VALDO 2021
Where is VALDO? VAscular Lesions DetectiOn and segmentation challenge at MICCAI 2021
Brain MRI with manual labels of enlarged perivascular spaces, microbleeds and lacunes, the training data of a MICCAI 2021 challenge on detecting small markers of cerebral small vessel disease. Cases from three European cohorts.
Overview
VALDO 2021 ("Where is VALDO?") was a MICCAI 2021 challenge run by University College London and Erasmus MC on the automated detection and segmentation of three small markers of cerebral small vessel disease on brain MRI: enlarged perivascular spaces (Task 1), cerebral microbleeds (Task 2) and lacunes (Task 3). The training data is public on Zenodo; the validation and test cases stayed with the organisers.
Composition
Cases come from three population cohorts: SABRE (London, UK), the Rotterdam Scan Study (RSS, Netherlands) and ALFA (Barcelona, Spain, cognitively normal relatives of people with Alzheimer's disease). Tasks 1 and 3 use SABRE and RSS with 40 training and 66 test cases each. Task 2 adds ALFA and has 72 training and 147 test cases. Each task also had 5 RSS validation cases. According to the challenge design document, the RSS training subjects are the same in all tasks and the training subjects of Tasks 1 and 3 are identical, so the 152 training cases are not 152 different people. No age, sex or clinical information is shipped with the images.
Acquisition
SABRE was scanned on a Philips 3T system, RSS on a dedicated GE 1.5T scanner and ALFA on a GE Discovery 3T. Tasks 1 and 3 provide T1w, T2w and FLAIR registered to T1w space. Task 2 provides T2*-weighted gradient echo with T1w and T2w registered to T2* space; slice thickness of the T2* ranges from 0.8 mm (RSS) to 3 mm (SABRE, ALFA). All images were defaced and RSS scans were bias-field corrected. Files are NIfTI.
Annotations
Task 1 labels are mixed: SABRE slabs segmented by two raters, six RSS cases segmented in selected slices and regions, and 28 RSS cases with counts only, plus region masks and per-region count tables. Task 2 has full-brain microbleed segmentations for every case, following BOMBS or the Rotterdam protocol. Task 3 has full-brain lacune segmentations from two raters for every case.
Known limitations
- Training sets are small and the test labels are not public.
- Task 1 training labels are weak (counts or partial segmentations), and the two cohorts used different PVS criteria.
- The Zenodo license field says CC BY-NC 4.0 while the record text and paper say CC BY-NC-SA 4.0.
- Overlap between the Task 1 and Task 3 test sets is not reported, so no total subject count is given.
Cohort
Aggregate numbers from the sources below. Bars are relative to the largest value.
Contrast / sequence
Groups can overlap
- T2*-weighted 224
Table 2 and Section 2.4.2 in Sudre et al. 2022, challenge overview paper (arXiv 2208.07167v1; published in Medical Image Analysis 2024)
Country by contrast / sequence
Reported cross table. A dot marks a cell the source does not give.
| T2*-weighted | |
|---|---|
| Netherlands | 107 |
| Spain | 86 |
| United Kingdom | 31 |
Table 2, Table 2 and Section 2.4.2 in Sudre et al. 2022, challenge overview paper (arXiv 2208.07167v1; published in Medical Image Analysis 2024)
Field strength by contrast / sequence
Reported cross table. A dot marks a cell the source does not give.
| T2*-weighted | |
|---|---|
| 3 T | 117 |
| 1.5 T | 107 |
Table 2 and Section 2.4.1 in Sudre et al. 2022, challenge overview paper (arXiv 2208.07167v1; published in Medical Image Analysis 2024)
Split by contrast / sequence
Reported cross table. A dot marks a cell the source does not give.
| T2*-weighted | FLAIR | |
|---|---|---|
| Test | 147 | · |
| Training | 72 | 40 |
| Validation | 5 | · |
Table 2, Section 2.4.2 in Sudre et al. 2022, challenge overview paper (arXiv 2208.07167v1; published in Medical Image Analysis 2024)
Scanner vendor by contrast / sequence
Reported cross table. A dot marks a cell the source does not give.
| T2*-weighted | |
|---|---|
| GE HealthCare | 193 |
| Philips | 31 |
Table 2 and Section 2.4.1 in Sudre et al. 2022, challenge overview paper (arXiv 2208.07167v1; published in Medical Image Analysis 2024)
Other reported numbers 8
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
Download without an account
Only the training set is on Zenodo, as one archive per task. The validation and test cases were never released.
Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International
Non-commercial use only, with credit. Anything you share that is built on the data must use the same license.
The Zenodo record text, the challenge paper and the challenge design document name CC BY-NC-SA 4.0. The license field in the Zenodo metadata shows CC BY-NC 4.0. Users are asked to acknowledge the listed funders, cite the cohort papers, and for Task 2 data add "for the ALFA Study" as corporate author with the listed contributors.
What you can do
- No
- Not stated
- Conditional
- Yes
- Yes
What you can share
- Conditional
- Conditional
- Conditional
What you must do
- Yes
- Yes
- No
- No
- Manuscript review No
- Release code No
- Return results No
- Delete after use No
Limits
- No
- Location limits No
Citation
Sudre CH, Van Wijnen K, Dubost F, et al. Where is VALDO? VAscular Lesions Detection and segmentatiOn challenge at MICCAI 2021. Medical Image Analysis 91, 103029 (2024). doi:10.1016/j.media.2023.103029
Sources
Every number on this page comes from one of these documents. Each chart names the table or page it is taken from. The raw numbers are in stats.csv.