---
license: cc-by-nc-4.0
task_categories:
  - other
language:
  - en
pretty_name: BrainAge Golden Preprocessed Cache
size_categories:
  - 1K<n<10K
tags:
  - neuroimaging
  - mri
  - brain-age
  - pytorch
  - preprocessed
  - tensor-cache
---

# BrainAge Golden Preprocessed Cache

**6,050 preprocessed brain MRI tensors** ready for training a brain-age
prediction model. Skip the 40+ hour preprocessing step and jump straight
to model training.

## What's inside

Each `.pt` file (one per subject) contains:

| Key | Type | Shape | Description |
|-----|------|-------|-------------|
| `volume` | float16 | (128, 144, 112) | Z-normed T1w brain in MNI space, trilinear-resized |
| `tab` | float32 | (86,) | 70 regional volumes (log1p/12) + 3 sex one-hot + 13 site one-hot |
| `age` | float32 | scalar | Chronological age in years |
| `meta` | dict | — | subject_id, site, sex, age, split |

## Stats

| Metric | Value |
|--------|-------|
| Total subjects | 6,050 |
| Age range | 0 – 86 years |
| Source datasets | 12 (BCP, Calgary, ds002726, ds000248, PTBP, IXI, MPI-Leipzig, AOMIC, NKI-Rockland, ABIDE-I, ABIDE-II, ADHD-200) |
| Volume shape | 128 × 144 × 112 (D × H × W) |
| Tabular dim | 86 (70 regions + 3 sex + 13 site) |
| File size | ~4 MB each |
| Total size | ~24 GB |

## Preprocessing pipeline applied

```
Raw T1w NIfTI
  → HD-BET skull-strip (GPU)
  → N4 bias correction (ANTs)
  → Affine registration to MNI152 1mm
  → Z-score intensity normalization
  → Harvard-Oxford atlas segmentation (69 regions)
  → Volume measurement + rescaling to native space
  → Tensor packaging (.pt)
```

## Quick start

```python
from huggingface_hub import snapshot_download
import torch

# Download (~24 GB)
snapshot_download(
    "bilalahmad176176/BrainAge-Golden-Preprocessed",
    repo_type="dataset",
    local_dir="cache/"
)

# Load one subject
data = torch.load("cache/cache/IXI002.pt", weights_only=False)
print(data["volume"].shape)  # (128, 144, 112) float16
print(data["tab"].shape)     # (86,) float32
print(data["age"])            # e.g. 36.2
print(data["meta"])           # {'subject_id': 'IXI002', 'site': 'DataSet-6_IXI', ...}
```

## Train a model

```bash
# Generate split
python -m pipeline_v2.data_split \
    --manifests Golden-0-to-25/manifest.csv Golden-25plus/manifest.csv \
    --out cache/split.csv

# Train
python -m pipeline_v2.train \
    --cache_dir cache/cache \
    --split_csv cache/split.csv \
    --out_ckpt brainage_sfcn.pt \
    --epochs 60 --batch 4
```

## Related

- Raw dataset: [bilalahmad176176/BrainAge-Golden-Raw](https://huggingface.co/datasets/bilalahmad176176/BrainAge-Golden-Raw)
- 3D Viewer demo: [bilalahmad176176/BrainAge-3D-Viewer](https://huggingface.co/spaces/bilalahmad176176/BrainAge-3D-Viewer)

## Citation

Please cite the original source studies listed in the raw dataset manifests.