PeTu

PeTu is a fully automated pipeline for segmenting pediatric brain tumors. It uses a 3D nnU-Net model trained on co-registered multi-parametric MRI scans, including T1c, T1n, T2w, and T2f sequences. Subsequently, the model provides segmented tumor regions, including:

1. T2-hyperintense region (T2H) – typically encompassing solid tumor mass and associated edema.
2. Enhancing tumor (ET) – regions with contrast uptake, indicative of active or aggressive tumor areas.
3. Cystic component (CC) – fluid-filled regions often seen in certain pediatric tumor types.

Features

Data Requirements

PeTu is trained on pediatric brain MRI data from the Children's Hospital Zurich (Kispi), including cases with optic glioma affecting the optic nerve.

The model expects multi-parametric MRI scans (T1c, T1n, T2w, T2f) that are co-registered to T1c and then brought into SRI-24 brain atlas space.

[!IMPORTANT]
Since PeTu handles optic gliomas affecting the optic nerve, input data should be brain scans without defacing or skull-stripping to preserve critical anatomical structures. However, it may be worth experimenting with skull-stripped (BET) or defaced brain scans depending on your specific use case.

We recommend using the preprocessing package, part of the BrainLesion Suite, to design custom preprocessing pipelines tailored to your specific needs.

Installation

With a Python 3.10+ environment, you can install petu directly from PyPI:

pip install petu

Use Cases and Tutorials

A minimal example to create a segmentation could look like this:

from petu import Inferer

inferer = Inferer()

# Save NIfTI files
inferer.infer(
    t1c="path/to/t1c.nii.gz",
    fla="path/to/fla.nii.gz",
    t1="path/to/t1.nii.gz",
    t2="path/to/t2.nii.gz",
    ET_segmentation_file="example/ET.nii.gz",
    CC_segmentation_file="example/CC.nii.gz",
    T2H_segmentation_file="example/T2H.nii.gz",
)

# Or directly use pre-loaded NumPy data. (Both outputs work as well)
et, cc, t2h = inferer.infer(
    t1c=t1c_np,
    fla=fla_np,
    t1=t1_np,
    t2=t2_np,
)

[!NOTE]
If you're interested in the PeTu package, the Pediatric Segmentation may also be of interest.

Citation

Please support our development by citing the following manuscripts:

Enhancing efficiency in paediatric brain tumour segmentation using a pathologically diverse single-center clinical dataset

@misc{piffer2025enhancingefficiencypaediatricbrain,
      title={Enhancing efficiency in paediatric brain tumour segmentation using a pathologically diverse single-center clinical dataset}, 
      author={A. Piffer and J. A. Buchner and A. G. Gennari and P. Grehten and S. Sirin and E. Ross and I. Ezhov and M. Rosier and J. C. Peeken and M. Piraud and B. Menze and A. Guerreiro Stücklin and A. Jakab and F. Kofler},
      year={2025},
      eprint={2507.22152},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2507.22152}, 
}

Contributing

We welcome all kinds of contributions from the community!

Reporting Bugs, Feature Requests and Questions

Please open a new issue here.

Code contributions

Nice to have you on board! Please have a look at our CONTRIBUTING.md file.