M4Raw
M4Raw: A multi-contrast, multi-repetition, multi-channel MRI k-space dataset for low-field MRI research
Raw four-channel brain k-space from 208 healthy volunteers scanned on a 0.3 T MRI system, with repeated T1w, T2w and FLAIR acquisitions per subject, for low-field MRI reconstruction, denoising and parallel imaging research.
Overview
M4Raw is a raw k-space collection of brain MRI acquired on a low-field (0.3 T) whole-body scanner, released by a team led from Shenzhen Technology University. High-field raw datasets such as fastMRI do not capture the signal and noise behaviour of low-field systems, so M4Raw was built to support learning-based reconstruction, denoising and parallel imaging methods for affordable MRI. Every contrast was scanned several times in a row, so single repetitions can serve as noisy inputs and their average as a cleaner target.
Composition
The original release (V1.1, described in the paper) covers 183 healthy volunteers, mostly college students aged 18 to 32, of whom 116 are male and 67 female. Each has three T1w, three T2w and two FLAIR repetitions of 18 axial slices. Volunteers with notable head motion were moved into a separate motion-corrupted subset; the rest were split at random into 128 training and 30 validation subjects (1,024 and 240 files). V1.5 added a T1-weighted gradient echo series with two repetitions for all 183 volunteers. V1.6 added a test subset of 25 new volunteers with six T1w, six T2w and four FLAIR repetitions each, together with the reference images. The current release has 2,230 files from 208 people.
Acquisition
All data come from one Oper-0.3 permanent-magnet scanner (Ningbo Xingaoyi) with a four-channel head coil. Slices are 5 mm thick with a 1 mm gap, with an in-plane resolution of about 0.94 x 1.23 mm. The k-space of every repetition was exported without averaging, shifted to correct left-right off-centering, and stored as HDF5 with an ISMRMRD-style header and a root-sum-of-squares image, in the array layout used by fastMRI.
Annotations
There are no clinical or anatomical labels. Reconstruction targets are the per-repetition root-sum-of-squares images and the averages over repetitions.
Known limitations
- One scanner model at one field strength, and a young, narrow-age cohort.
- Images were not defaced; the thick slices cover only the upper half of the head.
- Age and sex are published only for the original 183 volunteers.
- The paper gives 26 motion-corrupted subjects in one place and 25 in another; the released files and the creators' later inventory list 25.
- The test subset has twice as many repetitions, and its inter-contrast motion is about twice as large as in the training and validation subsets.
Cohort
Aggregate numbers from the sources below. Bars are relative to the 208 subjects.
Sex
- Male 116 63%
- Female 67 37%
Covers 183 of 208 subjects.
Age
mean 20.1± 1.5 · range 18 to 32No age bins reported.
Contrast combinations
How many subjects have exactly each set of contrasts.
| T1w | T2w | FLAIR | Subjects with exactly this set |
|---|---|---|---|
208 |
Contrast / sequence
subjects, values can overlap
- T1-weighted 208 100%
- T2-weighted 208 100%
- FLAIR 208 100%
Condition
subjects, values can overlap
- Healthy control 208 100%
Field strength
subjects
- 0.3 T 208 100%
Split
subjects
- Training 128 62%
- Validation 30 14%
- Test 25 12%
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
Creative Commons Attribution 4.0 International
Use, share and adapt the data for any purpose, including commercial use, as long as you credit the creators.
What you can do
- Yes
- Yes
- Yes
- Yes
What you can share
- Yes
- Yes
- Yes
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
Lyu M, Mei L, Huang S, et al. M4Raw: A multi-contrast, multi-repetition, multi-channel MRI k-space dataset for low-field MRI research. Sci Data 10, 264 (2023). https://doi.org/10.1038/s41597-023-02181-4
@article{lyu_m4raw_2023,
title = {{M4Raw}: {A} multi-contrast, multi-repetition, multi-channel {MRI} k-space dataset for low-field {MRI} research},
volume = {10},
doi = {10.1038/s41597-023-02181-4},
number = {1},
journal = {Scientific Data},
author = {Lyu, Mengye and Mei, Lifeng and Huang, Shoujin and Liu, Sixing and Li, Yi and Yang, Kexin and Liu, Yilong and Dong, Yu and Dong, Linzheng and Wu, Ed X.},
year = {2023},
pages = {264},
} All numbers
Every number on this page, as stored in stats.csv, with its source.
| Measure | Breakdown | Value | Source |
|---|---|---|---|
| Subjects | total 183 participants of the original cohort plus 25 test participants added in V1.6 | 208 | hf-m4raw-brain Release inventory |
| Subjects | condition=healthy all participants are healthy volunteers | 208 | hf-m4raw-brain Release inventory |
| Subjects | contrast=T1w the 183 original participants also have a T1-weighted gradient echo series (366 files) | 208 | hf-m4raw-brain Release inventory |
| Subjects | contrast=T2w | 208 | hf-m4raw-brain Release inventory |
| Subjects | contrast=FLAIR | 208 | hf-m4raw-brain Release inventory |
| Subjects | contrast_set=FLAIR+T1w+T2w | 208 | hf-m4raw-brain Release inventory |
| Subjects | split=train | 128 | lyu2023 Subset partition |
| Subjects | split=validation | 30 | lyu2023 Subset partition |
| Subjects | split=test new participants not in the paper; the remaining 25 original participants form a separate motion-corrupted subset | 25 | zenodo-m4raw-v1-6 V1.6 release notes |
| Subjects | sex=male original cohort of 183 only; sex of the 25 test participants is not reported | 116 | lyu2023 Imaging protocol |
| Subjects | sex=female original cohort of 183 only; sex of the 25 test participants is not reported | 67 | lyu2023 Imaging protocol |
| Subjects | field_strength=0.3 one 0.3 T Oper-0.3 scanner (Ningbo Xingaoyi) | 208 | hf-m4raw-brain Release inventory |
| Mean age | total original cohort of 183 only | 20.1 | lyu2023 Imaging protocol |
| Age SD | total original cohort of 183 only | 1.5 | lyu2023 Imaging protocol |
| Minimum age | total original cohort of 183 only | 18 | lyu2023 Imaging protocol |
| Maximum age | total original cohort of 183 only | 32 | lyu2023 Imaging protocol |
| Scans | total H5 files, one per single repetition of one contrast | 2,230 | hf-m4raw-brain Release inventory |
| Scans | split=train | 1,024 | lyu2023 Data Records |
| Scans | split=validation | 240 | lyu2023 Data Records |
| Scans | split=test six T1w, six T2w and four FLAIR repetitions per participant | 400 | hf-m4raw-brain Release inventory |
Sources
The keys used in the table above.
- lyu2023 Lyu et al. 2023, Scientific Data 10, 264 paper
- zenodo-m4raw-v1-6 Zenodo record of M4Raw V1.6 (release notes and files) data
- hf-m4raw-brain M4Raw Brain v1.6 dataset card on Hugging Face (by the first author) website