Asymmetric Regularization for Cone-Beam CT Image Reconstruction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Cone-beam CT systems face artifacts due to data truncation in the z-direction, mishandled data, and missing frequencies, particularly in axial scan trajectories, leading to imperfections in reconstructed images.
Innovation Solution
The method involves asymmetric regularization of voxels based on adjacent voxels, varying regularization strength based on available projection data, and generating proportional weights for projection rays outside the reconstruction volume, along with iterative reconstruction and data synthesis to address data truncation and missing frequencies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If conventional uniform regularization is applied to all voxels, then the reconstruction process is simple, but artifacts appear in regions with data truncation
Solution Approach 1:
The patent applies different regularization strengths to different spatial locations based on data availability. Voxels are categorized into fully-sampled regions (with complete projection data) and truncation regions (with incomplete data), and different regularization parameters are applied to each region. This local differentiation improves reconstruction accuracy in truncation regions while maintaining computational efficiency through region-based classification.
Solution Approach 2:
The patent introduces asymmetric regularization where the strength of regularization applied to a voxel depends on its neighboring voxels' data completeness. Specifically, the regularization strength is modulated by an asymmetry factor that considers whether adjacent voxels are in fully-sampled or truncation regions. This asymmetric approach prevents error propagation from truncation regions to fully-sampled regions while maintaining simplicity through a systematic classification scheme.
2Area of stationary object
If data truncation in the z-direction occurs, then the scan coverage is reduced, but artifacts and missing frequencies appear in reconstructed images
Solution Approach 1:
The patent identifies and separately processes voxels in truncation regions versus fully-sampled regions. By applying enhanced regularization specifically to truncation region voxels (where data is incomplete) while using standard processing for fully-sampled voxels, the method improves reconstruction quality in affected areas without compromising the overall scan coverage efficiency.
Solution Approach 2:
The patent uses an asymmetry factor as an intermediary parameter that mediates between the regularization term and the data completeness status of neighboring voxels. This asymmetry factor acts as a bridge to transfer information about data availability from neighboring voxels to the current voxel's regularization strength, enabling error prevention without requiring complex inter-voxel communication or iterative correction schemes.
3Manufacturing precision
If iterative reconstruction is performed without asymmetric regularization, then computational speed is maintained, but error propagation occurs from truncation regions
Solution Approach 1:
The patent applies asymmetric regularization only to voxels in truncation regions or adjacent to them, rather than uniformly to all voxels. This localized application prevents error propagation at the critical interfaces between truncation and fully-sampled regions while minimizing additional computational overhead. The method calculates regularization strengths once per iteration based on voxel classification, maintaining efficiency.
Solution Approach 2:
The patent pre-classifies voxels into fully-sampled and truncation regions before the iterative reconstruction begins, and pre-calculates the asymmetry factors based on the spatial distribution of these regions. This preliminary classification and factor calculation eliminates the need for complex real-time decisions during iteration, reducing computational overhead while maintaining artifact reduction effectiveness throughout the iterative process.
Data Source
AI summary
Approaches are described for addressing artifacts associated with iterative reconstruction of image data acquired using a cone-beam CT system. Such approaches include, but are not limited to, the use of asymmetric regularization during iterative reconstruction, the modulation of regularization strength for certain voxels, the modification of statistical weights, and/or the generation and use of synthesized data.