2.5D Dictionary Learning for Low-Dose CT Image Reconstruction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current reconstruction algorithms in CT imaging face challenges in reducing artifacts such as streaks and noise, particularly in low-dose imaging contexts, where there is a trade-off between computational efficiency, dose, and image quality.
Innovation Solution
The implementation of a 2.5D dictionary learning-based iterative reconstruction method that uses multiple two-dimensional dictionaries oriented in different directions to improve image quality and reduce noise, while maintaining computational efficiency by reducing the number of patches and dictionary size compared to 3D dictionary learning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If 3D dictionary learning is used to improve image quality and reduce artifacts, then image quality improves, but computational cost and dictionary size increase significantly
Solution Approach 1:
The patent segments the 3D volume data into multiple 2D slices and processes each slice independently using 2D dictionary learning. This segmentation approach maintains the ability to reduce artifacts and improve image quality while significantly reducing computational complexity and memory requirements compared to processing the entire 3D volume as a single entity.
Solution Approach 2:
The patent transitions from 3D dictionary learning to 2D dictionary learning by changing the dimensional approach. Instead of learning dictionaries in three dimensions, the method learns 2D dictionaries for each axial slice, reducing the dictionary size from O(N³) to O(N²) per slice, thereby reducing overall computational cost while maintaining image quality through the preservation of local structures in each slice.
2Manufacturing precision
If regularization based methods are used to reduce artifacts, then image quality improves, but there is a trade-off with computational efficiency and dose
Solution Approach 1:
The patent changes the parameter of dictionary dimensionality from 3D to 2D, which fundamentally alters the computational complexity of the regularization process. This parameter change enables the method to achieve artifact reduction with lower computational cost, improving productivity while maintaining image quality. The 2D dictionaries require fewer parameters to store and compute compared to 3D dictionaries.
3Manufacturing precision
If more views are acquired to improve image quality, then image quality improves, but radiation dose increases
Solution Approach 1:
The patent enables the reconstruction algorithm to self-correct and enhance image quality through dictionary learning-based regularization. By learning representative patterns from the data itself and using these to guide reconstruction, the method can achieve high image quality even with limited projection views, thereby reducing the required radiation dose while maintaining diagnostic image quality.
Data Source
AI summary
A computationally efficient dictionary learning-based term is employed in an iterative reconstruction framework to keep more spatial information than two-dimensional dictionary learning and require less computational cost than three-dimensional dictionary learning. In one such implementation, a non-local regularization algorithm is employed in an MBIR context (such as in a low dose CT image reconstruction context) based on dictionary learning in which dictionaries from different directions (e.g., x,y-plane, y,z-plane, x,z-plane) are employed and the sparse coefficients calculated accordingly. In this manner, spatial information from all three directions is retained and computational cost is constrained.