Image Domain De-noising Using Huber Roughness Penalty Minimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image reconstruction algorithms in CT scanners, such as iterative reconstruction (IR) and image domain de-noising methods, face challenges in efficiently reducing noise while maintaining image quality and minimizing radiation exposure, with IR algorithms being computationally expensive and image domain algorithms providing limited results.
Innovation Solution
An image data processing component employing image domain only iterative de-noising algorithms based on Huber roughness penalty minimization, which de-noises reconstructed image data solely in the image domain, reducing computational expense and noise while preserving edges, using a processor to execute algorithms that minimize the Huber roughness penalty.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If iterative reconstruction (IR) algorithms are used to reduce noise and improve image quality, then image quality and diagnostic value are improved, but computational expense and processing time increase significantly
Solution Approach 1:
The patent segments the de-noising process into two distinct stages: (1) an initial filtered backprojection reconstruction to generate a preliminary image, and (2) a subsequent image domain de-noising step using Huber roughness penalty minimization. This segmentation allows the computationally intensive IR to be replaced with a faster image domain method while preserving the benefits of iterative reconstruction through the two-stage approach.
Solution Approach 2:
The patent extracts the de-noising function from the projection domain (where iterative reconstruction operates) and relocates it to the image domain. By applying Huber roughness penalty minimization directly to the reconstructed image data rather than to projection data, the method achieves comparable noise reduction with significantly reduced computational expense.
2Object-affected harmful factors
If iterative reconstruction (IR) algorithms are used to reduce noise, then noise reduction is achieved, but radiation dose cannot be reduced due to computational constraints
Solution Approach 1:
The patent employs a computationally inexpensive image domain de-noising method (Huber roughness penalty minimization) as a disposable alternative to the expensive iterative reconstruction process. This lighter-weight algorithm achieves sufficient noise reduction without requiring the heavy computational resources of full IR, enabling its use in resource-constrained environments.
3Productivity
If traditional image domain de-noising algorithms are used, then processing speed is improved, but noise reduction effectiveness is limited compared to IR algorithms
Solution Approach 1:
The patent changes the mathematical parameters and objective function used in image domain de-noising by implementing Huber roughness penalty minimization. This parameter change transforms a simple smoothing operation into a more sophisticated de-noising process that better preserves edges and structures while effectively reducing noise, achieving results comparable to iterative reconstruction.
4Manufacturing precision
If Huber roughness penalty minimization is applied in the projection domain (iterative reconstruction), then optimal noise reduction is achieved, but computational complexity increases due to repeated switching between image and projection domains
Solution Approach 1:
The patent extracts the Huber roughness penalty minimization operation from the iterative reconstruction loop in the projection domain and applies it independently in the image domain. This extraction eliminates the need for repeated forward and back-projections, reducing algorithmic complexity while maintaining the noise reduction quality benefits of Huber penalty minimization.
Data Source
AI summary
An image data processing component (122) includes algorithm memory (212) including one or more image domain only iterative de-noising algorithms (214) based on the Huber roughness penalty minimization and a processor (206) which de-noises reconstructed image data solely in the image domain based on at least one of the Huber roughness penalty iterative minimization algorithms.