Brain-Creation / README.md
AthensJ's picture
Update README.md
a125ae7 verified
|
Raw History Blame Contribute Delete
3.79 kB
---
license: mit
language:
- en
tags:
- fMRI
- neuroscience
- neuroimaging
- mental-imagery
pretty_name: Brain-Creation
---
# Brain-Creation: Prompt-Guided Mental Creation fMRI
[Dataset repository](https://huggingface.co/datasets/AthensJ/Brain-Creation)
Brain-Creation follows a visual imagination paradigm that separates perception, memory maintenance, prompt comprehension, mental creation, and behavioral report. The current release contains fMRI series for data loading, temporal alignment, and analysis workflows.
## Dataset Overview
- **Subject series:** 6 (`sub-01` to `sub-06`)
- **Design:** 982 image indices, with 3 prompt conditions per image
- **Total trials:** 10,998
- **Runs:** 552 creation runs and 186 rest runs
- **Files:** 738 compressed NIfTI images and 738 JSON sidecars
- **Total size:** approximately 22.35 GB
## Data Organization
### Condition Assignment
- **Subjects 1-3:** each covers all 982 image indices and 2,946 conditions.
- **Subjects 4-6:** each covers 240 image indices and 720 conditions. Their image subsets are mutually disjoint.
Image indices are local design identifiers, not verified NSD image IDs. Stimulus images, prompt text, and trial-to-condition mappings are not included.
### Repository Contents
Representative layout; each subject folder contains all of its runs directly:
```text
Brain-Creation/
├── sub-01/
│ ├── sub-01_ses-01_task-creation_run-01_bold.nii.gz
│ └── sub-01_ses-01_task-creation_run-01_bold.json
├── sub-02/
├── sub-03/
├── sub-04/
├── sub-05/
└── sub-06/
```
Filenames identify the subject, session, task, and run. Tasks are `creation` and `rest`, with run indices counted separately for each task within a session. Each image has a matching JSON sidecar containing timing, volume counts, and protocol settings. Naming uses BIDS-style fields; this is not a complete BIDS dataset.
### Data Statistics
| Subject | Sessions | Creation Runs | Rest Runs | Trials |
|---------|---------:|--------------:|----------:|-------:|
| sub-01 | 25 | 148 | 50 | 2,946 |
| sub-02 | 25 | 148 | 50 | 2,946 |
| sub-03 | 25 | 148 | 50 | 2,946 |
| sub-04 | 6 | 36 | 12 | 720 |
| sub-05 | 6 | 36 | 12 | 720 |
| sub-06 | 6 | 36 | 12 | 720 |
| **Total** | **93** | **552** | **186** | **10,998** |
## Temporal Alignment and Trial Structure
Each trial lasts **24 seconds**, corresponding to **12 volume intervals** at TR = 2 seconds.
| Phase | Duration |
|-------|---------:|
| Fixation | 2 s |
| Image perception | 4 s |
| Memory maintenance | 2 s |
| Prompt comprehension | 6 s |
| Mental creation | 6 s |
| Behavioral report | 4 s |
A standard creation run contains **20 trials**, spanning **480 seconds / 240 volumes**. The final creation run for each of subjects 1-3 contains **6 trials / 72 volumes**. Each session has beginning and ending rest runs, each retaining **170 seconds / 85 volumes**.
## fMRI Specifications
- **Spatial dimensions:** 72 × 72 × 48 voxels
- **Voxel size:** 3 × 3 × 3 mm
- **TR:** 2 seconds
- **Storage:** NIfTI-1, `int16`, intensity scale approximately 0.1
- **Compression:** gzip (`.nii.gz`)
## Important Notes
- The initial five volumes have already been removed. Do not discard them again.
- Creation files contain the complete trial sequence, including all six phases.
- The release includes BOLD files and run-level JSON only; event tables, behavioral data, anatomical scans, and analysis derivatives are not included.
- NiBabel reads `.nii.gz` directly and applies NIfTI intensity scaling automatically.
## Additional Information
- **License:** [MIT](https://opensource.org/license/mit), as declared in the repository metadata.
- **Version:** Record the repository URL and exact revision when reporting use of the data.