cs.CVAug 24, 2026

VeCAS: Vessel-Focused Contrast-Free Angiogram Synthesis for Vascular Interventions

Authors: De-Xing Huang, Chen-Yu Wang, Hao Liang, Xiao-Hu Zhou, Mei-Jiang Gui, Tian-Yu Xiang, Qin-Yi Zhang, Chen Wang, +5 more

Organizations: State Key Laboratory of Multimodal Artificial Intelligence Systems, Institute of Automation, Chinese Academy of Sciences, Beijing 100190, China · School of Artificial Intelligence, University of Chinese Academy of Sciences, Beijing 100049, China · School of Advanced Interdisciplinary Sciences, University of Chinese Academy of Sciences, Beijing 101408, China · Beijing Friendship Hospital, Capital Medical University, Beijing 100050, China

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

X-ray angiography relies on iodinated contrast agents to visualize vascular structures during image-guided interventions. However, contrast administration carries risks of adverse events, motivating the development of contrast-free alternatives. Generating X-ray angiograms directly from non-contrast X-ray images offers a potential solution, but existing approaches remain limited by (i) insufficient control over vascular localization and (ii) inefficient modeling of redundant background content. To address these challenges, we propose VeCAS, a two-stage vessel-focused contrast-free angiogram synthesis framework that separates vascular structure localization from angiographic appearance synthesis. In Stage I, a discriminative model localizes vascular structures in non-contrast X-ray images, while cross-modality latent distillation transfers vessel-sensitive knowledge from X-ray angiograms during training. In Stage II, a vessel-focused inpainting model synthesizes angiographic appearance within the localized vascular regions while preserving the non-vascular background. Experiments on an in-house lower-limb vascular intervention dataset show that VeCAS outperforms the comparison methods in terms of vascular structural fidelity and image quality. Visual Turing tests and physician assessments indicate the perceptual realism of the synthesized angiograms. In addition, robotic guidewire navigation experiments in vascular phantoms show that VeCAS guidance reduces the time to target by 41.4% and the number of operation steps by 40.7% compared with non-contrast guidance. Together, these results suggest the potential of VeCAS to serve as ``meta contrast agent'' for vascular interventions.

