[Dataset] Add SOOP (ds004889) - Acute Stroke Imaging (1714 subjects)

Dataset Info

Field	Value
Name	Stroke Outcome Optimization Project (SOOP)
Source	OpenNeuro ds004889
Paper	PubMed 39095364
License	CC0 (Public Domain)
Subjects	1714 (1449 confirmed stroke)
Format	BIDS
Size	⚠️ >600 GB (large download)
HuggingFace Target	hugging-science/soop-acute-stroke

Description

SOOP is the acute stroke companion to ARC (chronic stroke). Together they enable acute→chronic outcome prediction research.

Data includes:

• T1-weighted structural MRI
• FLAIR
• Diffusion-weighted imaging (DWI) with ADC/TRACE maps
• Demographics (Safe Harbor compliant)
• Behavioral measures

Why This Matters

1. Completes the stroke timeline - ARC has chronic, SOOP has acute
2. Clinical ML applications - Acute stroke outcome prediction
3. Large scale - 1714 subjects for robust training
4. Proves pipeline - Tests bids-hub against a new dataset structure

Exact Schema

from datasets import Features, Value
from datasets.features import Nifti, Sequence

def get_soop_features() -> Features:
    """SOOP schema - one row per SUBJECT (most are single-session).

    Note: Most acute stroke patients have a single session. If multi-session
    exists, consider one row per SESSION like ARC.
    """
    return Features({
        "subject_id": Value("string"),      # e.g., "sub-M2001"
        # Structural
        "t1w": Nifti(),                     # *_T1w.nii.gz
        "flair": Nifti(),                   # *_FLAIR.nii.gz
        # Diffusion (critical for acute stroke)
        "dwi": Sequence(Nifti()),           # *_dwi.nii.gz (may have multiple)
        "adc": Nifti(),                     # *_adc.nii.gz (Apparent Diffusion Coefficient)
        "trace": Nifti(),                   # *_trace.nii.gz (if available)
        # Metadata from participants.tsv
        "age": Value("float32"),
        "sex": Value("string"),
        "nihss": Value("float32"),          # NIH Stroke Scale (verify column name)
    })

Implementation Note: Verify participants.tsv column names during implementation. NIHSS may be named differently (e.g., nihss_admission, NIHSS, etc.).

Directory Structure

ds004889/
├── participants.tsv
├── participants.json
├── dataset_description.json
└── sub-XXXXX/
    └── ses-XX/
        ├── anat/
        │   ├── sub-XXXXX_ses-XX_T1w.nii.gz
        │   └── sub-XXXXX_ses-XX_FLAIR.nii.gz
        └── dwi/
            ├── sub-XXXXX_ses-XX_dwi.nii.gz
            ├── sub-XXXXX_ses-XX_dwi.bval
            └── sub-XXXXX_ses-XX_dwi.bvec

Files to Create

src/bids_hub/datasets/soop.py          # Dataset module (follow arc.py pattern)
src/bids_hub/validation/soop.py        # Validation rules
scripts/download_soop.sh               # Download script (S3: s3://openneuro.org/ds004889)
tests/test_soop.py                     # Tests (follow test_arc.py pattern)
docs/dataset-cards/soop-acute-stroke.md  # Dataset card (follow arc-aphasia-bids.md pattern)

Implementation Steps

1. Create download script (scripts/download_soop.sh)

   # WARNING: This dataset is >600 GB. Ensure sufficient disk space.
   aws s3 sync --no-sign-request s3://openneuro.org/ds004889 "$TARGET_DIR"
   

2. Create dataset module (src/bids_hub/datasets/soop.py)

   • Copy arc.py as template
   • Implement build_soop_file_table() - walk sub-/ses-/anat/ and dwi/
   • Implement get_soop_features() - schema above
   • Implement build_and_push_soop()

3. Add CLI commands (in cli.py)

   @app.command()
   def soop():
       """SOOP dataset commands."""
   # Subcommands: validate, build, info
   

4. Add validation (src/bids_hub/validation/soop.py)

   • Check participants.tsv exists
   • Check at least T1w OR FLAIR per subject
   • Warn if DWI missing (important for acute stroke)

5. Add tests (tests/test_soop.py)

   • Follow test_arc.py pattern
   • Create synthetic BIDS fixture with SOOP structure

6. Upload to HuggingFace

   uv run bids-hub soop build /path/to/ds004889 --hf-repo hugging-science/soop-acute-stroke --no-dry-run
   

   • Use num_shards=1714 (one per subject) to prevent OOM

Acceptance Criteria

• scripts/download_soop.sh downloads dataset successfully
• uv run bids-hub soop validate <path> passes
• uv run bids-hub soop build <path> --dry-run succeeds
• uv run pytest tests/test_soop.py passes
• uv run ruff check . && uv run mypy src tests clean
• Dataset uploaded to hugging-science/soop-acute-stroke
• HuggingFace README.md with proper frontmatter, usage examples, and citation
• docs/dataset-cards/soop-acute-stroke.md added (follow arc-aphasia-bids.md pattern)

Resources

• OpenNeuro ds004889
• SOOP Processing Demo (Chris Rorden)
• Existing ARC implementation - use as template
• arc-aphasia-bids on HF - reference

Citation

@dataset{soop2024,
  title={Stroke Outcome Optimization Project (SOOP)},
  author={Rorden, Chris and colleagues},
  year={2024},
  publisher={OpenNeuro},
  doi={10.18112/openneuro.ds004889.v1.0.0}
}