Tomographic Image Reconstruction with Spatially Variant Hyper-parameters
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current tomographic image reconstruction algorithms face challenges in optimizing the penalty term parameter, which is manually selected and dependent on noise levels, leading to irregular noise levels in reconstructed images and potential image quality degradation.
Innovation Solution
A spatially variant hyper-parameter approach is introduced, where the hyper-parameter controlling the data fidelity and penalty terms is auto-estimated and tuned iteratively, using Fisher information and local impulse response functions to optimize noise reduction and ensure convergence.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a fixed penalty term parameter is used in tomographic image reconstruction, then the reconstruction process is simple and fast, but the noise level becomes irregular and image quality degrades
Solution Approach 1:
The patent applies dynamics by transforming the fixed penalty term parameter into a dynamic, spatially variant parameter that adapts to local image characteristics. The parameter β is updated iteratively based on the current image estimate and noise statistics, allowing the reconstruction process to adjust parameters in real-time rather than using a static value throughout.
Solution Approach 2:
The patent implements local quality by making the penalty term parameter spatially variant rather than uniform. Different regions of the image receive different parameter values based on local noise characteristics and image features, with the parameter β(r) varying across spatial position r to optimize reconstruction quality locally.
2Ease of operation
If manual parameter selection is used, then the system is easy to operate, but the parameter optimization is time-consuming and subjective
Solution Approach 1:
The patent applies self-service by implementing automated parameter selection through the iterative update process. The system automatically estimates noise levels and computes optimal parameter values based on the current image state and measurement statistics, eliminating the need for manual parameter tuning by operators.
Solution Approach 2:
The patent implements feedback by using the reconstructed image and measurement data to continuously update the penalty term parameter. The parameter β is refined in each iteration based on feedback from the current image estimate and noise statistics, creating a closed-loop optimization process that automatically converges to optimal values.
Data Source
AI summary
The present disclosure a system and method for generating a tomographic image of a subject. In some aspects, the method includes receiving an initial image acquired from a subject using the tomographic imaging system, and performing, using the initial image and a cost function model, a penalty calculation based on a spatially variant hyper-parameter. The method also includes generating an updated image using the penalty calculation, and generating a finalized image by iteratively updating the updated image until a stopping criterion is met.