Radboudumc TBI SWI Microbleed Set
Radboudumc moderate-to-severe traumatic brain injury SWI cerebral microbleed dataset
Single-centre 3 T SWI and T1 MRI from Radboudumc, Nijmegen: 81 studies of 45 patients with moderate to severe traumatic brain injury and 18 healthy controls, with microbleed segmentations and six-observer point annotations. Not publicly released.
Overview
This is an in-house MRI collection from Radboud University Medical Center in Nijmegen, the Netherlands, used to develop deep learning detectors of cerebral microbleeds after traumatic brain injury. In trauma, microbleeds are taken as a marker of traumatic axonal injury, and severe cases can show a hundred or more of them, often in clusters or with a curvilinear shape. The 2022 paper compares a patch classifier, a segmentation CNN and a 3D U-Net on these data.
Composition
The text describes 45 patients with moderate or severe TBI (Glasgow Coma Scale 3 to 12), 20 of them scanned at two timepoints, plus 18 healthy volunteers, for 81 MRI studies in total. Table 1 of the paper lists 46 TBI subjects across the splits, so the patient count differs by one between text and table. The studies are split into 45 for training (41 TBI, 4 control), 16 for validation (12 TBI, 4 control) and 20 for testing (10 TBI, 10 control). The data were collected retrospectively with a waiver of informed consent.
Acquisition
All scans come from one 3 T Siemens Magnetom Trio. Each study has an SWI scan (TR 27 ms, TE 20 ms, flip angle 15 degrees, 0.98 x 0.98 x 1.00 mm voxels) and a T1 MP-RAGE (1 mm isotropic). Only SWI is used by the deep learning models.
Annotations
Training and validation studies have voxel segmentations of the full blooming extent of each microbleed. An experienced neuroradiologist sorted the output of an earlier computer-aided detection system into definite, possible and false positive microbleeds and added missed ones, and a medical student then outlined each lesion under her supervision. For 10 test studies, six observers of differing experience placed points on definite and possible microbleeds; region growing from these points and a majority vote give the test reference. Agreement between observers was low (Fleiss' kappa 0.19 to 0.24).
Known limitations
- Not publicly available. The paper announced a Zenodo release and a challenge, but neither could be found.
- One scanner, one SWI protocol and one centre.
- Only moderate and severe TBI; mild TBI is not included.
- Small test set with low inter-observer agreement on what counts as a microbleed.
Cohort
Aggregate numbers from the sources below. Bars are relative to the 63 subjects.
Contrast / sequence
studies, values can overlap
- Susceptibility-weighted 81 100%
- T1-weighted 81 100%
Condition
subjects, values can overlap
- Traumatic brain injury 45 71%
- Healthy control 18 29%
Scanner vendor
studies
- Siemens Healthineers 81 100%
Field strength
studies
- 3 T 81 100%
Country
studies
- Netherlands 81 100%
Split
studies
- Training 45 56%
- Test 20 25%
- Validation 16 20%
License and access
Our reading of the license, not legal advice. Before you use the data, read the original license and confirm that your use is allowed. We take no responsibility for how you use a dataset. Full disclaimer
Ask the authors or the data holder. No published process or terms, and access is at their discretion
No published license or data use terms
The data holder has published no license and no data use terms. Any use, sharing or commercial right has to be agreed with the data holder, and nothing can be assumed to be allowed. Default copyright and data protection law still apply.
The paper has no data availability statement. A footnote says the MRI data without the test set "will be made available on Zenodo (10.5281/zenodo.6535523)" for a challenge at traumatic-cmb.grand-challenge.org. On 2026-10-08 that DOI was not registered and the challenge site was not publicly reachable. The CC BY-NC-ND 4.0 license of the article covers the text only.
What you can do
- Commercial use Not stated
- Train ML models Not stated
- Create derived data Not stated
- Publish results Not stated
What you can share
- Share the data Not stated
- Share derived data Not stated
- Share trained models Not stated
What you must do
- Cite or credit Not stated
- Share alike Not stated
- Sign an agreement Not stated
- Ethics approval Not stated
- Manuscript review Not stated
- Release code Not stated
- Return results Not stated
- Delete after use Not stated
Limits
- No re-identification Not stated
- Location limits Not stated
Citation
Koschmieder K, Paul MM, van den Heuvel TLA, van der Eerden AW, van Ginneken B, Manniesing R. Automated detection of cerebral microbleeds via segmentation in susceptibility-weighted images of patients with traumatic brain injury. NeuroImage: Clinical 35, 103027 (2022). doi:10.1016/j.nicl.2022.103027
All numbers
Every number on this page, as stored in stats.csv, with its source.
| Measure | Breakdown | Value | Source |
|---|---|---|---|
| Subjects | total 45 TBI patients plus 18 healthy controls as stated in the text; Table 1 lists 24 + 12 + 10 = 46 TBI subjects across the splits | 63 | koschmieder2022 Section 2.1.1 |
| Subjects | condition=traumatic_brain_injury moderate (GCS 9-12) to severe (GCS 3-8) TBI; Table 1 lists 46 | 45 | koschmieder2022 Section 2.1.1 |
| Subjects | condition=healthy healthy volunteers | 18 | koschmieder2022 Section 2.1.1 |
| Studies | total 20 patients were scanned at two timepoints | 81 | koschmieder2022 Section 2.1.1 and Table 1 |
| Studies | condition=traumatic_brain_injury 41 training + 12 validation + 10 test studies | 63 | koschmieder2022 Table 1 |
| Studies | condition=healthy 4 training + 4 validation + 10 test | 18 | koschmieder2022 Table 1 |
| Studies | contrast=swi every study has an SWI scan | 81 | koschmieder2022 Section 2.1.1 and Table 2 |
| Studies | contrast=T1w every study has a T1 MP-RAGE scan, not used by the deep learning models | 81 | koschmieder2022 Section 2.1.1 and Table 2 |
| Studies | contrast_set=swi+T1w | 81 | koschmieder2022 Section 2.1.1 |
| Studies | field_strength=3 | 81 | koschmieder2022 Section 2.1.1 |
| Studies | vendor=siemens Siemens Magnetom Trio | 81 | koschmieder2022 Section 2.1.1 |
| Studies | country=NL | 81 | koschmieder2022 Section 2.1.1 |
| Studies | split=train 41 TBI + 4 control studies from 24 TBI patients | 45 | koschmieder2022 Abstract and Table 1 |
| Studies | split=validation 12 TBI + 4 control studies | 16 | koschmieder2022 Abstract and Table 1 |
| Studies | split=test 10 TBI + 10 control studies | 20 | koschmieder2022 Abstract and Table 1 |
| Studies | split=train;condition=traumatic_brain_injury | 41 | koschmieder2022 Table 1 |
| Studies | split=validation;condition=traumatic_brain_injury | 12 | koschmieder2022 Table 1 |
| Studies | split=test;condition=traumatic_brain_injury | 10 | koschmieder2022 Table 1 |
| Studies | split=test;condition=healthy | 10 | koschmieder2022 Table 1 |
Sources
The keys used in the table above.
- koschmieder2022 Koschmieder et al. 2022, NeuroImage Clinical (PMC9127224) paper