pzarzycki/mri-oasis-1-ixi-pre

---
license: other
task_categories:
- image-classification
- other
language:
- en
tags:
- medical
- mri
- brain
- neuroimaging
- oasis
- ixi
- alzheimer
- dementia
- t1
- numpy
pretty_name: OASIS-1 + IXI Preprocessed MRI
size_categories:
- 1K<n<10K
---

# OASIS-1 + IXI Preprocessed MRI

Preprocessed T1-weighted brain MRI volumes from two widely-used public neuroimaging datasets,
ready for machine learning and 3-D visualisation.

Volumes are stored as `float32` NumPy arrays of shape **(96 × 128 × 96)** (D × H × W),
so loading is as simple as `np.load("OAS1_0001_MR1_t1.npy")`.

---

## Datasets

### OASIS-1 — Open Access Series of Imaging Studies

| Property | Value |
|---|---|
| Subjects | 436 (ages 18–96) |
| Population | Cognitively normal adults + early-stage Alzheimer's disease |
| Clinical labels | CDR, MMSE, age, sex, education, SES, eTIV, nWBV |
| Channels per subject | 3 (see below) |

**Channels:**

| Channel | Description |
|---|---|
| `t1` | Talairach-registered T1 (skull present, brain-masked approximation applied) |
| `t1_masked` | Skull-stripped T1 (FSL brain extraction) |
| `fseg` | FSL tissue segmentation: 1 = CSF, 2 = GM, 3 = WM |

**Clinical metadata** is included as `oasis_preprocessed/oasis_cross-sectional.csv`
with columns: `ID, M/F, Hand, Age, Educ, SES, MMSE, CDR, eTIV, nWBV, ASF`.

**Citation:**
> Marcus, D. S., Wang, T. H., Parker, J., Csernansky, J. G., Morris, J. C., & Buckner, R. L. (2007).
> Open Access Series of Imaging Studies (OASIS): Cross-sectional MRI data in young, middle aged,
> nondemented, and demented older adults.
> *Journal of Cognitive Neuroscience*, 19(9), 1498–1507.
> https://doi.org/10.1162/jocn.2007.19.9.1498

Original data: https://sites.wustl.edu/oasisbrains/

---

### IXI — Information eXtraction from Images

| Property | Value |
|---|---|
| Subjects | 581 healthy volunteers |
| Sites | Hammersmith Hospital, Guy's Hospital, Institute of Psychiatry (London) |
| Channels per subject | 1 |

**Channels:**

| Channel | Description |
|---|---|
| `t1` | T1-weighted volume (skull-strip approximation + z-score normalised) |

**Citation:**
> IXI Dataset, Brain Development. Imperial College London.
> https://brain-development.org/ixi-dataset/

Original data: https://biomedic.doc.ic.ac.uk/brain-development/downloads/IXI/IXI-T1.tar

---

## Preprocessing pipeline

All T1 channels follow the same pipeline:

1. Load with [nibabel](https://nipy.org/nibabel/) → squeeze singleton dimensions
2. Bounding-box crop (non-zero voxels + 8-voxel margin)
3. Approximate brain masking — zero out voxels below the 15th percentile of positive signal
(applied to `t1` channels that still contain skull)
4. Z-score normalise (computed on non-zero voxels only)
5. Transpose (x, y, z) → (z, y, x)
6. Trilinear resize to **(96, 128, 96)**

The `fseg` (segmentation) channel skips steps 2–4 and uses **nearest-neighbour** resize
to preserve integer label values.

Preprocessing code: [`preprocess_oasis.py`](https://github.com/pzarzycki/contrib-experiment/blob/main/preprocess_oasis.py)
and [`download_ixi.py`](https://github.com/pzarzycki/contrib-experiment/blob/main/download_ixi.py).

---

## File structure

```
oasis_preprocessed/
├── oasis_cross-sectional.csv # Clinical metadata (CDR, MMSE, age, …)
├── OAS1_0001_MR1_t1.npy # shape (96, 128, 96), float32
├── OAS1_0001_MR1_t1_masked.npy
├── OAS1_0001_MR1_fseg.npy
└── … (436 subjects × 3 channels = 1 308 files, ~5.8 GB)

ixi_preprocessed/
├── IXI002-Guys-0828-T1.npy # shape (96, 128, 96), float32
└── … (581 subjects × 1 channel = 581 files, ~2.6 GB)
```

Total: **~8.4 GB**

---

## Usage

```python
import numpy as np
from huggingface_hub import hf_hub_download

# Download a single volume
path = hf_hub_download(
repo_id="pzarzycki/mri-oasis-1-ixi-pre",
repo_type="dataset",
filename="oasis_preprocessed/OAS1_0001_MR1_t1.npy",
)
volume = np.load(path) # (96, 128, 96) float32
```

Or with `hf-mount` for lazy access (no full download):

```bash
hf-mount start repo datasets/pzarzycki/mri-oasis-1-ixi-pre /data
ls /data/oasis_preprocessed/ | head
python -c "import numpy as np; print(np.load('/data/oasis_preprocessed/OAS1_0001_MR1_t1.npy').shape)"
```

---

## Interactive viewer

This dataset powers the [MRI Viewer space](https://huggingface.co/spaces/pzarzycki/mri-viewer) —
a browser-based 3-D MRI viewer with patient browser, CDR/MMSE filtering, per-channel
colormap/opacity controls, and WebGL2 volume rendering.

Source: [github.com/pzarzycki/contrib-experiment](https://github.com/pzarzycki/contrib-experiment) — `webapp/` directory.

---

## Licence and attribution

This is a **derivative dataset** intended for non-commercial research and educational use only.
Please respect the original licences:

- **OASIS-1**: free for non-commercial research; cite Marcus et al. (2007); registration at [oasisbrains.org](https://sites.wustl.edu/oasisbrains/) required to access the raw data
- **IXI**: [Creative Commons CC BY-SA 3.0](https://creativecommons.org/licenses/by-sa/3.0/)