Back Projection Voxel Computation for Parallel Image Reconstruction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The sequential image reconstruction method faces challenges in reducing processing time and improving image accuracy, particularly when dealing with large-sized images, due to difficulties in creating and efficiently computing matrix elements for forward and back projection, which often requires sacrificing parallel computation efficiency.
Innovation Solution
An information processing apparatus that performs image reconstruction using pixel values from a radiation source and detection unit, with a forward projection unit creating forward projection data and a back projection unit projecting onto smaller back projection voxels, allowing for efficient parallel computation and improved accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If matrix elements for forward projection and back projection are stored in storage device, then transposed matrix relationship is satisfied, but storage capacity requirement becomes excessively large for large-sized images
Solution Approach 1:
The patent extracts and removes the back projection matrix from storage requirements by computing it dynamically during the back projection process. Instead of storing the entire back projection matrix, the system computes matrix elements on-demand based on voxel positions and projection geometry, significantly reducing storage capacity requirements while maintaining the transposed matrix relationship for accurate image reconstruction
Solution Approach 2:
The patent implements a dynamic computation approach where the back projection matrix is not static but computed dynamically during the reconstruction process. Matrix elements are generated as needed based on current processing requirements, allowing the system to adapt to large-sized images without requiring proportional increases in storage capacity
2Productivity
If computation method is optimized for parallel computation, then processing speed improves, but transposed matrix relationship between forward and back projection is abandoned
Solution Approach 1:
The patent segments the back projection computation into independent voxel-wise operations that can be executed in parallel. Each voxel's contribution to the projection data is computed independently, allowing full parallelization across multiple processing units while maintaining the transposed matrix relationship through consistent application of the same computation formula to all voxels
Solution Approach 2:
The patent changes the computational parameters and organization of the back projection process to enable parallel execution. By reorganizing the computation to operate on voxel groups or individual voxels in parallel, and by using efficient memory access patterns, the system achieves high processing speed while preserving the mathematical relationship required for accurate reconstruction
3Productivity
If forward projection or back projection is optimized for parallel computation, then parallel computation efficiency improves, but the other computation sacrifices parallelism due to atomic operations
Solution Approach 1:
The patent segments both forward projection and back projection into independent, non-overlapping computational units that can be executed in parallel without atomic operations. By dividing the projection space into discrete regions or voxel groups that can be processed independently, the system eliminates the need for atomic operations and synchronization, allowing both forward and back projection to achieve high parallel computation efficiency
Data Source
AI summary
Back projection voxels smaller in size than forward projection voxels are created. Back projection is performed by the use of computed pixel values and measured pixel values at intersection points between straight lines connecting an X-ray source with the centroids of the back projection voxels and an X-ray detection panel.