Iterative CT Image Reconstruction via Voxel Block Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current x-ray computed tomography (CT) image reconstruction methods, such as simple back-projection and iterative algorithms, fail to produce satisfactory results with incomplete data and do not fully account for statistical properties, leading to suboptimal image quality and increased computational intensity.
Innovation Solution
An iterative reconstruction method that divides the initial image dataset into blocks of voxels, updates voxel values by minimizing a weighted sum of a geometry term and a regularization term, and repeats this process until all groups of voxels are updated, allowing for fast convergence and improved handling of incomplete data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If iterative image reconstruction methods are used to handle incomplete data and account for statistical properties, then image quality is improved, but computation time and computational intensity increase significantly
Solution Approach 1:
The image dataset is divided into multiple blocks of voxels, and each block is further divided into groups where voxels share the same position relative to their block. This segmentation allows the computational workload to be distributed and processed in parallel, significantly reducing computation time while maintaining the iterative reconstruction quality for handling incomplete data
Solution Approach 2:
Instead of updating all voxels simultaneously in each iteration, the method updates only one group of voxels per iteration step. This partial action approach reduces the computational burden per iteration while still achieving convergence through multiple iterations, balancing image quality with computational efficiency
2Speed
If FISTA algorithm is used for fast iterative reconstruction, then convergence speed is improved, but the need to solve non-linear image denoising problem with correlated voxels at each iteration increases complexity
Solution Approach 1:
By dividing voxels into independent groups based on their relative positions within blocks, the complex non-linear denoising problem with correlated voxels is decomposed into simpler, independent sub-problems that can be solved separately for each group, reducing algorithmic complexity while maintaining fast convergence
Solution Approach 2:
The voxel groups are pre-defined based on their spatial relationships within blocks before the iterative process begins. This preliminary organization allows the algorithm to efficiently access and update specific groups without complex correlation calculations during each iteration, simplifying the overall algorithm structure
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This method achieves fast and stable image reconstruction, reducing patient dose and accounting for statistical properties, resulting in improved image quality even with sparse or incomplete data.
Implementation Method 1
the object from which projection data is to be acquired is generally exposed to x-ray radiation from a number of projection directions
Data Source
AI summary
A method and calculation unit are disclosed for reconstructing an image from measured projection data. In an embodiment, the method includes reconstructing an initial image from the projection data via back-projection; dividing the initial image into contiguous blocks of voxels; defining groups of voxels, wherein each group is defined by selecting one voxel from each block, each voxel having the same position relative to the other voxels within its block; updating the voxel values for all voxels within one group by minimizing a function consisting of a weighted sum of a first term representing the geometry of the data acquisition process, and a second regularization term; and repeating the updating of the voxel values for all voxels within one group until all groups have been updated at least once.