2D-3D Image Registration Using Contour Feature Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The accuracy of 2D-3D image registration in medical imaging, particularly during interventions, is compromised by initial differences between preoperative 3D volume images and intraoperative 2D fluoroscopy images, leading to reduced robustness and increased computing effort and radiation exposure.
Innovation Solution
A computer-implemented method that generates input data from 2D and 3D images using contour pixels and voxels, applies a trained function to identify corresponding features, and determines a transformation instruction for registering the 2D image with the 3D image, utilizing a projection direction that ensures contour voxels have a perpendicular surface normal, thereby enhancing registration accuracy and reducing computational complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple fluoroscopy images from different perspectives are recorded to improve registration robustness, then the reliability of 2D-3D registration is improved, but the radiation exposure to the patient increases
Solution Approach 1:
The patent extracts only the essential contour information from fluoroscopy images rather than using complete multi-perspective image sets. By focusing on contour pixels and their geometric relationships, the method achieves reliable registration with reduced radiation exposure compared to conventional multi-perspective approaches.
Solution Approach 2:
The patent performs preliminary extraction of contour features and computation of surface normals before the actual registration process. This preliminary action prepares optimized input data that improves registration robustness while avoiding the need for multiple fluoroscopy acquisitions, thereby reducing radiation exposure.
2Reliability
If multiple fluoroscopy images from different perspectives are recorded to improve registration robustness, then the reliability of 2D-3D registration is improved, but the computational effort increases
Solution Approach 1:
The patent extracts only contour pixels and their geometric properties from fluoroscopy images, discarding redundant internal pixel information. This extraction reduces the volume of data requiring computational processing while maintaining registration robustness through the use of geometrically significant contour features.
Solution Approach 2:
The patent segments the fluoroscopy image data to isolate only contour pixels that are relevant for registration. By dividing the complete image data into essential contour elements and non-essential interior elements, the method reduces computational effort while preserving registration reliability.
3Ease of operation
If large initial differences between preoperative 3D volume image and intraoperative 2D fluoroscopy image are present, then the ease of operation is improved (simpler setup), but the measurement precision of registration deteriorates
Solution Approach 1:
The patent changes the parameters used for registration by focusing on geometric properties of contour pixels (surface normals, curvature, orientation) rather than relying on intensity-based matching. This parameter transformation enables accurate registration even when large initial differences exist between preoperative and intraoperative images.
Solution Approach 2:
The patent transitions from 2D intensity-based matching to 3D geometric property matching by computing surface normals and curvature from contour pixels. This dimensional elevation from pixel intensity space to geometric property space improves measurement precision while maintaining ease of operation.
Data Source
AI summary
The disclosure relates to a computer-implemented method for the provision of a transformation instruction for registering a 2D image with a 3D image. The method includes: receiving the 2D image and the 3D image; generating input data based on the 2D image having contour pixels and the 3D image having contour voxels; applying a trained function to the input data for identification of contour pixels of the 2D image and contour voxels of the 3D image, wherein at least one parameter of the trained function is adjusted based on a comparison of training contour pixels with comparison contour pixels and a comparison of training contour voxels corresponding thereto with comparison contour voxels; determining the transformation instruction based on the identified contour pixels of the 2D image and the contour voxels corresponding thereto of the 3D image for registering the 2D image with the 3D image; and providing the transformation instruction.