VesselBoost

VesselBoost is a Python-based software package utilizing deep learning techniques to segment high-resolution time-of-flight MRI angiography data, with high sensitivity towards small vessels (An experimental pretrained model is available for T2*-weighted imaging). The software suite encompasses three essential functional modules: (1) predict, (2) test-time adaptation (TTA), and (3) boost. By leveraging these modules, users can efficiently segment high-resolution time-of-flight data or conveniently leverage our command line interface to boost segmentations for other vascular MRI image contrasts.

Table of Contents

• Update History
• Purpose
• Current Version
• Requirements
• Software container
• Installation
• Citation
• Contact

Update History

• 2.0.2 - patch release: for details see Update Log - 13/Apr/2026
• 2.0.1 - patch release: for details see Update Log - 06/Mar/2026
• 2.0.0 - official release: for details see Update Log - 18/Feb/2026
• 2.0.0 - pre-release: for details see Update Log - 16/Sept/2025
• 1.0.0: Initial release, for details see Citation

Purpose

VesselBoost is a Python-based software package leveraging a UNet3D-based segmentation pipeline that utilizes data augmentation and test-time adaptation (TTA) to enhance segmentation quality and is generally applicable to high-resolution magnetic resonance angiograms (MRAs).
This repository contains 3 major modules:

1. Predict. With this module, users can segment high-resolution time-of-flight using our pre-trained models. It can be used to generate intermediate proxy segmentations as well as the final ones.
2. Test-time-adaptation. This module allows the user to provide a proxy segmentation or generate a proxy with our pre-trained model (Module 1), to drive further adaptation of the pre-trained models.
3. Booster. Boost allows users to train a segmentation model on a single or more data using existing imperfect segmentation.

Current Version

VesselBoost 2.0.2

Requirements

• Docker / Singularity container

Availability

Docker

The Dockerhub container is available at Dockerhub. To download the container, run the following command:

docker pull vnmd/vesselboost_2.0.1

Neurodesk

To predict vessel segmentation using your data and the latest version of VesselBoost on Neurodesk, you can run the following code snippet:

ml vesselboost
path_to_model=/cvmfs/neurodesk.ardc.edu.au/containers/vesselboost_2.0.0_20250916/vesselboost_2.0.0_20250916.simg/opt/VesselBoost/saved_models/
prediction.py --image_path /path/ --output_path /path/ --pretrained "$path_to_model"/BM_VB2_aug_all_ep2k_bat_10_0903 --prep_mode 4

For more information, please check our notebooks.

VesselBoost Webapp

VesselBoost is also available as a web application. To access the webapp, please visit the VesselBoost Webapp.

OpenRecon

VesselBoost is also available on Siemens OpenRecon. To run VesselBoost on OpenRecon enabled scanners (>XA60), please refer to the open recon container.

Installation

This is a Python-based software package. To successfully run this project on your local machine, please follow the following steps to set up the necessary software environment.

1. Clone this repository to your local machine For latest version:

   git clone https://github.com/KMarshallX/VesselBoost.git
   
   

   To clone the previous version (VesselBoost 1.0.0):

   git clone -b stable_ver_1_0_0_hpc --single-branch https://github.com/KMarshallX/VesselBoost.git
   

2. Install miniconda:

   cd VesselBoost
   bash miniconda-setup.sh
   

3. Then set your current working directory as the cloned repository, and install the remaining required packages

   conda env create -f environment.yml
   conda activate vessel_boost
   

4. (CI or CPU-only) If you run the CI tests or run VesselBoost on a CPU-only machine, please install the CPU version of PyTorch by running the following command:

   conda env create -f environment-ci.yml
   conda activate vessel_boost_ci
   

Brain extraction in offline environments

Brain extraction uses FreeSurfer's SynthStrip and requires the synthstrip.1.pt weights file. If the file is not available locally, VesselBoost tries to download it from the FreeSurfer server at runtime. When there is no internet connection and no local weights file, brain extraction fails with an error.

On a connected machine, download the weights into the standard VesselBoost location:

mkdir -p saved_models
curl -L \
  -o saved_models/synthstrip.1.pt \
  https://surfer.nmr.mgh.harvard.edu/docs/synthstrip/requirements/synthstrip.1.pt

If curl is unavailable, use wget:

wget \
  -O saved_models/synthstrip.1.pt \
  https://surfer.nmr.mgh.harvard.edu/docs/synthstrip/requirements/synthstrip.1.pt

For airgapped or offline deployments, copy saved_models/synthstrip.1.pt into the deployment image or runtime directory before running VesselBoost. Alternatively, set VESSELBOOST_SYNTHSTRIP_WEIGHTS to the weights file path or to the directory containing it.

Citation

VesselBoost paper is now published! Please cite us if you use VesselBoost in your research:

@article{xuVesselBoostPythonToolbox2024a,
  title = {{{VesselBoost}}: {{A Python Toolbox}} for {{Small Blood Vessel Segmentation}} in {{Human Magnetic Resonance Angiography Data}}},
  shorttitle = {{{VesselBoost}}},
  author = {Xu, Marshall and Ribeiro, Fernanda L. and Barth, Markus and Bernier, Micha{\"e}l and Bollmann, Steffen and Chatterjee, Soumick and Cognolato, Francesco and Gulban, Omer F. and Itkyal, Vaibhavi and Liu, Siyu and Mattern, Hendrik and Polimeni, Jonathan R. and Shaw, Thomas B. and Speck, Oliver and Bollmann, Saskia},
  year = {2024},
  month = sep,
  journal = {Aperture Neuro},
  volume = {4},
  publisher = {Organization for Human Brain Mapping},
  issn = {2957-3963},
  doi = {10.52294/001c.123217},
  urldate = {2024-09-17},
  copyright = {http://creativecommons.org/licenses/by/4.0},
  langid = {english}
}

Contact

Marshall Xu <marshall.xu@uq.edu.au>

Saskia Bollmann <saskia.bollmann@uq.edu.au>

Fernanda Ribeiro <fernanda.ribeiro@uq.edu.au>