Brain MRI Inference Pipeline

CLI pipeline for Brain MRI inference and SLURM submission.

Flow: input NIfTI -> preprocess -> model -> outputs

Quick Start

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
./brain run-tasks segmentation brainage idh mci stroke sequence survival \
  --input-dir /path/to/sub-01/anat \
  --output-dir /path/to/outputs

Use only validated tasks:

./brain run-tasks segmentation brainage idh mci stroke \
  --input-dir /path/to/sub-01/anat \
  --output-dir /path/to/outputs

Submit to SLURM:

./brain submit-models \
  --input-dir /path/to/cohort \
  --tasks segmentation,brainage,idh,mci,stroke \
  --output-dir /path/to/outputs \
  --job-prefix brain_task

Check jobs/logs:

./brain status
./brain status --job-ids 12345,12346
./brain logs --path ./slurm_logs --lines 80

Outputs

Task folders are created under --output-dir (for example: segmentation, brainage, idh, mci, stroke). Quantitative summaries are created automatically:

• quantitative_summary.csv
• quantitative_summary.json

Included fields:

• brainage: predicted_age
• idh, mci, stroke: pred_prob, pred_label, pred_logit
• segmentation: mask_nonzero_voxels, mask_nonzero_fraction, mask_shape

Known Limits

• Validated tasks in this environment: segmentation, brainage, idh, mci, stroke
• sequence is skipped when the checkpoint is unreadable
• survival requires a ViT-compatible os.ckpt; incompatible files are skipped

Optional Generic Path

If you need TorchScript inference:

python infer_pipeline.py \
  --input /path/to/subject.nii.gz \
  --model-path /path/to/model.pt \
  --output-dir ./outputs \
  --save-preprocessed