TL;DR This project page provides the code to generate VesselSim's high tortuosity synthetic blood vessels and skull-injected samples.

---

VesselSim is a two-stage framework for universal 3D blood vessel segmentation that eliminates the need for real annotated data during training. First, we introduce a stochastic, geometry-driven vascular simulation framework that models recursive branching, curvature-controlled growth, and collision-aware topology, followed by domain-randomized intensity synthesis to generate 16,500 anatomically plausible 3D angiographic volumes . Second, a 3D U-Net is trained solely on this synthetic data. To bridge the domain gap from synthetic to real images at inference time, we introduce a test-time adaptation strategy via a self-supervised mask reconstruction decoder, enabling adaptation to unseen clinical scans without prior domain knowledge.

For more details, please refer to our manuscript, our publicly available data, and the full project page.

Installation

First, set up a conda environment and install dependencies:

conda create -n vesselsim python=3.12
conda activate vesselsim
pip install -r requirements.txt

Data Generation Instructions

The full VesselSim dataset is seperated into low-tortuosity vessels, high-tortuosity vessels, background samples, and skull-injected samples.

1. The low-tortuosity vessel samples are generated using the VascuSynth vascular simulation framework for straighter, tree-like structures. Please refer to their software and instructions to generate these samples.
2. The high tortuosity vessel sample instructions and code are available here.
3. The background samples are 96x96x96-voxel empty .npy volumes, one sample volume is available here.
4. The skull-injected samples are any of the three other vessel volumes augmented with ellipsoidal shell structures designed to simulate skull or bone anatomy in angiographic imaging. The instructions and code are available here.

Each segmentation mask is then paired with a corresponding synthetic angiography image generated using the domain-randomized rendering pipeline introduced in VesselFM: D_drand. The high tortuosity code generates the volumes in the correct format to be easily run with VesselFM. Please refer to their code and instructions to generate the angiographic image volumes.

Citing VesselSim

The VesselSim paper has been accepted into the MICCAI 2026 conference. The peer reviewed, finalized copy will be available in October 2026. The unofficial version is currently available on arXiv. If you find our work useful for your research, please cite:

@misc{rainville2026vesselsim,
      title={VesselSim: learning 3D blood vessel segmentation without expert annotations}, 
      author={Erin Rainville and Melissa Ananian and Tristan Mirolla and Hassan Rivaz and Yiming Xiao},
      year={2026},
      eprint={2605.26277},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2605.26277}, 
}

Notes

The index.html file and static folder contain the source code for the VesselSim project page based on the Nerfies website.