Datasetpublic

MedOtter/RHUH-GBM

RHUH-GBM — Rio Hortega University Hospital Glioblastoma dataset Longitudinal multi-parametric MRI (mpMRI) of glioblastoma patients from Rio Hortega University Hospital (Valladolid, Spain), with expert tumor sub-region segmentations at three timepoints per patient: preoperative, early postoperative (< 72 h), and follow-up at recurrence. This is the NIfTI release from TCIA — images are skull-stripped and co-registered to the SRI24 atlas, and the segmentations are aligned to them.… See the full description on the dataset page: https://huggingface.co/datasets/MedOtter/RHUH-GBM.

sourceHugging Facecc-by-4.0updated 2mo agoView on Hugging Face

0likes489downloads

Dataset Card

RHUH-GBM — Rio Hortega University Hospital Glioblastoma dataset

Longitudinal multi-parametric MRI (mpMRI) of glioblastoma patients from Rio Hortega University Hospital (Valladolid, Spain), with expert tumor sub-region segmentations at three timepoints per patient: preoperative, early postoperative (< 72 h), and follow-up at recurrence. This is the NIfTI release from TCIA — images are skull-stripped and co-registered to the SRI24 atlas, and the segmentations are aligned to them.

Distinguishing feature vs. preop-only glioma datasets (e.g. BraTS): RHUH-GBM provides post-resection and recurrence scans with expert-validated masks — exactly the timepoints where automated tools usually fail.

Dataset Details

Field	Value
Modality	Brain mpMRI — T1, T1-Gd (T1CE), T2, T2-FLAIR, ADC
Body part	Brain (glioblastoma, WHO grade 4)
Task	3D multi-class tumor sub-region segmentation
Patients	40
Timepoints / patient	3 (preop / early postop / recurrence)
Studies (timepoints)	120
Imaging volumes	600 (5 sequences x 120 studies)
Segmentations	120 (one per study)
Volume geometry	240 x 240 x 155 (one study, RHUH-0028/0, is 230 x 230 x 138)
Format	NIfTI (`.nii.gz`)
License	CC BY 4.0

Label Scheme

Value	Tumor sub-region
0	Background
1	Necrosis (necrotic tumor core)
2	Peritumoral region (edema / non-enhancing signal alteration)
3	Enhancing tumor

Evaluation regions (BraTS-style): WT (whole tumor) = 1+2+3, TC (tumor core) = 1+3, ET (enhancing tumor) = 3.

Label encoding note (verified against the released masks). The upstream preprocessing pipeline documents enhancing tumor as BraTS label 4, but the distributed NIfTI masks use 3 — value 4 never appears in any of the 120 masks. Loaders should treat 3 as enhancing tumor for RHUH-GBM.

Ground Truth

A single, expert-corrected segmentation tier (no separate automated tier). Masks were initialized with a DeepMedic CNN and then reviewed and manually corrected by two neurosurgeons specializing in neuroimaging. Every study has exactly one segmentation; no GT-tier filtering is required.

Structure

RHUH-NNNN/<tp>/RHUH-NNNN_<tp>_t1.nii.gz
RHUH-NNNN/<tp>/RHUH-NNNN_<tp>_t1ce.nii.gz
RHUH-NNNN/<tp>/RHUH-NNNN_<tp>_t2.nii.gz
RHUH-NNNN/<tp>/RHUH-NNNN_<tp>_flair.nii.gz
RHUH-NNNN/<tp>/RHUH-NNNN_<tp>_adc.nii.gz
RHUH-NNNN/<tp>/RHUH-NNNN_<tp>_segmentations.nii.gz   # expert GT
subjects_manifest.json                               # per-study paths + legends

<tp> is the timepoint index: 0 = preoperative, 1 = early postoperative (< 72 h), 2 = follow-up / recurrence. subjects_manifest.json lists, for every study, the five modality paths and the segmentation path, plus the label and timepoint legends — so loaders need not re-derive them.

Cohort Overlap

No known overlap with BraTS2023 or UCSF-PDGM. RHUH-GBM is a single-institution Spanish cohort (Rio Hortega U. Hospital, 2018-2022) and is not among the contributing sites of either dataset; the released data carries no BraTS/UCSF cross-reference identifiers (patient IDs are RHUH-00NN). No cases need to be excluded when benchmarking alongside those datasets.

Notes for Loaders

—Images and masks share an identical grid+affine within each study — no resampling or axis permutation is needed between a scan and its mask.
—Do not hardcode the volume shape: most studies are 240x240x155, but RHUH-0028/0 is 230x230x138. Read the shape per study (or from the manifest).
—One non-standard filename: RHUH-0035/2's mask is segmentation.nii.gz (not ..._segmentations.nii.gz). subjects_manifest.json records the real path; prefer the manifest over globbing.
—The NIfTI images are SRI-registered/skull-stripped and do not align with the TCIA DICOM package by design.
—Multi-channel input: stack T1/T1CE/T2/FLAIR (+ADC) as channels (BraTS-style).

Source

—TCIA collection: https://www.cancerimagingarchive.net/collection/rhuh-gbm/
—DOI: 10.7937/4545-c905
—Public, no registration required (TCIA fully public since 2025-07-07).

License & Attribution

Released under CC BY 4.0. This mirror contains only the skull-stripped NIfTI derivative; the raw DICOM MRI is not included. Attribution to the original creators is given in the citation below.

Changes made in this mirror: the original TCIA NIfTI files were reorganized into the RHUH-NNNN/<tp>/ per-study directory layout, a subjects_manifest.json index was added, and a small 2D preview (data/train-*.parquet) was generated for the dataset viewer. No voxel intensities, affines, or segmentation labels were altered — the volumes and masks are the unmodified TCIA release. Full attribution to the original creators is given in the citation below.

Citation

bibtex

@article{cepeda2023rhuhgbm,
  author  = {Cepeda, Santiago and Garc\'ia-Garc\'ia, Sergio and Arrese, Ignacio
             and Herrero, Francisco and Escudero, Trinidad and Zamora, Tom\'as
             and Pastor, Roberto and others},
  title   = {The R\'io Hortega University Hospital Glioblastoma dataset: A
             comprehensive collection of preoperative, early postoperative and
             recurrence MRI scans (RHUH-GBM)},
  journal = {Data in Brief},
  volume  = {50},
  pages   = {109617},
  year    = {2023},
  doi     = {10.1016/j.dib.2023.109617}
}