AortaDiff: Volume-Guided Conditional Diffusion Models for Multi-Branch Aortic Surface Generation

Area of Science:

• Medical imaging and computational modeling
• Cardiovascular engineering
• Artificial intelligence in healthcare

Background:

• Accurate 3D aortic reconstruction is vital for clinical diagnosis, surgical planning, and computational fluid dynamics (CFD) simulations.
• Current methods often require large datasets and manual input, limiting geometric consistency for CFD analysis.

Purpose of the Study:

• To introduce AortaDiff, a diffusion-based framework for generating smooth, CFD-compatible 3D aortic surfaces directly from medical imaging volumes.
• To overcome limitations of existing methods regarding dataset size and manual intervention.

Main Methods:

• A volume-guided conditional diffusion model (CDM) generates aortic centerlines from CT/MRI data.
• Centerlines are used to extract vessel contours, ensuring precise boundary delineation.
• Extracted contours are fitted into smooth, continuous 3D surfaces for CFD-compatible meshing.

Main Results:

• AortaDiff successfully generates smooth, geometrically accurate 3D aorta meshes suitable for CFD analysis.
• The framework demonstrates effectiveness with limited training data, producing high-fidelity reconstructions.
• Successfully reconstructed both normal and pathological aorta cases, including aneurysms and coarctation.

Conclusions:

• AortaDiff provides an end-to-end workflow for generating CFD-compatible aorta meshes with minimal reliance on large labeled datasets.
• The method offers high geometric fidelity and adaptability to various aortic conditions.
• Positions AortaDiff as a practical tool for cardiovascular research, enhancing visualization and simulation capabilities.