Homographic Resampling for CT Image Reconstruction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Geometry-based models for medical imaging in computed tomography (CT) face challenges in scalability and implementation on massively parallel computing architectures, limiting precision and computational speed, especially when using graphics processing units (GPUs).
Innovation Solution
The use of one-dimensional homographic resampling transforms allows for arbitrary precision in image reconstruction by recasting continuous functions without assuming geometric representations of lines of integration, facilitating easier adaptation to deep learning paradigms and improving processing efficiency through massively parallelizable coefficient determination.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If geometry-based models are used for CT image reconstruction, then image reconstruction can be performed, but computational speed is limited and scalability to GPU architectures is difficult
Solution Approach 1:
The patent replaces geometry-based computational models with a homographic transform approach that uses algebraic operations. This substitution enables the system to achieve high computational speed on GPU architectures by eliminating complex geometric calculations while maintaining image reconstruction accuracy.
Solution Approach 2:
The patent changes the mathematical parameters and transformation methods from traditional geometry-based models to homographic transforms. This parameter change allows for more efficient computation on parallel architectures like GPUs, significantly improving computational speed while simplifying implementation.
2Measurement precision
If geometry-based models are used for CT image reconstruction, then image reconstruction can be performed, but precision and adaptability to deep learning techniques are limited
Solution Approach 1:
The homographic transform framework provides a universal mathematical representation that can accommodate various imaging geometries and is easily adaptable to deep learning techniques. This multi-functional approach allows the same mathematical foundation to support both traditional CT reconstruction and modern deep learning methods, enhancing both precision and adaptability.
3Productivity
If traditional CT reconstruction algorithms are used, then image data can be processed, but processing efficiency and computational performance are suboptimal
Solution Approach 1:
The patent segments the image reconstruction process into discrete homographic transform operations that can be independently optimized and executed in parallel on GPU architectures. This segmentation enables highly efficient processing by allowing simultaneous computation of multiple transform operations, significantly reducing total computational time while improving overall processing efficiency.
Data Source
Figure 1~2
Figure 3~4A
Figure 4B~5
AI summary
Systems and methods are provided for reprojection and backprojection of objects of interest via homographic transforms, and particularly one-dimensional homographic transforms. In one example, a method may include acquiring imaging data corresponding to a plurality of divergent X-rays, assigning a single functional form to the plurality of divergent X-rays, determining, via a homographic transform, weights of interaction between a plurality of distribution samples and a plurality of X-ray detector bins based on the single functional form, and reconstructing an image based on the weights of interaction.