VasTSD: Learning 3D Vascular Tree-state Space Diffusion Model for Angiography Synthesis

TL;DR

VasTSD tackles the challenge of synthesizing angiography from non-angiographic inputs to reduce iodinated contrast exposure. It marries a pre-trained vision embedder with a 3D diffusion process guided by a dynamic vascular tree-state space, enabling anatomically continuous 3D vasculature across multiple modalities. The approach introduces a vascular tree-scanning algorithm and cross-slice attention to preserve geometric fidelity, achieving state-of-the-art results on brain and lung vasculature datasets with improved connectivity and vessel continuity. This work has potential clinical impact by enabling safer, multi-modal angiography synthesis and supporting diagnostic workflows, though it acknowledges anisotropy in medical data acquisition as a limitation for future work.

Abstract

Angiography imaging is a medical imaging technique that enhances the visibility of blood vessels within the body by using contrast agents. Angiographic images can effectively assist in the diagnosis of vascular diseases. However, contrast agents may bring extra radiation exposure which is harmful to patients with health risks. To mitigate these concerns, in this paper, we aim to automatically generate angiography from non-angiographic inputs, by leveraging and enhancing the inherent physical properties of vascular structures. Previous methods relying on 2D slice-based angiography synthesis struggle with maintaining continuity in 3D vascular structures and exhibit limited effectiveness across different imaging modalities. We propose VasTSD, a 3D vascular tree-state space diffusion model to synthesize angiography from 3D non-angiographic volumes, with a novel state space serialization approach that dynamically constructs vascular tree topologies, integrating these with a diffusion-based generative model to ensure the generation of anatomically continuous vasculature in 3D volumes. A pre-trained vision embedder is employed to construct vascular state space representations, enabling consistent modeling of vascular structures across multiple modalities. Extensive experiments on various angiographic datasets demonstrate the superiority of VasTSD over prior works, achieving enhanced continuity of blood vessels in synthesized angiographic synthesis for multiple modalities and anatomical regions.

Figures (7)

• Figure 1: (a) In 3D medical volumes, vascular have relatively consistent contextual semantics and structural connectivity. (b) The proposed tree-state space model for simulating 3D vasculature.
• Figure 2: The overall framework of VasTSD. VasTSD contains a pre-trained vision embedder and a 3D vascular tree-state space diffusion module. The vision embedder encodes 3D medical data and generates embeddings for the diffusion process. The 3D vascular state space diffusion module consists of a forward diffusion and a 3D denoising process based on tree-state space.
• Figure 3: Vascular Tree-state space architecture. A dynamic tree is constructed from vessel features and spatial relationships, with embeddings guided by prior slice information.
• Figure 4: Comparison of 2D angiographic slice modality transformation and ours on the ITKTubeTK dataset. From left to right: MRA results corresponding to T1-Flash, T1-MPRAG, and T2, with comparisons between the cGAN dar2019image, SynDiff ozbey2023unsupervised, and our method.
• Figure 5: Synthesized CTA from the pulmonary artery (non-angiographic) CTs, by cGAN dar2019image, SynDiff ozbey2023unsupervised, DiffMa wang2024soft, and our method. From top to bottom, visualized from three different anatomical coordinate systems of the body.