CT Scatter Estimation With Cascaded Neural Refinement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current deep learning methods for scatter estimation in CT imaging suffer from insufficient accuracy and inefficient processing time due to the direct use of measured projection data, which lacks information on scattered radiation distribution.
Innovation Solution
A cascaded neural network framework is employed, utilizing a physical-process-based model to calculate a first-order scatter distribution, followed by a neural network to determine a total scatter distribution, and adaptive sparse sampling to reduce processing time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If deep learning methods directly use measured projection data as input to estimate scatter distribution, then processing speed is improved, but estimation accuracy deteriorates
Solution Approach 1:
The patent segments the scatter estimation process into multiple stages: first estimating a low-resolution scatter distribution quickly using deep learning on down-sampled projection data, then refining this estimate on selected detector pixels using the full-resolution projection data. This segmentation allows the system to benefit from both the speed of deep learning and the accuracy of detailed calculation.
Solution Approach 2:
The patent performs preliminary scatter estimation using down-sampled projection data before the final scatter correction. This preliminary estimate provides an initial scatter distribution that guides subsequent refinement steps, enabling faster processing while maintaining accuracy in the final result.
2Measurement precision
If full-resolution projection data is used for scatter estimation, then estimation accuracy is improved, but processing time increases
Solution Approach 1:
The patent applies partial action by performing detailed scatter estimation only on a selected subset of detector pixels rather than all pixels. Pixels are selected based on their scatter contribution, allowing the system to achieve sufficient accuracy for diagnostic purposes while significantly reducing processing time.
Solution Approach 2:
The patent divides the detector pixels into different groups based on their scatter characteristics. Full-resolution processing is applied only to pixels with significant scatter contribution, while other pixels use the lower-resolution estimate, thereby balancing accuracy and processing time.
3Measurement precision
If scatter correction is performed using traditional methods, then processing accuracy is maintained, but processing speed deteriorates
Solution Approach 1:
The patent replaces traditional iterative physics-based scatter estimation methods with a deep learning-based approach. The neural network is trained to predict scatter distributions directly from projection data, substituting the mechanical iterative calculation process with a faster learning-based inference process that maintains sufficient accuracy.
Solution Approach 2:
The patent performs preliminary scatter estimation using down-sampled projection data before the final scatter correction. This preliminary estimate provides an initial scatter distribution that guides subsequent refinement steps, enabling faster processing while maintaining accuracy in the final result.
Data Source
AI summary
A method for scatter estimation in a CT including a detector having multiple detector pixels includes: obtaining projection data by scanning an imaging object; reconstructing image data from the projection data; estimating, based on the projection data, a first scatter distribution; selecting, based on the first scatter distribution, a first subset of the pixels; calculating, based on the projection data and the image data, a second scatter distribution with respect to the selected first subset, the second scatter distribution having higher accuracy than the first scatter distribution; acquiring, based on the second scatter distribution, a third scatter distribution with respect to a second subset of the pixels, the third scatter distribution having higher spatial resolution than the second scatter distribution.