GPU Parallel PET Image Reconstruction via Particle Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for reconstructing Positron Emission Tomography (PET) images are inefficient due to their reliance on serial computing, leading to long reconstruction times and low-quality images, especially when using traditional CPU computing or iterative methods that fail to incorporate physiological and structural information effectively.
Innovation Solution
A method utilizing GPU parallel computing to reconstruct PET images through particle filtering, which models noise as a Poisson distribution and employs a linear sampling method and estimation algorithm to calculate intensity and weight values, significantly reducing computational time while maintaining image quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional CPU serial computing is used for PET image reconstruction, then the reconstruction algorithm can be implemented, but the reconstruction time is very long and unsuitable for clinical usage
Solution Approach 1:
The patent replaces the traditional CPU serial computing system with a GPU parallel computing system. This substitution leverages the massively parallel architecture of GPUs to perform simultaneous computations across multiple data elements, transforming the reconstruction process from sequential to parallel execution, thereby achieving significant speedup in reconstruction time while maintaining clinical applicability
Solution Approach 2:
The patent divides the PET image reconstruction task into multiple independent computational segments that can be processed in parallel. By segmenting the large-scale matrix operations and iterative calculations into smaller, concurrently executable units, the system充分利用 GPU's parallel processing capabilities to reduce overall reconstruction time without sacrificing image quality
2Measurement precision
If iterative methods are used for PET image reconstruction, then statistical information can be incorporated, but the reconstruction time is long and image quality is limited due to inability to use physiological and structural information
Solution Approach 1:
The patent incorporates physiological and structural information as prior knowledge before the reconstruction process begins. By pre-integrating this auxiliary information into the reconstruction algorithm, the system guides the iterative process more effectively, reducing the number of iterations needed to achieve high-quality images and thereby shortening reconstruction time
Solution Approach 2:
The patent replaces traditional iterative reconstruction methods with a GPU-accelerated parallel computing approach. This substitution maintains the ability to incorporate statistical information while dramatically reducing computation time through parallel processing, enabling both high image quality and clinical feasibility
3Productivity
If FBP method is used for PET image reconstruction, then the reconstruction process is fast, but the image quality is low due to noise and inability to suppress noise
Solution Approach 1:
The patent merges the speed advantage of FBP method with the image quality benefits of iterative methods. By combining FBP as an initialization step with subsequent iterative refinement using GPU parallel computing, the system achieves both fast reconstruction and high image quality with noise suppression capabilities
Solution Approach 2:
The patent replaces the noise-prone FBP method with a GPU-accelerated iterative reconstruction approach. This substitution maintains computational efficiency through parallel processing while adding noise suppression capabilities and the ability to incorporate prior information, thereby achieving both speed and image quality
Data Source
AI summary
A method for reconstructing a PET image using GPU parallel computing including the following steps: 1) sampling particles of each voxel and obtaining an intensity value of each particle of the voxel; 2) calculating a prior intensity value of the voxel, and calculating a weight value of each particle of the voxel corresponding to the intensity value of each particle; 3) resampling the intensity value and the weight value of the particle, and obtaining the resampled intensity value of the particle and the resampled weight value of the particle; 4) repeating step 2) and step 3) until the resampled intensity value of the particle is converged to a certain value; and 5) calculating the voxel value according to a true intensity value of the particle and a true weight value of the particle.