Figures & tables

Fig. 1: Iodinated Contrast Agent vs . Meta Contrast Agent. Top: In conventional X-ray angiography, iodinated contrast agents are injected to enhance vessel visibility, but their use may introduce contrast-related risks, such as kidney injury and allergic reactions. Bottom: The proposed meta contrast agent uses a trained deep neural network to synthesize X-ray angiograms from non-contrast X-ray images, providing a computational alternative with the potential to reduce contrast administration and improve procedural efficiency.
Fig. 2: Overview of VeCAS. VeCAS performs contrast-free angiogram synthesis using a two-stage framework. Stage I: A vascular structure localization model predicts the vessel mask from a non-contrast X-ray image. Stage II: A vessel-focused inpainting model synthesizes angiographic appearance within the predicted vascular regions while preserving the non-vascular background from the input image.
Fig. 3: Stage I: Cross-Modality Latent Distillation for Vascular Structure Localization. A frozen angiogram-domain segmentation model DA(EA(⋅))\mathcal{D}_{\rm A}\left(\mathcal{E}_{\rm A}(\cdot)\right) provides vessel-sensitive multi-scale features from X-ray angiograms. The trainable non-contrast encoder EN(⋅)\mathcal{E}_{\rm N}(\cdot) and multi-scale predictor P(⋅)\mathcal{P}(\cdot) learn to predict angiogram-domain latent features from non-contrast X-ray images. The predicted features are then decoded by the frozen decoder DA(⋅)\mathcal{D}_{\rm A}(\cdot) for vessel mask prediction.
Fig. 4: Stage II: Vessel-Focused Inpainting. (a) Training: A vessel-focused denoising loss restricts the diffusion learning objective to vascular regions, encouraging the model to learn angiographic vessel appearance rather than redundant background reconstruction. (b) Inference: At each reverse diffusion step, the denoised vascular output is fused with the forward-diffused non-contrast X-ray image according to the predicted vessel mask from Stage I, thereby synthesizing contrast-enhanced appearance while preserving the non-vascular background.
MetricR 11R 22R 33R 44mean±std{\rm mean\pm std}
Sen. (%)62.0062.0076.0076.0094.0094.0078.0078.0077.50±13.1077.50\pm 13.10
Spe. (%)50.0050.0052.0052.0062.0062.0078.0078.0060.50±12.7960.50\pm 12.79
Acc. (%)56.0056.0064.0064.0078.0078.0078.0078.0069.00±10.8969.00\pm 10.89
Positives: Real X-ray angiograms.
Negatives: Synthesized X-ray angiograms.
TABLE I: Results of the Visual Turing Test. Sensitivity (Sen.), specificity (Spe.), and accuracy (Acc.) are reported for each reader.
Fig. 5: Physician Assessments of Synthesized Angiograms. Three physicians independently scored the synthesized angiograms using three criteria: vascular morphological realism (VMR), contrast opacification realism (COR), and procedural usability (PU).
MethodDSC (%) ↑\uparrowclDice (%) ↑\uparrow
Generative Models
Pix 22 Pix [ 20 ] [ CVPR ’17]9.75±0.92∗∗∗{9.75\pm 0.92}^{\ast\ast\ast}9.50±1.10∗∗∗{9.50\pm 1.10}^{\ast\ast\ast}
CycleGAN [ 21 ] [ ICCV ’17]2.30±2.47∗∗∗{2.30\pm 2.47}^{\ast\ast\ast}2.89±3.42∗∗∗{2.89\pm 3.42}^{\ast\ast\ast}
ResViT [ 27 ] [ TMI ’22]37.03±7.09∗∗∗{37.03\pm 7.09}^{\ast\ast\ast}42.69±7.93∗∗∗{42.69\pm 7.93}^{\ast\ast\ast}
BBDM [ 31 ] [ CVPR ’23]5.43±1.67∗∗∗{5.43\pm 1.67}^{\ast\ast\ast}5.38±1.86∗∗∗{5.38\pm 1.86}^{\ast\ast\ast}
ControlNet [ 33 ] [ ICCV ’23]8.93±0.64∗∗∗{8.93\pm 0.64}^{\ast\ast\ast}8.49±0.76∗∗∗{8.49\pm 0.76}^{\ast\ast\ast}
TABLE II: Quantitative comparison of different methods for Stage I. Results are reported as “ mean±std{\rm mean}\pm{\rm std} ” over five-fold cross-validation. The best and second-best results are highlighted in bold and underlined , respectively. Statistical significance against VeCAS is assessed using Wilcoxon signed-rank tests on patient-level scores ( ∗\ast : p<0.05p<0.05 , ∗∗\ast\ast : p<0.01p<0.01 , ∗∗∗\ast\ast\ast : p<0.001p<0.001 ).
