LinFlo-Net: Pre-trained model weights for whole-heart mesh generation from CT and MR images

LinFlo-Net Pre-trained Weights

This record provides pre-trained PyTorch model weights for LinFlo-Net, a deep learning method for automatically generating simulation-ready 3D whole-heart meshes from cardiac CT and MR images.

Contents

What the models do

Given a 3D cardiac image and a template heart mesh, LinFlo-Net predicts a patient-specific deformed mesh (`.vtp`) and an associated segmentation rasterized to image space. The models were developed for whole-heart meshing workflows, including computational modeling and simulation preparation.

Training data

Models were trained on the Multi-Modality Whole Heart Segmentation (MMWHS) challenge dataset, with data augmentation following the MeshDeformNet procedure. CT and MR models were trained separately with modality-appropriate preprocessing and normalization.

Training data are **not** included in this record. MMWHS must be obtained separately from the dataset providers.

Software requirements

Use these weights with the LinFlo-Net package:

- PyPI: https://pypi.org/project/linflonet/
- Source code: https://github.com/ArjunNarayanan/LinFlo-Net
- Quick start: https://github.com/ArjunNarayanan/LinFlo-Net/blob/main/docs/quick_start.md

Example inference:

pip install linflonet
linflonet predict \
  --image /path/to/scan.nii.gz \
  --model /path/to/best_model.pth \
  --modality ct \
  --output /path/to/output