Related Experiment Video
Updated: Nov 20, 2025
05:05
Four-Dimensional CT Analysis Using Sequential 3D-3D Registration
Published on: November 23, 2019
8.3K
Deep learning-based X-ray inpainting for improving spinal 2D-3D registration
Hooman Esfandiari1, Simon Weidert2, István Kövesházi2
1School of Biomedical Engineering, Surgical Technologies Lab, Centre for Hip Health and Mobility, University of British Columbia, Vancouver, British Columbia, Canada.
Summary
Deep learning inpainting removes implant projections from X-rays, significantly improving 2D-3D registration accuracy. This technique enhances the capture range for medical image registration by up to 85%.
Area of Science:
- Medical imaging
- Computer vision
- Artificial intelligence
Background:
- Two-dimensional (2D)-3D registration is crucial for image-guided surgery.
- Implant projections on intraoperative X-rays hinder registration accuracy and limit capture range.
- Novel methods are needed to overcome challenges posed by artifacts in medical imaging.
Purpose of the Study:
- To investigate the efficacy of deep-learning-based inpainting for removing implant projections from X-rays.
- To assess the impact of inpainting on the performance of 2D-3D registration.
- To determine if inpainting can improve the capture range of the registration process.
Main Methods:
- Trained deep learning models for inpainting implant projections on X-ray images.
- Utilized clinical datasets for evaluating inpainting performance using six image similarity metrics.
- Assessed the effect of inpainting on the capture range of 2D-3D registration.
Main Results:
- X-ray inpainting significantly enhanced image similarity between inpainted images and ground truth.
- Inpainting prior to 2D-3D registration led to a substantial recovery of the capture range.
- Capture range improvement reached up to 85% with the application of inpainting.
Conclusions:
- Deep-learning-based inpainting effectively removes implant artifacts from X-ray images.
- This method markedly improves the capture range for 2D-3D registration tasks.
- Inpainting offers a promising solution for enhancing medical image registration accuracy in the presence of implants.
