VesselFusion: Diffusion Models for Vessel Centerline Extraction from 3D CT Images

Soichi Mita; Shumpei Takezaki; Ryoma Bise

cs.CVMar 9, 2026

VesselFusion: Diffusion Models for Vessel Centerline Extraction from 3D CT Images

Authors: Soichi Mita, Shumpei Takezaki, Ryoma Bise

Organizations: Kyushu University, Fukuoka, Japan · German Research Center for Artificial Intelligence (DFKI), Kaiserslautern, Germany

Abstract

Vessel centerline extraction from 3D CT images is an important task because it reduces annotation effort to build a model that estimates a vessel structure. It is challenging to estimate natural vessel structures since conventional approaches are deterministic models, which cannot capture a complex human structure. In this study, we propose VesselFusion, which is a diffusion model to extract the vessel centerline from 3D CT image. The proposed method uses a coarse-to-fine representation of the centerline and a voting-based aggregation for a natural and stable extraction. VesselFusion was evaluated on a publicly available CT image dataset and achieved higher extraction accuracy and a more natural result than conventional approaches.

Figures & tables

Figure 1: Overview of VesselFusion : A conditional diffusion model extracts a vessel centerline from a 3D CT image.

Figure 2: Stochastic generation from Gaussian distribution to a vessel distribution. VesselFusion infrequently generates unnatural structure (e.g., tears, loops, noise).

Figure 3: (a) Representation of centerline coordinates by a coarse-to-fine style. (b) Voting and aggregation of some estimations from different initial noises.

Method	$R=1$			$R=2$			$R=3$			Betti
Method	Prec. $\uparrow$	Rec. $\uparrow$	F1 $\uparrow$	Prec. $\uparrow$	Rec. $\uparrow$	F1 $\uparrow$	Prec. $\uparrow$	Rec. $\uparrow$	F1 $\uparrow$	Betti-0 $\downarrow$	Betti-1 $\downarrow$
Baseline	0.838	0.651	0.733	0.851	0.789	0.819	0.856	0.829	0.842	190.66	0.04
VesselFormer	0.651	0.616	0.633	0.739	0.806	0.771	0.776	0.879	0.824	68.29	4.51
VesselFusion	0.855	0.679	0.757	0.880	0.786	0.830	0.887	0.830	0.858	67.66	0.09

Table 1: Comparison of vessel extraction accuracy (Prec., Rec., F1 for different radii

R

) and structural consistency (Betti numbers).

C2F	Vote	$R=1$			$R=2$			$R=3$			Betti
C2F	Vote	Prec. $\uparrow$	Rec. $\uparrow$	F1 $\uparrow$	Prec. $\uparrow$	Rec. $\uparrow$	F1 $\uparrow$	Prec. $\uparrow$	Rec. $\uparrow$	F1 $\uparrow$	Betti-0 $\downarrow$	Betti-1 $\downarrow$
		$0.459\,406$	$0.299\,848$	$0.362\,861$	$0.624\,654$	$0.605\,313$	$0.614\,832$	$0.702\,071$	$0.781\,351$	$0.739\,592$	509.35	2.53
✓		$0.638\,089$	$0.612\,302$	$0.624\,930$	$0.693\,412$	0.806	$0.745\,560$	$0.717\,571$	0.867	$0.785\,346$	115.06	3.16
	✓	$0.679\,408$	$0.358\,461$	$0.469\,310$	$0.842\,009$	$0.582\,883$	$0.688\,884$	0.893	$0.715\,583$	$0.794\,350$	105.65	0.07
✓	✓	0.855	0.679	0.757	0.880	$0.785\,814$	0.830	$0.887\,398$	$0.829\,663$	0.858	67.66	0.09

Table 2: Ablation study on the effects of coarse-to-fine representation (C2F) and voting-based aggregation. Higher values (

\uparrow

) indicate better accuracy, while lower values (

\downarrow

) indicate better structural consistency.

Figure 4: Qualitative comparison of vessel centerline extraction: (a) Baseline U-Net, (b) VesselFormer, (c) VesselFusion. Green : True positive (correctly extracted vessels). Red : False positive. Gray : Ground truth. Blue : Areas that were missed by the comparison method but were correctly detected by VesselFusion.

Figure 5: The effect of the number of samples $K$ in VesselFusion. The horizontal axis shows the number of samples $K$ used for voting, the left vertical axis shows the F1 score at $R=1$ (the higher the better), and the right vertical axis shows Betti-0 (the lower the better).

3D Medical Imaging

3D medical imaging develops methods to analyze, reconstruct, and visualize volumetric scans such as CT, MRI, PET, and ultrasound.

See all 173 papers

vesselFM-CT: Segmenting All Blood Vessels in CT Images for System-Level Cardiovascular Analysis
Jun 8, 2026Bastian Wittmann, Chinmay Prabhakar, Suprosanna Shit +1Medical Image Analysis Medical Image Segmentation
Department of Quantitative Biomedicine, University of Zurich, Zurich, Switzerland
VascuSR: Iterative Anatomy-Aware Diffusion for CT Super-Resolution
Oct 14, 2026Hairong Ni, Jiahe Huang, Minghao Guo +2Medical Image Super-Resolution 3D Medical Imaging
University of Michigan · University of California, San Diego · MIT CSAIL
Trexplorer Super: Topologically Correct Centerline Tree Tracking of Tubular Objects in CT Volumes
Jul 15, 2025Roman Naeem, David Hagerman, Jennifer Alvén +2Medical Image Analysis 3D Medical Imaging
Chalmers University of Technology, 412 96 Gothenburg, Sweden

Jun 8, 2026J/K move · Enter open · S save

vesselFM-CT: Segmenting All Blood Vessels in CT Images for System-Level Cardiovascular Analysis

Bastian Wittmann, Chinmay Prabhakar, Suprosanna Shit, Bjoern Menze

Department of Quantitative Biomedicine, University of Zurich, Zurich, Switzerland

Medical Image Analysis Medical Image Segmentation 3D Medical Image Segmentation Computed Tomography

Open paper AskPDF arXiv

The vascular network in the human body is characterized by blood vessels exhibiting drastic structural variations in radius, length, topological properties, and branching patterns. This heterogeneity, together with location-specific anatomical background variations, poses a significant challenge for robust, large-scale analysis of the entire cardiovascular system. As a result, most research has focused on narrow, isolated segments of the vascular network. While such targeted studies provide valuable insights, they inherently limit the ability to assess the systemic health and functional integrity of the vascular network as a whole. In this work, we aim to bridge this gap to advance both clinical diagnostics and our fundamental understanding of vascular physiology. We propose the task of segmenting all vessels in CT images, ranging from the largest components of the cardiovascular system to even minuscule mesenteric vessels. To this end, we introduce vesselFM-CT, the first model capable of robustly segmenting all blood vessels in 3D CT images. VesselFM-CT is trained via an iterative, multi-step process and optimizes our proposed TubeLoss loss function, effectively addressing the inherent heterogeneity of the cardiovascular system. We demonstrate that vesselFM-CT outperforms all baselines and enables automated, precise extraction of the cardiovascular system from CT images, thereby unlocking a wide range of clinical and technical perspectives, including automated disease classification and synthetic CT image generation.

Abstract

3D Medical Imaging

Explore similar work

vesselFM-CT: Segmenting All Blood Vessels in CT Images for System-Level Cardiovascular Analysis

VascuSR: Iterative Anatomy-Aware Diffusion for CT Super-Resolution

Trexplorer Super: Topologically Correct Centerline Tree Tracking of Tubular Objects in CT Volumes