PINGU Docker image

Portable CPU-only Docker image for PINGU (Perivascular space Identification Nnunet for Generalised Usage), a nnUNet-based tool for segmentation of enlarged perivascular spaces from 3D T1-weighted MRI.

• Reference: https://doi.org/10.48550/arXiv.2405.08337
• PINGU project: https://github.com/iBrain-Lab/PINGU

Find the image in Docker Hub here.

• Debian 11 (Bullseye)
• Python 3.10
• PyTorch 2.5.1 (CPU)
• nnUNet 1.7.1
• Pretrained model: PINGU-All_v0

• Model: PINGU-All_v0
• DOI: 10.26180/27176523
• Dataset URL: https://bridges.monash.edu/articles/dataset/PINGU_Weights/27176523?file=49632888
• MD5: 97247ee4ca0f62c864550be39c47cc6e

The container entrypoint is nnUNet_predict, with PINGU default parameters pre-set:

• trainer: nnUNetTrainerV2
• cascade trainer: nnUNetTrainerV2CascadeFullRes
• model: 3d_fullres
• plans: nnUNetPlansv2.1_trgSp_8x8x8
• task: Task101_LabelsSpecificplusWP5FCV

Minimal usage:

docker run --rm -it \
  -v $(pwd)/data:/data \
  jhuguetn/pingu:latest \
  -i /data/in \
  -o /data/out

Input files must follow nnUNet naming: *_0000.nii.gz. Example:

sub-001_T1w_0000.nii.gz

The pretrained PINGU weights included in this image are licensed under:

Creative Commons Attribution 4.0 International (CC BY 4.0)

Author: Ben Sinclair
Source: https://bridges.monash.edu/articles/dataset/PINGU_Weights/27176523
Citation: Sinclair, Ben (2024). PINGU Weights. Monash University. Dataset. https://doi.org/10.26180/27176523.v1

• CPU-only image
• Model weights are included in the image
• Intended for reproducible inference workflows (HPC compatible)

Jordi Huguet (BarcelonaBeta Brain Research Center)