ATLAS v2.0
Anatomical Tracings of Lesions After Stroke, Release 2.0
T1-weighted brain MRI with manually traced stroke lesion masks from 44 research cohorts in 11 countries: 1,271 subjects, of which 655 training cases with masks and 300 test scans without masks are released.
Overview
ATLAS (Anatomical Tracings of Lesions After Stroke) is a collection of stroke brain MRIs with hand-drawn lesion masks, assembled at the University of Southern California with the ENIGMA Stroke Recovery working group. Release 2.0 replaced the 304-subject v1.2 and was built to give lesion segmentation methods a larger, more varied training set and a proper held-out test. It was part of the ISLES challenge at MICCAI 2022.
Composition
The release covers 1,271 people, each with a single T1-weighted scan. Of these, 655 training cases come with lesion masks and 300 test cases come as images only, both drawn at random from the same 33 cohorts. A further 316 cases from 11 other cohorts form a generalizability set that is never released and exists only for challenge evaluation. Training metadata give scanner information per cohort, lesion count and location, and days since stroke where known. Age, sex and clinical outcomes are not shared. Nine repeat scans from v1.2 were dropped so that every subject appears once.
Acquisition
Scans were collected for different research studies at 1.5T and 3T, mostly at 1 mm resolution or better, and range from the first day after stroke to the chronic stage. Each cohort used one scanner and protocol, with two exceptions. INDI hosts a version registered to MNI-152 space, bias-corrected and defaced; the native-space images are kept at ICPSR under stricter terms.
Annotations
Trained team members traced lesions in ITK-SNAP following a standard protocol, excluding white matter hyperintensities and perivascular spaces where possible. Two further team members reviewed every mask, and the original tracer made any corrections. For some subjects the contributing site supplied a mask that the team then edited.
Known limitations
- Each lesion has only one tracing, so inter-rater agreement is unknown.
- Cohorts come from research studies with their own inclusion criteria and may not represent all stroke patients.
- No demographics are provided.
- Release 2.0 has since been superseded by R2.1 (test masks released) and R3.0 (1,453 scans).
Cohort
Aggregate numbers from the sources below. Bars are relative to the 1,271 subjects.
Contrast combinations
How many subjects have exactly each set of contrasts.
| T1w | Subjects with exactly this set |
|---|---|
1,271 |
Contrast / sequence
subjects, values can overlap
- T1-weighted 1,271 100%
Condition
subjects, values can overlap
- Stroke 1,271 100%
Split
subjects
- Training 655 52%
- Test 300 24%
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
Sign a data use agreement, often reviewed by the provider
ATLAS Terms of Use (University of Southern California)
Research or statistical use only, after submitting a request form that the ATLAS team reviews before sending a per-person decryption key. No redistribution without written agreement of USC, no re-identification, and publications must cite ATLAS and be reported to the ATLAS team.
Native-space raw data are archived at ICPSR/ADDEP (doi:10.3886/ICPSR36684.v4) under what the data descriptor calls "a more stringent restricted data use agreement". That agreement could not be read for this entry.
What you can do
- Not stated
- Not stated
- Not stated
- Conditional
What you can share
- No
- Not stated
- Share trained models Not stated
What you must do
- Yes
- Conditional
- Yes
- No
- Manuscript review No
- Release code No
- Conditional
- Conditional
Limits
- Yes
- No
Commercial license: Not stated
Citation
Liew SL, Lo BP, Donnelly MR, et al. A large, curated, open-source stroke neuroimaging dataset to improve lesion segmentation algorithms. Scientific Data 9, 320 (2022). doi:10.1038/s41597-022-01401-7
All numbers
Every number on this page, as stored in stats.csv, with its source.
| Measure | Breakdown | Value | Source |
|---|---|---|---|
| Subjects | total one T1w MRI per subject from 44 cohorts; includes 316 hidden generalizability cases that are not released | 1,271 | liew2022 Abstract |
| Subjects | split=train T1w and lesion masks released | 655 | liew2022 Abstract |
| Subjects | split=test T1w released; masks hidden in R2.0 (released later in R2.1) | 300 | liew2022 Abstract |
| Subjects | contrast=T1w | 1,271 | liew2022 Methods, Data overview and Data characteristics |
| Subjects | contrast_set=T1w | 1,271 | liew2022 Methods, Data overview and Data characteristics |
| Subjects | condition=stroke | 1,271 | liew2022 Methods, Data overview and Data characteristics |
Sources
The keys used in the table above.
- liew2022 Liew et al. 2022, A large, curated, open-source stroke neuroimaging dataset (Scientific Data) paper
- indi-atlas-2021 INDI ATLAS page as archived in December 2021 (R2.0) website