MeshGrow

Unified pipeline for patient-specific cardiac + vascular simulation mesh construction:

1. nnU-Net — binary cardiac localization
2. Crop — subvolume around the heart
3. LinFlo-Net — whole-heart mesh and segmentation
4. SeqSeg — aortic/vascular tracing (seeded from cardiac mesh)
5. Combine — merged simulation-ready model ({case_id}_LV_aorta.vtp)

Quick start

python3 -m venv .venv
source .venv/bin/activate   # Windows: .venv\Scripts\activate

pip install -e .

# Pipeline tools (nnU-Net, SeqSeg, LinFlo-Net)
pip install seqseg linflonet nnunetv2

# PyTorch — install before pytorch3d; use a CUDA wheel on Linux if you have a GPU
pip install torch

# pytorch3d (required by LinFlo-Net; not on PyPI for all platforms)
pip install --no-build-isolation \
  "git+https://github.com/facebookresearch/pytorch3d.git@stable"

meshgrow download-weights --dest models/
meshgrow doctor

meshgrow run \
  --input /path/to/images \
  --output /path/to/results \
  --modality ct

Install torch before pytorch3d. On macOS, use the default CPU torch build; MeshGrow runs nnU-Net on CPU automatically (runtime.device: auto). GPU builds are strongly recommended on Linux for nnU-Net and SeqSeg.

meshgrow doctor checks nnU-Net, SeqSeg, and LinFlo-Net CLIs but not pytorch3d — verify with:

python -c "import torch; import pytorch3d; print(torch.__version__, pytorch3d.__version__)"

No config file is required — built-in defaults are used after weights are downloaded.

Optional project scaffold:

meshgrow init --dest ./my_project
meshgrow run --config ./my_project/pipeline.yaml \
  --input ./my_project/images \
  --output ./results \
  --modality mr

Model weights

Optional: use --cardiac-path ./nnUNet_cardiac_weights instead of downloading from Zenodo.

CLI

Resume from a step:

meshgrow run --output results/ --input images/ --modality ct \
  --case case_001 --from-step seqseg

Documentation

• Installation
• Pipeline

Citation

When using this workflow, please cite SeqSeg, LinFlo-Net, and the underlying nnU-Net models. See the respective project pages for BibTeX entries.