PET Image Reconstruction With Adaptive Coordinate Descent
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing PET image reconstruction methods, such as ML-EM and OSEM, lack control over convergence speed and are inconvenient for computer implementation, leading to image artifacts and distortion.
Innovation Solution
A discrete-to-discrete data model with an iterative coordinate descent strategy and Tikhonov-type regularization, allowing for controlled convergence and reduced image artifacts through the use of a coefficient ρ and parameter β in the iterative update formula.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the standard ML-EM algorithm is used for image reconstruction, then statistical accuracy of measurements is improved, but convergence speed cannot be controlled and computation time increases
Solution Approach 1:
The patent applies dynamic regularization where the regularization parameter β is not fixed but adapts during iterations. The algorithm dynamically adjusts β based on convergence behavior, allowing early stopping when sufficient accuracy is achieved, thus controlling computation time while maintaining statistical accuracy through the adaptive regularization strategy
Solution Approach 2:
The patent changes the regularization parameter β from a static value to a dynamic one that evolves during the iterative process. By modifying β at each iteration based on the current state of the reconstruction, the algorithm achieves both statistical accuracy and controlled convergence speed, resolving the contradiction between measurement precision and computation time
2Reliability
If the standard ML-EM algorithm is used for image reconstruction, then statistical modeling of Poisson distribution is improved, but image artifacts and distortion increase due to lack of regularization control
Solution Approach 1:
The patent implements feedback through the adaptive regularization mechanism where the algorithm monitors reconstruction quality and adjusts β accordingly. This feedback loop allows the system to maintain statistical modeling accuracy while suppressing image artifacts by increasing regularization when artifacts appear and reducing it when the image quality improves
Solution Approach 2:
The patent applies preliminary anti-action by preemptively applying regularization to prevent image artifacts before they fully develop. The adaptive β parameter is adjusted in advance to counteract the tendency of iterative algorithms to produce artifacts, thereby maintaining both statistical accuracy and image quality throughout the reconstruction process
3Manufacturing precision
If the regularization form from Eq. (2) is used, then image quality is improved, but computer implementation becomes inconvenient due to complex calculations
Solution Approach 1:
The patent simplifies the regularization implementation by changing the approach from complex analytical derivatives to a more straightforward iterative update formula. The adaptive β parameter is incorporated directly into the update equation, making the algorithm easier to implement while maintaining image quality through the same regularization effect
Data Source
AI summary
An iterative statistical algorithm based on a discrete-to-discrete data model with iterative coordinate descent optimization for image reconstruction from radiation measurements obtained in emission tomography, i.e. in a Positron Emission Tomography scanner, is described in this invention. The method presented here allows for the presence of the regularization as an additive term. Furthermore, this method makes it possible to control the convergence of the algorithm using a specific parameter. These improvements are due to the signals obtained regarding the given statistics of this imaging technique.