MethodDSC (%) ↑\uparrowclDice (%) ↑\uparrowPSNR (dB) ↑\uparrowSSIM ↑\uparrow
Pix 22 Pix [ 20 ] [ CVPR ’17]13.11±2.26∗∗∗{13.11\pm 2.26}^{\ast\ast\ast}14.78±2.46∗∗∗{14.78\pm 2.46}^{\ast\ast\ast}36.06±1.20∗∗∗{36.06\pm 1.20}^{\ast\ast\ast}0.9355±0.0173∗∗∗{0.9355\pm 0.0173}^{\ast\ast\ast}
CycleGAN [ 21 ] [ ICCV ’17]26.41±4.07∗∗∗{26.41\pm 4.07}^{\ast\ast\ast}31.93±5.05∗∗∗{31.93\pm 5.05}^{\ast\ast\ast}35.71±0.74∗∗∗{35.71\pm 0.74}^{\ast\ast\ast}0.9498±0.0075‾∗∗∗\underline{0.9498\pm 0.0075}^{\ast\ast\ast}
CUT [ 25 ] [ ECCV ’20]26.72±2.96∗∗∗{26.72\pm 2.96}^{\ast\ast\ast}32.13±3.71∗∗∗{32.13\pm 3.71}^{\ast\ast\ast}35.34±0.51∗∗∗{35.34\pm 0.51}^{\ast\ast\ast}0.9015±0.0574∗∗∗{0.9015\pm 0.0574}^{\ast\ast\ast}
RegGAN [ 26 ] [ NeurIPS ’21]35.60±4.75∗∗∗{35.60\pm 4.75}^{\ast\ast\ast}41.78±6.33‾∗∗∗\underline{41.78\pm 6.33}^{\ast\ast\ast}33.27±1.16∗∗∗{33.27\pm 1.16}^{\ast\ast\ast}0.9430±0.0065∗∗∗{0.9430\pm 0.0065}^{\ast\ast\ast}
ResViT [ 27 ] [ TMI ’22]31.06±3.96∗∗∗{31.06\pm 3.96}^{\ast\ast\ast}35.78±5.20∗∗∗{35.78\pm 5.20}^{\ast\ast\ast}33.56±1.01∗∗∗{33.56\pm 1.01}^{\ast\ast\ast}0.9400±0.0080∗∗∗{0.9400\pm 0.0080}^{\ast\ast\ast}
BBDM [ 31 ] [ CVPR ’23]30.80±3.95∗∗∗{30.80\pm 3.95}^{\ast\ast\ast}36.54±5.23∗∗∗{36.54\pm 5.23}^{\ast\ast\ast}37.67±0.55\bm{37.67\pm 0.55}0.9488±0.0201∗∗∗{0.9488\pm 0.0201}^{\ast\ast\ast}
TABLE III: Quantitative comparison of different methods for Stage I ++ II. Results are reported as “ mean±std{\rm mean}\pm{\rm std} ” over five-fold cross-validation. The best and second-best results are highlighted in bold and underlined , respectively. Statistical significance against VeCAS is assessed using Wilcoxon signed-rank tests on patient-level scores ( ∗\ast : p<0.05p<0.05 , ∗∗\ast\ast : p<0.01p<0.01 , ∗∗∗\ast\ast\ast : p<0.001p<0.001 ).
Lseg\mathcal{L}_{\rm seg}Lpred\mathcal{L}_{\rm pred}DSC (%) ↑\uparrowclDice (%) ↑\uparrow
✗✓29.49±3.5329.49\pm 3.5335.52±2.7435.52\pm 2.74
✓✗51.49±3.47‾\underline{51.49\pm 3.47}59.66±4.63‾\underline{59.66\pm 4.63}
✓✓52.56±3.32\bm{52.56\pm 3.32}60.62±4.42\bm{60.62\pm 4.42}
TABLE IV: Ablation study of the distillation loss Ldistill\mathcal{L}_{\rm distill} in Stage I. The best and second-best results are highlighted in bold and underlined , respectively.
Fig. 6: Qualitative Results of the Complete Pipeline (Stage I ++ II). Compared with baseline methods, VeCAS synthesizes clearer and more continuous vascular structures while preserving the non-vascular background. Most baseline methods retain the background appearance but produce weak, incomplete, or barely visible vessels.
Fig. 7: Error Propagation Analysis under Different Mask Conditions. Stage II is evaluated using predicted masks from VeCAS, perturbed ground-truth masks, and oracle ground-truth masks to analyze how mask quality affects performance.
Vessel-focusedDSC (%) ↑\uparrowclDice (%) ↑\uparrowPSNR (dB) ↑\uparrowSSIM ↑\uparrow
LossSampling
✗✗11.43±3.7611.43\pm 3.7612.55±4.2312.55\pm 4.2329.69±0.3429.69\pm 0.340.9257±0.01520.9257\pm 0.0152
✗✓11.38±3.0311.38\pm 3.0311.89±3.5611.89\pm 3.5636.19±0.93‾\underline{36.19\pm 0.93}0.9508±0.0178‾\underline{0.9508\pm 0.0178}
✓✗20.51±4.32‾\underline{20.51\pm 4.32}22.96±5.08‾\underline{22.96\pm 5.08}31.66±0.5831.66\pm 0.580.9322±0.01690.9322\pm 0.0169
✓✓42.27±4.99\bm{42.27\pm 4.99}47.70±6.29\bm{47.70\pm 6.29}36.78±0.91\bm{36.78\pm 0.91}0.9537±0.0183\bm{0.9537\pm 0.0183}
TABLE V: Ablation study of the vessel-focused design in Stage II. The best and second-best results are highlighted in bold and underlined , respectively.
Fig. 8: Phantom Experiment. (a) Phantom experimental setup. (b) Navigation efficiency under different guidance conditions, evaluated by time to target and the number of operation steps. ❶ non-contrast guidance, ❷ VeCAS guidance, and ❸ contrast guidance.

