Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

UDCR: Unsupervised Aortic DSA/CTA Rigid Registration Using Deep Reinforcement Learning and Overlap Degree Calculation

Wentao Liu, Bowen Liang, Weijin Xu, Tong Tian, Qingsheng Lu, Xipeng Pan, Haoyuan Li, Siyu Tian, Huihua Yang, Ruisheng Su

TL;DR

UDCR presents an unsupervised deep reinforcement learning framework for rigidly registering aortic DSA to CTA, avoiding reliance on ground-truth annotations or DSA segmentation. It introduces an overlap-degree reward within a cross-dimensional 2D/3D DRL environment to quantify registration quality, enabling both pre-training and online learning. Evaluated on 61 paired cases, the method achieved a translation MAE of $2.85$ mm and a rotation MAE of $4.35^{\circ}$, demonstrating potential for intraoperative guidance in aortic interventions. The approach is adaptable to other modalities and does not hinge on landmarks or centerlines, supporting robust cross-modal registration in clinical workflows.

Abstract

The rigid registration of aortic Digital Subtraction Angiography (DSA) and Computed Tomography Angiography (CTA) can provide 3D anatomical details of the vasculature for the interventional surgical treatment of conditions such as aortic dissection and aortic aneurysms, holding significant value for clinical research. However, the current methods for 2D/3D image registration are dependent on manual annotations or synthetic data, as well as the extraction of landmarks, which is not suitable for cross-modal registration of aortic DSA/CTA. In this paper, we propose an unsupervised method, UDCR, for aortic DSA/CTA rigid registration based on deep reinforcement learning. Leveraging the imaging principles and characteristics of DSA and CTA, we have constructed a cross-dimensional registration environment based on spatial transformations. Specifically, we propose an overlap degree calculation reward function that measures the intensity difference between the foreground and background, aimed at assessing the accuracy of registration between segmentation maps and DSA images. This method is highly flexible, allowing for the loading of pre-trained models to perform registration directly or to seek the optimal spatial transformation parameters through online learning. We manually annotated 61 pairs of aortic DSA/CTA for algorithm evaluation. The results indicate that the proposed UDCR achieved a Mean Absolute Error (MAE) of 2.85 mm in translation and 4.35° in rotation, showing significant potential for clinical applications.

UDCR: Unsupervised Aortic DSA/CTA Rigid Registration Using Deep Reinforcement Learning and Overlap Degree Calculation

TL;DR

UDCR presents an unsupervised deep reinforcement learning framework for rigidly registering aortic DSA to CTA, avoiding reliance on ground-truth annotations or DSA segmentation. It introduces an overlap-degree reward within a cross-dimensional 2D/3D DRL environment to quantify registration quality, enabling both pre-training and online learning. Evaluated on 61 paired cases, the method achieved a translation MAE of mm and a rotation MAE of , demonstrating potential for intraoperative guidance in aortic interventions. The approach is adaptable to other modalities and does not hinge on landmarks or centerlines, supporting robust cross-modal registration in clinical workflows.

Abstract

The rigid registration of aortic Digital Subtraction Angiography (DSA) and Computed Tomography Angiography (CTA) can provide 3D anatomical details of the vasculature for the interventional surgical treatment of conditions such as aortic dissection and aortic aneurysms, holding significant value for clinical research. However, the current methods for 2D/3D image registration are dependent on manual annotations or synthetic data, as well as the extraction of landmarks, which is not suitable for cross-modal registration of aortic DSA/CTA. In this paper, we propose an unsupervised method, UDCR, for aortic DSA/CTA rigid registration based on deep reinforcement learning. Leveraging the imaging principles and characteristics of DSA and CTA, we have constructed a cross-dimensional registration environment based on spatial transformations. Specifically, we propose an overlap degree calculation reward function that measures the intensity difference between the foreground and background, aimed at assessing the accuracy of registration between segmentation maps and DSA images. This method is highly flexible, allowing for the loading of pre-trained models to perform registration directly or to seek the optimal spatial transformation parameters through online learning. We manually annotated 61 pairs of aortic DSA/CTA for algorithm evaluation. The results indicate that the proposed UDCR achieved a Mean Absolute Error (MAE) of 2.85 mm in translation and 4.35° in rotation, showing significant potential for clinical applications.
Paper Structure (12 sections, 3 equations, 2 figures, 2 tables)

This paper contains 12 sections, 3 equations, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Problem Definition
  4. Data Preprocessing
  5. DSA/CTA Registration Environment
  6. Action and Observations
  7. Reward Fuction
  8. Experiments
  9. Dataset
  10. Implementation Details
  11. Comparing Results and Discussion
  12. Conclusions

Figures (2)

  • Figure 1: Deep reinforcement learning-based rigid registration of aortic DSA/CTA
  • Figure 2: Visualization of registration results using the PPO algorithm. (PCC: PPO+Concat+CNN, PFC: PPO+Fuse+CNN, PFP: PPO+Fuse+PCM)