Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

3D/2D Registration of Angiograms using Silhouette-based Differentiable Rendering

Taewoong Lee, Sarah Frisken, Nazim Haouchine

TL;DR

This work addresses 3D/2D registration of Digital Subtraction Angiography (DSA) by casting it as a two-view pose-estimation problem using differentiable silhouette rendering. It introduces a silhouette-based differentiable rendering framework that projects a 3D vascular mesh $\mathbf{M}$ via projection $P = K[\mathbf{R}|\mathbf{t}]$ and optimizes the poses $\mathbf{P}_{AP}$ and $\mathbf{P}_{L}$ through a gradient-based loss that couples both AP and LAT images, including a $90^{\circ}$ rotation approach with Rodrigues' formula to handle lateral views. The method demonstrates improved registration accuracy over single-view cases on synthetic data and shows qualitative feasibility on real DSA images, highlighting the potential for clinical applicability in brain hemodynamics assessment. By leveraging two-view information and differentiable rendering, the approach offers robust pose estimation and precise alignment of angiographic data to a 3D vascular model, enabling better interpretation of angioarchitecture. Future enhancements aim to integrate advanced segmentation, optimize inter-view geometry, and develop adaptive rasterization to better handle vessels of varying thickness.

Abstract

We present a method for 3D/2D registration of Digital Subtraction Angiography (DSA) images to provide valuable insight into brain hemodynamics and angioarchitecture. Our approach formulates the registration as a pose estimation problem, leveraging both anteroposterior and lateral DSA views and employing differentiable rendering. Preliminary experiments on real and synthetic datasets demonstrate the effectiveness of our method, with both qualitative and quantitative evaluations highlighting its potential for clinical applications. The code is available at https://github.com/taewoonglee17/TwoViewsDSAReg.

3D/2D Registration of Angiograms using Silhouette-based Differentiable Rendering

TL;DR

This work addresses 3D/2D registration of Digital Subtraction Angiography (DSA) by casting it as a two-view pose-estimation problem using differentiable silhouette rendering. It introduces a silhouette-based differentiable rendering framework that projects a 3D vascular mesh via projection and optimizes the poses and through a gradient-based loss that couples both AP and LAT images, including a rotation approach with Rodrigues' formula to handle lateral views. The method demonstrates improved registration accuracy over single-view cases on synthetic data and shows qualitative feasibility on real DSA images, highlighting the potential for clinical applicability in brain hemodynamics assessment. By leveraging two-view information and differentiable rendering, the approach offers robust pose estimation and precise alignment of angiographic data to a 3D vascular model, enabling better interpretation of angioarchitecture. Future enhancements aim to integrate advanced segmentation, optimize inter-view geometry, and develop adaptive rasterization to better handle vessels of varying thickness.

Abstract

We present a method for 3D/2D registration of Digital Subtraction Angiography (DSA) images to provide valuable insight into brain hemodynamics and angioarchitecture. Our approach formulates the registration as a pose estimation problem, leveraging both anteroposterior and lateral DSA views and employing differentiable rendering. Preliminary experiments on real and synthetic datasets demonstrate the effectiveness of our method, with both qualitative and quantitative evaluations highlighting its potential for clinical applications. The code is available at https://github.com/taewoonglee17/TwoViewsDSAReg.
Paper Structure (4 sections, 6 equations, 2 figures, 1 table)

This paper contains 4 sections, 6 equations, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. Method
  3. Results
  4. Conclusion

Figures (2)

  • Figure 1: Overview of the proposed silhouette-based differentiable rendering method for 3D/2D angiographic registration using two views. The framework optimizes the alignment of 2D DSA images with the 3D vascular model. Click the image to play the video in a browser.
  • Figure 2: First (top) and final (bottom) iterations of the registration process using the two AP and LAT images for synthetic data. Click the image to play the video in a browser.