Explore similar work

CardsList
Oct 8, 2026

Temporal Diffusion Bridge Model for Vessel Extraction Using Physics-informed Synthetic Data

Accurate artery extraction from X-ray coronary angiography (XCA) is essential for coronary artery disease diagnosis and intervention. However, low vessel visibility, respiratory and cardiac motion, and limited annotations challenge existing methods. Moreover, most deep learning approaches produce binary vessel masks without preserving clinically relevant intra-arterial contrast information. We propose a Temporal Diffusion Bridge Model (TDBM) that formulates vessel extraction as image translation and learns from physics-informed synthetic data without manual vessel annotations. A physics-based XCA synthesis engine generates diverse, perfectly aligned pairs of contrast-enhanced and contrast-free images. Using these pairs, TDBM learns cross-domain mappings to recover high-fidelity vessel information. Temporal conditioning incorporates adjacent frames to improve consistency, while a vessel-aware contrastive loss exploits background-invariant synthetic pairs to enhance vascular discrimination. We evaluate TDBM on two XCA vessel segmentation datasets and conduct a reader study using two real-patient XCA sequence datasets. Zero-shot segmentation is further assessed on retinal imaging and fluorescence microscopy datasets. Results demonstrate that TDBM consistently outperforms existing methods and generalizes effectively across imaging modalities. Code and trained weights will be available at: https://github.com/Advanced-AI-in-Medicine-and-Physics-Lab/TDBM

Jinkui Hao, Donald R. Cantrell, Ramez Abdalla +2

Department of Radiology, Northwestern University, Chicago, IL, USA

Coronary Artery SegmentationCoronary AngiographyDiffusion ModelsCross-Modality Medical Image SegmentationPhysics-Informed Generative Modeling
AskPDF
Aug 31, 2026

Contrast-Free Autonomous Navigation of Untethered Endovascular Microrobots Using Single-Plane Fluoroscopy

Reliable three-dimensional (3D) navigation of magnetically actuated untethered microrobots remains a major barrier to clinical translation. X-ray fluoroscopy is the standard real-time imaging modality for endovascular procedures, but single-plane fluoroscopy provides only a two-dimensional (2D) projection, eliminating depth information and complicating autonomous navigation. Recovering this information through biplane imaging or repeated contrast-enhanced angiography increases procedural complexity, radiation exposure, or contrast burden. Here, we introduce VISTA (Virtual Integration for Spatial Tracking and Autonomy), a digital twin framework enabling contrast-free autonomous navigation under single-plane fluoroscopy. VISTA reconstructs vascular anatomy as a 3D digital twin, discretizes the vessel centerline into navigation milestones, and assigns the detected 2D robot position to the nearest projected milestone. Consecutive milestones define the local vessel orientation used to generate magnetic actuation commands, converting single-plane fluoroscopic observations into topology-constrained navigation states without requiring contrast injection during navigation. VISTA is demonstrated across anatomically distinct vascular phantoms under continuous flow and within the inferior vena cava of a live rat in vivo. Compared with conventional fluoroscopic human-in-the-loop control, VISTA reduced navigation time by up to 62%, corrective actuation commands by up to 98%, and radiation exposure by up to 57%. These results establish VISTA as a digital twin-guided framework for contrast-free autonomous navigation of untethered endovascular microrobots using widely available single-plane fluoroscopy.

Husnu Halid Alabay, Tuan-Anh Le, Ping Wang +1

Department of Physiology and Biomedical Engineering, Mayo Clinic, Scottsdale, Arizona, USA · School of Biological and Health Systems Engineering, Arizona State University, Tempe, AZ, USA

Robotic Endovascular InterventionDigital TwinsVision-Based Robot Control
AskPDF
Aug 25, 2026

MoE-based Feature Adapter for Prompt-free Binary Coronary Artery Segmentation in X-ray Angiography

Accurate segmentation of coronary arteries in X-ray angiography videos is essential for quantitative coronary analysis and image-guided interventions. However, accurate segmentation remains challenging because coronary vessels are thin and exhibit low contrast, while the presence of catheters, guidewires, and complex anatomical background structures can further interfere with vessel delineation. Existing U-Net- and Transformer-based models provide strong baselines, but their shared feature-adaptation pathways may be insufficient for heterogeneous angiographic appearances. In this paper, we propose a prompt-free mixture-of-experts (MoE) feature adapter for binary coronary artery segmentation. Built upon parameter-efficient Vision Transformer adapters, the proposed method uses multiple lightweight experts with input-dependent top-kk routing to adaptively refine vessel-related features while limiting active computational cost. Experiments on MOSXAV and external evaluation on XACV show that the proposed method outperforms representative baselines and improves cross-dataset generalisation. These results suggest that MoE-based adapter learning is effective for robust coronary artery segmentation in X-ray angiography videos.

Lin Xi, Yingliang Ma

University College London, United Kingdom · University of East Anglia, United Kingdom · King’s College London, United Kingdom

Coronary Artery SegmentationCoronary AngiographyMixture of ExpertsVision Foundation Model Adaptation
AskPDF