Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

CAS-GAN for Contrast-free Angiography Synthesis

De-Xing Huang, Xiao-Hu Zhou, Mei-Jiang Gui, Xiao-Liang Xie, Shi-Qi Liu, Shuang-Yi Wang, Hao Li, Tian-Yu Xiang, Zeng-Guang Hou

TL;DR

CAS-GAN addresses the problem of reducing iodinated contrast in X-ray angiography by learning to synthesize angiographies from non-contrast X-rays. It introduces a disentangled latent space separating background and vessel components, coupled with a predictor that maps background representations to vessel representations, and a vessel semantic-guided generator with a corresponding semantic adversarial loss to improve vascular realism. The model optimizes a composite objective including prediction, image- and latent-cycle consistency, and reconstruction losses, achieving state-of-the-art FID ($5.87$) and MMD ($0.016$ in abstract/table, 0.16 reported) on the XCAD dataset and showing positive ablation and external validation results. This approach demonstrates strong potential for clinical use by providing high-fidelity, contrast-free angiography synthesis and guiding safer interventional procedures.

Abstract

Iodinated contrast agents are widely utilized in numerous interventional procedures, yet posing substantial health risks to patients. This paper presents CAS-GAN, a novel GAN framework that serves as a "virtual contrast agent" to synthesize X-ray angiographies via disentanglement representation learning and vessel semantic guidance, thereby reducing the reliance on iodinated contrast agents during interventional procedures. Specifically, our approach disentangles X-ray angiographies into background and vessel components, leveraging medical prior knowledge. A specialized predictor then learns to map the interrelationships between these components. Additionally, a vessel semantic-guided generator and a corresponding loss function are introduced to enhance the visual fidelity of generated images. Experimental results on the XCAD dataset demonstrate the state-of-the-art performance of our CAS-GAN, achieving a FID of 5.87 and a MMD of 0.016. These promising results highlight CAS-GAN's potential for clinical applications.

CAS-GAN for Contrast-free Angiography Synthesis

TL;DR

CAS-GAN addresses the problem of reducing iodinated contrast in X-ray angiography by learning to synthesize angiographies from non-contrast X-rays. It introduces a disentangled latent space separating background and vessel components, coupled with a predictor that maps background representations to vessel representations, and a vessel semantic-guided generator with a corresponding semantic adversarial loss to improve vascular realism. The model optimizes a composite objective including prediction, image- and latent-cycle consistency, and reconstruction losses, achieving state-of-the-art FID () and MMD ( in abstract/table, 0.16 reported) on the XCAD dataset and showing positive ablation and external validation results. This approach demonstrates strong potential for clinical use by providing high-fidelity, contrast-free angiography synthesis and guiding safer interventional procedures.

Abstract

Iodinated contrast agents are widely utilized in numerous interventional procedures, yet posing substantial health risks to patients. This paper presents CAS-GAN, a novel GAN framework that serves as a "virtual contrast agent" to synthesize X-ray angiographies via disentanglement representation learning and vessel semantic guidance, thereby reducing the reliance on iodinated contrast agents during interventional procedures. Specifically, our approach disentangles X-ray angiographies into background and vessel components, leveraging medical prior knowledge. A specialized predictor then learns to map the interrelationships between these components. Additionally, a vessel semantic-guided generator and a corresponding loss function are introduced to enhance the visual fidelity of generated images. Experimental results on the XCAD dataset demonstrate the state-of-the-art performance of our CAS-GAN, achieving a FID of 5.87 and a MMD of 0.016. These promising results highlight CAS-GAN's potential for clinical applications.

Paper Structure

This paper contains 26 sections, 13 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related works
  3. Generative adversarial networks
  4. Image-to-image translation
  5. Disentanglement representation learning
  6. Methodology
  7. Disentanglement representation learning
  8. Vessel semantic-guided generator
  9. Vessel semantic-guided loss
  10. Training objective
  11. Prediction loss
  12. Adversarial loss
  13. Cycle-consistency loss
  14. Reconstruction loss
  15. Experimental setup
  16. ...and 11 more sections

Figures (4)

  • Figure 1: Illustration of two vascular visualization methods.
  • Figure 2: The overall architecture of our CAS-GAN. (a) The forward cycle (Non-contrast X-ray $\rightarrow$ X-ray angiography). (b) The backward cycle (X-ray angiography $\rightarrow$ Non-contrast X-ray). Solid black and dashed gray arrows indicate data flow and loss functions, respectively.
  • Figure 3: Qualitative comparisons of X-ray angiographies generated by different models.
  • Figure 4: Qualitative comparisons on the external dataset.