PET and SPECT Image Denoising With Iterative Neural Reconstruction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing noise removal techniques for PET and SPECT images, while superior in performance, require significant processing time and there is a need for further improvement in noise removal performance.
Innovation Solution
An image processing apparatus and method utilizing a feature extraction neural network and reconstruction neural network to iteratively refine tomographic images, reducing noise while optimizing performance and time efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing noise removal techniques (guided filter, deep image prior) are used, then noise removal performance is improved, but processing time increases significantly
Solution Approach 1:
The patent divides the list data into multiple frames and processes each frame independently through separate neural networks. This segmentation allows parallel processing of multiple frames, reducing overall processing time while maintaining noise removal quality for each individual frame.
Solution Approach 2:
The patent performs preliminary feature extraction on input images before reconstruction, extracting essential features in advance. This preliminary action reduces the computational burden during the actual reconstruction phase, thereby decreasing processing time while preserving noise removal effectiveness.
2Measurement precision
If more frames are processed individually, then noise removal performance is improved, but device complexity increases
Solution Approach 1:
The patent employs a universal neural network architecture that can process multiple frames with different parameters. The same network structure is reused across all frames, allowing the system to handle variable frame configurations without increasing fundamental system complexity, while still achieving improved noise removal through multi-frame processing.
Data Source
AI summary
An image processing apparatus includes a feature extraction unit, a reconstruction unit, an evaluation unit, and a control unit. The feature extraction unit inputs an input image to a feature extraction NN, and outputs an intermediate image from the feature extraction NN. The reconstruction unit inputs the intermediate image to an m-th reconstruction NN, and outputs an m-th output image from the m-th reconstruction NN. The evaluation unit obtains an evaluation value based on a sum of differences between the m-th tomographic image and the m-th output image. The control unit repeatedly performs processes of the feature extraction unit and the reconstruction unit, calculation of the evaluation value by the evaluation unit, and training of the feature extraction NN and the m-th reconstruction NN based on the evaluation value.