VesselGPT: Autoregressive Modeling of Vascular Geometry

2025-05-19

AI Summary

To address the challenge of high-fidelity geometric modeling of anatomical structures such as vascular trees, this paper proposes the first autoregressive sequence-based vascular synthesis method. It unifies vascular centerlines, branching topology, and cross-sectional geometry into a discrete sequence representation; employs B-splines to parameterize cross-sections for preserving morphological details; and introduces a learnable, compact discrete vocabulary. Geometric quantization is achieved via a VQ-VAE, while GPT-2 models long-range dependencies and topological consistency. The method achieves high-fidelity reconstruction across multiple clinical vascular datasets, significantly outperforming existing generative and reconstruction approaches in both qualitative assessment and quantitative metrics—including Hausdorff distance and branch matching rate. The model is lightweight, exhibits strong generalizability across diverse vascular geometries, and will be released with open-source code, datasets, and pretrained models.

Technology Category

• Machine Learning: Deep Generative Models & Autoencoders
• Computer Vision: Diffusion Models for Vision
• Natural Language Processing: Code Generation / Program Synthesis from Natural Language

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Web Mining and Content Analysis: Web data generation and simulation

Abstract

Anatomical trees are critical for clinical diagnosis and treatment planning, yet their complex and diverse geometry make accurate representation a significant challenge. Motivated by the latest advances in large language models, we introduce an autoregressive method for synthesizing anatomical trees. Our approach first embeds vessel structures into a learned discrete vocabulary using a VQ-VAE architecture, then models their generation autoregressively with a GPT-2 model. This method effectively captures intricate geometries and branching patterns, enabling realistic vascular tree synthesis. Comprehensive qualitative and quantitative evaluations reveal that our technique achieves high-fidelity tree reconstruction with compact discrete representations. Moreover, our B-spline representation of vessel cross-sections preserves critical morphological details that are often overlooked in previous' methods parameterizations. To the best of our knowledge, this work is the first to generate blood vessels in an autoregressive manner. Code, data, and trained models will be made available.

Problem

• Accurate representation of complex vascular geometry
• Autoregressive modeling of anatomical tree synthesis
• Preservation of morphological details in vessel cross-sections

Innovation

• Autoregressive modeling with GPT-2 for vascular trees
• VQ-VAE embedding for discrete vessel structure representation
• B-spline representation preserving morphological details