Best Performing TopBrain Segmentation Dockers

This Zenodo upload contains the best performing segmentation dockers from the TopBrain 2025 challenge, along with instructions and scripts to help you run them locally.

The dockers are from the top-3 teams (for more info please refer to the TopBrain challenge paper in Citation) for CTA and MRA modalities: team ARG, KDH, and UZH. The docker files are named as 

Team_{team-name}_{year}_topbrain_segmentation_{modality}.tar.gz

The `modality` in the docker file name indicates whether the docker is for CT or MR angiographies. The docker from team UZH works for both CTA and MRA modalities.

How to Run Predictions Yourself

Pre-requisite for the Input Images: LPS+

The only pre-requisite for the images is orientation. The image MUST be in LPS+ orientation. We provide a Python script `reorient_nii.py` that you can directly use to re-orient your input images to LPS+.

python3 reorient_nii.py <input_nifti_file> LPS

The input image to the dockers can be images of these types: "*.nii.gz", "*.nii", and "*.mha".

Optional crop to braincase region: For best results and also to reduce memory, you can crop the input images to the braincase region.

Docker Load Image and `run_docker_topbrain_2025.py`

Once you have downloaded a team's docker, first load the docker image with `docker image load -i`:

docker image load -i <downloaded_docker.tar.gz>

Then you need to note down the loaded docker image's REPOSITORY:TAG from `docker images`. For example, when you run:

$ docker images
REPOSITORY    TAG       IMAGE ID       CREATED        SIZE
topbrain-ct   v2.0      a8961a3dd01c   7 months ago   9.59GB

In the above example, the REPOSITORY:TAG will be `topbrain-ct:v2.0`. You need to know the repo:tag pair to run the docker container below.

With the folder containing your input images to be predicted, the modality, and the noted repo:tag, you can simply run the provided Python script `run_docker_topbrain_2025.py` as follows:

python3 run_docker_topbrain_2025.py \
    --img_src <input_img_folder> \
    --modality <modality> \
    --repo_tag <REPOSITORY:TAG>

The predictions are saved in the folder Saved_predictions_<modality>_<timestamp>.

TopBrain Data

The TopBrain data used to train these Docker images is available on another Zeonodo at: https://zenodo.org/records/16878417

Citation

The dockers in this Zenodo upload were submitted to the TopBrain challenge for benchmarking. For more details on the algorithms and teams of the best performing dockers, please refer to our TopBrain challenge summary paper:

Yang, Kaiyuan, Pengcheng Shi, Houjing Huang, Fabio Musio, Hakim Baazaoui, Orhun Utku Aydin, Adam Hilbert et al. "TopBrain segmentation challenge for whole brain vessel anatomy." medRxiv (2026): 2026-05.

Please cite the above TopBrain paper if you use the Docker images from this Zenodo upload.