Denoise node positions
A coordinate diffusion model samples spatial locations for graph vertices in 3D.
MICCAI 2024
3D Vessel Graph Generation
3D Vessel Graph Generation Using Denoising Diffusion
Figure preview is unavailable.View the original figure ↗
Two-stage diffusion pipeline, from the official code repository. · Original figure ↗
Overview
Why this work matters
Many existing graph generators focus on tree-like structures, yet real vascular networks may contain cycles. This work introduces a two-stage denoising diffusion approach that first generates 3D node coordinates and then samples edge connectivity to build diverse, anatomically plausible vessel graphs.
Approach
How the method works
Generate connectivity
A subsequent discrete diffusion model predicts graph edges while keeping the generated node coordinates fixed.
Evaluate generated graphs
Compare generated vessel networks with real-world vascular graphs, including their geometry and connectivity.
Research directions
Applications and evaluation
Microscopic capillaries
Model vascular networks in which cycles are a natural topological feature.
Cerebral vasculature
Generate major-vessel structures including the Circle of Willis.
Synthetic graph modeling
Study diverse, previously unseen vessel graph samples and their structural statistics.
Results
Visual comparison
Figure preview is unavailable.View the original figure ↗
Graph-property comparisons from the official repository. · Original image ↗
Resources
Paper and code
The official repository contains code for coordinate and edge diffusion, instructions for preparing the vascular datasets, and tools for graph visualization and statistical comparison.
Reference
Citation
@inproceedings{prabhakar2024vesseldiffusion,
title = {3D Vessel Graph Generation Using Denoising Diffusion},
author = {Prabhakar, Chinmay and Shit, Suprosanna and Musio, Fabio and Yang, Kaiyuan and Amiranashvili, Tamaz and Paetzold, Johannes C. and Li, Hongwei Bran and Menze, Bjoern},
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
pages = {3--13},
year = {2024},
publisher = {Springer},
doi = {10.1007/978-3-031-72120-5_1}
}