CT Image Reconstruction Using Differential Projection and Non-Quadratic Smoothing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current CT image reconstruction methods, such as weighted Filtered Back-Projection, suffer from cone artifacts, unequal signal-to-noise ratios, and inflexibility in simulating the scanning process, leading to suboptimal spatial resolution and image quality, while iterative statistical methods require excessive computing resources.
Innovation Solution
An iterative method for reconstructing CT image data that incorporates differential projection data and a non-quadratic correction operator to reduce noise and artifacts, using a combination of filtered back-projection and local contrast-dependent smoothing, which significantly reduces computing time compared to statistical methods.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If iterative statistical reconstruction methods are used, then image quality and spatial resolution are improved, but computing time increases significantly (approximately 100 times)
Solution Approach 1:
The patent segments the reconstruction process into two distinct parts: a filtering step using filtered back-projection and an iterative optimization step using statistical methods. This segmentation allows the computationally intensive iterative process to be applied selectively and efficiently, reducing overall computing time while maintaining image quality improvements.
Solution Approach 2:
The patent performs preliminary filtering of the projection data using filtered back-projection before applying iterative statistical reconstruction. This preliminary action prepares the data in a way that reduces the computing burden of the subsequent iterative step, achieving faster convergence and reducing total computing time by a factor of approximately 100 compared to pure iterative methods.
2Productivity
If weighted filtered back-projection is used, then computing effort is kept low, but cone artifacts appear and spatial resolution deteriorates
Solution Approach 1:
The patent implements a feedback mechanism where the filtered image is reprojected and compared with the original projection data. The difference (residual) is fed back into the iterative optimization process, allowing continuous refinement of the reconstruction. This feedback loop eliminates cone artifacts and improves spatial resolution while maintaining computational efficiency through the initial filtered back-projection step.
3Ease of operation
If all x-ray beams are incorporated with equal weight in filtered back-projection, then the algorithm is simple to implement, but beams with poor signal-to-noise ratio degrade image quality
Solution Approach 1:
The patent applies local quality weighting where each x-ray beam is weighted according to its individual signal-to-noise ratio. The weighting factor is calculated based on the attenuation properties of the object along each beam path. This allows beams with poor signal-to-noise ratio to be appropriately down-weighted, preventing them from degrading overall image quality, while maintaining algorithmic tractability through the filtered back-projection framework.
Data Source
AI summary
A method is disclosed for reconstructing CT image data. In at least one embodiment, the method includes provisioning CT projection data p. Secondly, it includes reconstruction of first image data fk=1 based on the CT projection data p. Thirdly, it includes iterative determination of k+1-th CT image data fk+1 on the basis of the first CT image data fk=1 as a function of: k-th CT image data fk, a reconstruction of differential projection data, the differential projection data being produced as the difference between reprojected CT image data fk and the CT projection data p, as well as a local contrast-dependent smoothing of the CT image data fk using a non-quadratic correction operator R(fk). Besides suppressing “cone” artifacts, the proposed method of at least one embodiment exhibits a significant reduction in image noise after just a few iterations.