AIBL Synthetic Cerebral Microbleeds on SWI
Synthetic Cerebral Microbleed on SWI images (CSIRO Data Access Portal collection 50304)
370 SWI brain scans from the AIBL study, 57 with expert-located cerebral microbleeds and 313 without, plus 3,700 copies with 10 synthetic microbleeds each and their locations, for training and benchmarking microbleed detectors.
Overview
This CSIRO collection supports work on detecting cerebral microbleeds in susceptibility-weighted MRI when few labelled lesions exist. It holds SWI scans from the Australian Imaging, Biomarker and Lifestyle (AIBL) study together with versions of the same scans into which a mathematical model has inserted artificial microbleeds at known positions. The accompanying paper (Momeni et al., 2021) reports that a classifier trained only on such synthetic lesions can detect real ones.
Composition
The file list on the Data Access Portal contains 57 SWI scans with real microbleeds and 313 scans without, each as a bias-corrected and histogram-matched NIfTI volume. Separate folders hold the copies with synthetic lesions: 570 for the microbleed scans and 3,130 for the scans without microbleeds, ten per scan. Three spreadsheets with real and synthetic microbleed information come with the scans. In total the collection has 4,073 files and about 17.7 GB. File names carry a numeric prefix and a suffix from T0 to T7; the collection does not document what they mean.
The collection description speaks of 75 scans with 175 real microbleeds, which does not match the 57 microbleed scans in the file list. The paper abstract states that the release includes a version with 37,000 synthetic lesions.
Acquisition
The scans come from AIBL, according to the collection's lineage statement. The collection does not state the scanner, field strength or sequence parameters, and the full paper is not openly accessible.
Annotations
Experts defined the locations of the real microbleeds; the labels are positions, not voxel masks. The scans with microbleeds are filed as definite cases. Synthetic microbleeds were modelled with a random Gaussian shape and placed in healthy brain locations, ten per scan, with their positions provided.
Known limitations
- Point locations only; no segmentation masks.
- No demographic or clinical data are included, and the number of distinct participants is not stated.
- The collection description and the files disagree on the number of real microbleed scans.
- The CSIRO Data Licence allows non-commercial use only and forbids passing the data on.
Cohort
Aggregate numbers from the sources below. Bars are relative to the largest value.
Contrast / sequence
scans, values can overlap
- Susceptibility-weighted 370 100%
Condition
scans, values can overlap
- Cerebral microbleeds 57 15%
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
CSIRO Data Licence
A personal, non-transferable licence for non-commercial use of data on the CSIRO Data Access Portal. The data may not be passed on, and no patent or other IP rights may be claimed over it. Free hardcopy products and non-editable images based on the data may be shared, and CSIRO must not be presented as endorsing the use or its results.
What you can do
- No
- Conditional
- Conditional
- Conditional
What you can share
- No
- Conditional
- Share trained models Not stated
What you must do
- Yes
- Share alike No
- No
- No
- Manuscript review No
- Release code No
- Return results No
- Delete after use No
Limits
- No
- Location limits No
Citation
Momeni S, Fazlollahi A, Yates P, Rowe C, Gao Y, Liew AWC, Salvado O. Synthetic microbleeds generation for classifier training without ground truth. Computer Methods and Programs in Biomedicine 207, 106127 (2021). doi:10.1016/j.cmpb.2021.106127. Data: Momeni S, et al. (2021) Synthetic Cerebral Microbleed on SWI images. v1. CSIRO. Data Collection. https://doi.org/10.25919/aegy-ny12
All numbers
Every number on this page, as stored in stats.csv, with its source.
| Measure | Breakdown | Value | Source |
|---|---|---|---|
| Scans | total real SWI scans; the 3,700 copies with synthetic microbleeds are not counted | 370 | csiro-dap-50304-files folders rCMB_DefiniteSubject and NoCMBSubject |
| Scans | contrast=swi | 370 | csiro-dap-50304-files folders rCMB_DefiniteSubject and NoCMBSubject |
| Scans | condition=cerebral_microbleeds real scans with definite microbleeds; the collection description says 75 scans with 175 microbleeds | 57 | csiro-dap-50304-files folder rCMB_DefiniteSubject |
Sources
The keys used in the table above.
- csiro-dap-50304 CSIRO Data Access Portal collection 50304 (description, lineage, licence, size) website
- csiro-dap-50304-files File list of collection 50304 from the Data Access Portal API (CSIRO Data Licence, no registration), NIfTI files counted per folder computed
- momeni2021 Momeni et al. 2021, Computer Methods and Programs in Biomedicine (abstract, PubMed 34051412) paper