Ordered-Subset Image Reconstruction via Denoised Algebraic Updates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current medical image reconstruction techniques face a bottleneck due to the tradeoff between convergence speed and parallelizability, resulting in slow reconstruction speeds that hinder clinical utility, especially in low-dose CT imaging where high-quality images with low noise levels are desired.
Innovation Solution
The method decomposes the image reconstruction process into iterative steps with a statistically-penalized algebraic reconstruction update sequence and denoising steps, using a noise-weighted first derivative of the cost function to update the image estimate, allowing for high convergence speed and parallelizability, transforming nonlinear optimization problems into separable convex sub-problems that can be solved efficiently using parallelizable architectures.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If statistical imaging reconstruction with highly nonlinear regularization is used to reduce noise and improve image quality, then image quality and signal-to-noise ratio are improved, but reconstruction speed deteriorates to the order of hours
Solution Approach 1:
The patent segments the reconstruction process into ordered subsets, where each subset processes a portion of the projection data independently. This allows parallel computation across multiple processors while maintaining convergence to the optimal solution, thereby improving reconstruction speed without sacrificing image quality.
Solution Approach 2:
The patent introduces an intermediary denoising step that operates on the reconstructed image from the algebraic reconstruction technique. This denoising operation serves as a mediator that enhances image quality by reducing noise while preserving the computational efficiency of the segmented approach.
2Productivity
If optimization techniques with high convergence speed are used, then reconstruction speed is improved, but parallelizability deteriorates
Solution Approach 1:
The patent divides the projection data into ordered subsets that can be processed independently and in parallel. Each subset generates an update to the reconstruction, and these updates are combined to converge to the final solution. This segmentation enables high parallelizability while maintaining convergence speed through the mathematical properties of the ordered subset approach.
Data Source
AI summary
Described here are systems and methods for iteratively reconstructing images from data acquired using a medical imaging system. The image reconstruction is decomposed into separate linear sub-problems that can be more efficiently solved. A statistical image reconstruction process is decomposed into a statistically-weighted algebraic reconstruction update sequence. After this step, the reconstructed image is denoised using a regularization function.