Epipolar Consistency for CT Image Value Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional inpainting methods and artificial intelligence-based approaches often fail to maintain epipolar consistency in estimating image values for marked pixels, leading to low-quality image reconstruction and artifacts in three-dimensional image datasets, especially in the presence of metal artifacts or defective detectors.
Innovation Solution
The method incorporates epipolar consistency conditions into the determination of image values by formulating a linear equation system based on Radon transform and derivation, ensuring consistency across projection images and using relevant conditions to improve estimation quality, which can be integrated with existing inpainting methods or artificial intelligence algorithms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional interpolation methods are used to estimate image values in marked pixels, then the process is simple and fast, but the precision and image quality are degraded
Solution Approach 1:
The patent introduces epipolar consistency conditions as an intermediary constraint between the projection images. These conditions act as a mediator that enforces geometric consistency when estimating image values in marked pixels, bridging the gap between simple interpolation and accurate reconstruction without requiring complex iterative optimization
Solution Approach 2:
The patent transforms the image estimation problem by changing the parameter space - instead of directly interpolating pixel values, it formulates a linear equation system based on epipolar geometry parameters. This parameter transformation enables accurate estimation while maintaining computational efficiency through direct solution methods
2Measurement precision
If normalization methods (NMAR) are used to improve image quality, then metal artifacts are reduced, but the method requires laterally untruncated scans which limits applicability
Solution Approach 1:
The patent extracts and utilizes only the essential epipolar consistency constraints from the full normalization process. By taking out the core geometric consistency requirements and formulating them as a linear system, the method achieves artifact reduction without requiring the complete normalization procedure that demands untruncated scans
Solution Approach 2:
The patent applies epipolar consistency conditions locally at each marked pixel location rather than requiring global normalization. This localized approach allows the method to be applied to truncated scans where only local geometric consistency can be enforced, significantly improving adaptability while maintaining image quality
3Measurement precision
If artificial intelligence methods (CNN) are used to estimate missing image values, then estimation quality improves, but computational complexity and processing time increase
Solution Approach 1:
The patent replaces the complex mechanical learning process of CNNs with a deterministic mathematical system based on epipolar geometry. Instead of using neural networks with multiple layers and parameters that require extensive training, the invention uses a linear equation system that can be solved directly, substituting AI complexity with elegant mathematical formulation
4Measurement precision
If multiple epipolar consistency conditions are enforced to ensure accuracy, then image reconstruction quality improves, but the linear equation system becomes more complex
Solution Approach 1:
The patent segments the epipolar consistency conditions into distinct linear equations, each corresponding to specific geometric relationships between projection images. By dividing the complex consistency requirements into manageable linear components, the system can enforce multiple conditions simultaneously while maintaining computational tractability through standard linear algebra techniques
Data Source
AI summary
A method for determining image values in marked pixels of at least one projection image is provided. The at least one projection image is part of a projection image set provided for reconstruction of a three-dimensional image dataset and acquired in each case using a projection geometry in an acquisition procedure. The image values are determined through evaluation of at least one epipolar consistency condition that is to be at least approximately fulfilled, that results from the projection geometries of the different projection images of the projection image set, and that requires the agreement of two transformation values in transformation images determined from different projection images by Radon transform and subsequent derivation as a condition transformation.