JHU Acute Stroke MRI
Annotated Clinical MRIs and Linked Metadata of Patients with Acute Stroke, Baltimore, Maryland, 2009-2019
Clinical MRI of 2,888 patients admitted with acute or early subacute stroke to the Johns Hopkins Comprehensive Stroke Center (2009-2019), with manual DWI lesion masks, native and MNI-space images and clinical metadata, in BIDS.
Overview
This collection holds routine clinical brain MRI of patients admitted with acute stroke to the Johns Hopkins Comprehensive Stroke Center in Baltimore between 2009 and 2019. Its creators released it so that segmentation, mapping and scoring tools can be trained and tested on real clinical scans rather than uniform research protocols. It is archived at ICPSR as a restricted-use collection.
Composition
There are 2,888 patients, one MRI each, aged 18 to 99 (mean 62.2), 1,527 men and 1,361 women. The largest racial group is African American (43.5%). By radiological appearance, 1,878 lesions are ischemic, 540 show any bleeding (including hemorrhagic transformation and extra-parenchymal blood) and 470 are not visible on imaging, mostly transient ischemic attacks or very small strokes. Every patient has DWI with ADC. Most also have FLAIR (2,746), T2 (2,581), SWI (2,106) and T1-weighted images, either conventional (2,373) or MPRAGE (1,298); 531 have perfusion images of usable quality. Per-patient metadata include NIHSS, stroke type, 90-day modified Rankin score for part of the cohort, hospital stay, blood tests, blood pressure, BMI, prior conditions and medication.
Acquisition
Scans come from 11 scanners over ten years, 1,766 at 1.5T and 1,122 at 3T, mostly Siemens with some GE and Philips systems. Most were done 6 or more hours after symptom onset, and 43% after acute treatment. DWI has about 1 mm in-plane resolution with 2 to 7 mm slices. DICOM files were converted to NIfTI with dcm2niix in a BIDS layout, and images with full head coverage were defaced. DWI, B0, ADC and the masks are also provided in MNI space, together with intensity-normalized DWI, lesion frequency maps and average DWI templates.
Annotations
Two experienced tracers outlined lesions on DWI and ADC in ROIEditor, using other contrasts to exclude chronic or white matter lesions. A neuroradiologist reviewed every mask, and cases were redrawn until tracers and reviewer agreed. Intraventricular and subarachnoid blood was not traced. On 220 cases traced twice, inter-rater Dice was 0.68 and intra-rater Dice 0.72. Brain masks come from a U-Net trained on corrected level-set masks.
Known limitations
- Single center, with more Black and fewer Hispanic and Asian patients than many urban stroke centers.
- Protocols vary widely across scanners and years, and slice thickness is clinical.
- Lesion type is a radiological label and does not always match the clinical diagnosis.
- Paper and table differ on the scanner vendor split, so no per-vendor counts are given here.
Cohort
Aggregate numbers from the sources below. Bars are relative to the 2,888 subjects.
Sex
- Female 1,361 47%
- Male 1,527 53%
Age
mean 62.16± 14.68 · range 18 to 99No age bins reported.
Contrast / sequence
subjects, values can overlap
- Diffusion-weighted 2,888 100%
- FLAIR 2,746 95%
- T2-weighted 2,581 89%
- Susceptibility-weighted 2,106 73%
Condition
subjects, values can overlap
- Stroke 2,888 100%
- Ischemic stroke 1,878 65%
- Intracranial hemorrhage 540 19%
Field strength
subjects
- 1.5 T 1,766 61%
- 3 T 1,122 39%
Country
subjects
- United States 2,888 100%
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
A research proposal is reviewed and approved
Restricted-use collection at ICPSR. Requests need a Restricted Data Use Agreement, the reason for the request, and IRB approval or a notice of exemption.
ICPSR Terms of Use and Restricted Data Use Agreement for ICPSR 38464
Restricted-use data. Access needs a Restricted Data Use Agreement, a stated reason for the request and IRB approval or an exemption notice. The ICPSR terms limit use to research or statistics, forbid re-identification and redistribution without ICPSR's written agreement, and require citing the data and reporting publications to ICPSR.
What you can do
- Not stated
- Not stated
- Create derived data Not stated
- Conditional
What you can share
- No
- Not stated
- Share trained models Not stated
What you must do
- Yes
- Conditional
- Yes
- Yes
- Not stated
- Not stated
- Conditional
- Conditional
Limits
- Yes
- Conditional
Commercial license: Not stated
Citation
Liu CF, Leigh R, Johnson B, et al. A large public dataset of annotated clinical MRIs and metadata of patients with acute stroke. Scientific Data 10, 548 (2023). doi:10.1038/s41597-023-02457-9. Faria AV. Annotated Clinical MRIs and Linked Metadata of Patients with Acute Stroke, Baltimore, Maryland, 2009-2019. ICPSR (2025-10-08). doi:10.3886/ICPSR38464.v6
All numbers
Every number on this page, as stored in stats.csv, with its source.
| Measure | Breakdown | Value | Source |
|---|---|---|---|
| Subjects | total one clinical MRI per patient | 2,888 | liu2023 Methods, Cohort; Table 1 |
| Subjects | condition=stroke clinical diagnosis of acute or early subacute stroke at admission | 2,888 | liu2023 Methods, Cohort |
| Subjects | condition=ischemic_stroke radiological lesion type "ischemic"; not necessarily the clinical stroke type | 1,878 | liu2023 Table 1 |
| Subjects | condition=intracranial_hemorrhage radiological lesion type "hemorrhage": any intra- or extra-parenchymal bleeding, including hemorrhagic transformation | 540 | liu2023 Table 1 |
| Subjects | sex=female | 1,361 | liu2023 Table 1 |
| Subjects | sex=male | 1,527 | liu2023 Table 1 |
| Subjects | contrast=dwi inclusion required DWI; lesion masks are drawn on DWI | 2,888 | liu2023 Methods, Cohort |
| Subjects | contrast=FLAIR passed visual quality control | 2,746 | liu2023 Methods, MRIs |
| Subjects | contrast=T2w passed visual quality control | 2,581 | liu2023 Methods, MRIs |
| Subjects | contrast=swi passed visual quality control | 2,106 | liu2023 Methods, MRIs |
| Subjects | field_strength=1.5 | 1,766 | liu2023 Table 1 |
| Subjects | field_strength=3 | 1,122 | liu2023 Table 1 |
| Subjects | country=US Johns Hopkins Comprehensive Stroke Center | 2,888 | liu2023 Methods, Cohort |
| Mean age | total | 62.2 | liu2023 Methods, Cohort |
| Age SD | total | 14.7 | liu2023 Methods, Cohort |
| Median age | total | 62 | liu2023 Table 1 |
| Minimum age | total | 18 | liu2023 Methods, Cohort |
| Maximum age | total | 99 | liu2023 Methods, Cohort |
Sources
The keys used in the table above.