar carrying frame, units, labels, licence and provenance, plus a manifest for the folder. Label maps become exact surfaces with to_mesh(label=k), DICOM series become one volume, raw arrays are decoded through the layout above.

ds = pl.load("hf_openmind").take(2)
ds.export("./standardized")        # PLY / MetaImage + sidecars + manifest.json

Algorithms and models that run on it

• labels roles: mask.to_mesh(label=k), mask.to_numpy(), metrics: dice, hd95 (evaluation-metrics-3d) ; catalog modules: pyvista_volume_processing, simpleitk_registration, simpleitk_segmentation
• volume roles: volume.align(moving), volume.segment(lower, upper), volume.to_mesh(isovalue=...), volume.to_sitk() ; catalog modules: pyvista_volume_processing, simpleitk_registration, simpleitk_segmentation
• Supervised segmentation: prepare_nnunet_dataset.py writes this dataset in the nnU-Net v2 raw layout; PointNet trains on the label surfaces.
• Self-supervised pretraining: 3D JEPA reads the volumes directly.

Related

• Datasets catalog · How to download a medical imaging dataset in Python · Add a dataset
• Tags: medical, mri, head, brain, nifti, hf, self-supervised

_{Generated from src/pardesline/data/manifests/hf_openmind.json on 2026-10-14.}