liamchalcroft/FOMO300K-jvol

---
license: cc-by-nc-sa-4.0
tags:
- brain
- mri
- ssl
- foundation_model
- 3d
- image
- jvol
- compressed
pretty_name: FOMO-300K (jvol-compressed)
size_categories:
- 100K<n<1M
task_categories:
- image-feature-extraction
- zero-shot-classification
viewer: false
---

# FOMO-300K (jvol-compressed)

This is a **wavelet-compressed version** of the [FOMO-300K](https://huggingface.co/datasets/FOMO-MRI/FOMO300K) brain MRI dataset, using the [jvol](https://github.com/fepegar/jvol-rust) compression format.

All NIfTI files (`.nii.gz`) have been compressed to `.jvol` format using **quality 60** (lossy), achieving approximately **4.1x compression** over the original gzipped NIfTI data.

## What is jvol?

[jvol](https://github.com/fepegar/jvol-rust) is a wavelet-based JPEG compression format for 3D medical images. It supports both lossy and lossless compression and can achieve significant compression ratios while preserving image quality suitable for most research and training applications.

Install jvol:
```bash
pip install jvol
```

### Reading jvol files in Python

```python
import jvol

# Load a jvol file as a nibabel-compatible image
img = jvol.read("path/to/scan.jvol")

# Access the data as a numpy array
data = img.get_fdata()

# Access the affine matrix
affine = img.affine
```

### Converting back to NIfTI

```bash
# Command line
jvol decompress scan.jvol scan.nii.gz

# Or in Python
import jvol
import nibabel as nib

img = jvol.read("scan.jvol")
nib.save(img, "scan.nii.gz")
```

## Compression details

| | Original (FOMO-300K) | Compressed (jvol q=60) |
|---|---|---|
| **Total size** | ~2.3 TB | ~560 GB |
| **Compression ratio** | — | ~4.1x |
| **Format** | `.nii.gz` | `.jvol` |
| **Quality setting** | — | 60 (lossy) |

## Dataset structure

The directory structure mirrors the original FOMO-300K layout, with `.nii.gz` files replaced by `.jvol`:

```
PT001_ClevelandCCF/
sub-01/
ses-01/
t1w.jvol
PT002_Nigerian_Clinical/
sub-01/
ses-01/
t1w.jvol
...
```

The metadata files (`participants.tsv`, `mapping.tsv`, `mri_info.tsv`) are included unchanged from the original dataset.

## Original dataset

This is a derived version of **FOMO-300K**, a large-scale dataset of brain MRI scans containing:
- 59,969 subjects
- 82,678 sessions
- 318,877 scans

across 37 dataset collections, including both clinical and research-grade scans with a wide range of sequences (T1, MPRAGE, T2, T2*, FLAIR, SWI, T1c, PD, DWI, ADC, and more).

For full details about the original dataset, see:
- **Dataset**: [FOMO-MRI/FOMO300K](https://huggingface.co/datasets/FOMO-MRI/FOMO300K)
- **Paper**: [A large-scale heterogeneous 3D magnetic resonance brain imaging dataset for self-supervised learning](https://arxiv.org/abs/2506.14432)

## License & attribution

This compressed version is distributed under the same [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/) license as the original FOMO-300K dataset. All original dataset licenses, data use agreements, and citation requirements apply. Please refer to the [original dataset page](https://huggingface.co/datasets/FOMO-MRI/FOMO300K) for full terms.

## Citation

If you use this dataset, please cite the original FOMO-300K paper:

```bibtex
@article{Cerri2026large,
title={A large-scale heterogeneous 3D magnetic resonance brain imaging dataset for self-supervised learning},
author={Cerri, Stefano and Munk, Asbj{\o}rn and Llambias, Sebastian N{\o}rgaard and Ambsdorf, Jakob and Machnio, Julia and Nersesjan, Vardan and Hedeager Krag, Christian and Liu, Peirong and Rocamora Garc{\'\i}a, Pablo and Mehdipour Ghazi, Mostafa and Boesen, Mikael and Benros, Michael Eriksen and Iglesias, Juan Eugenio and Nielsen, Mads},
journal={arXiv preprint arXiv:2506.14432},
year={2026},
url={https://arxiv.org/abs/2506.14432}
}
```

And cite the jvol compression tool:

```bibtex
@software{jvol,
author={P\'{e}rez-Garc\'{i}a, Fernando},
title={jvol: Wavelet-based compression for 3D medical images},
url={https://github.com/fepegar/jvol-rust}
}
```