CT Image Reconstruction Using Low-Frequency Data Replacement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for correcting image uniformity in CT reconstructed images, such as adjacent channel interpolation and external information-based corrections, fail to effectively address low-frequency ring artifacts and uniformity issues, especially in scenarios with large non-uniformity and high computational demands.
Innovation Solution
Perform dimension downsampling on initial scanning data to merge pixel values, extract low-frequency information, and replace original low-frequency components with intermediate low-frequency information to improve image uniformity, applicable to both traditional CT and Photon Counting CT (PCCT) systems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If traditional methods (adjacent channel interpolation or external reference information) are used to correct image uniformity, then some correction is achieved, but the correction results are still poor and low-frequency ring artifacts remain
Solution Approach 1:
The patent merges data from multiple detector units through dimension downsampling to increase statistics. Specifically, it combines data from adjacent detector units along the channel direction and/or row direction to create intermediate scanning data with improved statistical reliability, which then enables effective low-frequency artifact removal and uniformity correction
Solution Approach 2:
The patent applies dimension downsampling to transform the data from high-dimensional raw scanning data to lower-dimensional intermediate scanning data by merging along channel and/or row directions. This dimensional transformation enables better statistical properties while preserving essential image information for uniformity correction
2Manufacturing precision
If dimension downsampling is performed to merge pixel values, then detector unit statistics are increased and low-frequency artifacts are eliminated, but computational resources and processing time are consumed
Solution Approach 1:
The patent performs dimension downsampling and low-frequency information extraction as preliminary processing steps before final image reconstruction. By pre-processing the scanning data to remove low-frequency artifacts and improve uniformity beforehand, the subsequent reconstruction process works with already-corrected data, potentially reducing iterative correction time
Solution Approach 2:
The patent segments the correction process into distinct stages: dimension downsampling to merge pixel values, low-frequency information extraction from the intermediate data, and final image reconstruction. This segmentation allows each stage to be optimized independently and enables parallel processing where applicable
3Manufacturing precision
If dimension downsampling merges pixel values from multiple pixels, then data statistics are improved and uniformity is enhanced, but image resolution may be reduced
Solution Approach 1:
The patent applies dimension downsampling selectively to specific portions of the scanning data (first portion along channel direction and/or row direction) while preserving other portions (second portion). This localized application maintains uniformity correction where needed while preserving high-resolution information in other regions
Solution Approach 2:
The patent creates composite scanning data by combining down-sampled merged data with original second portion data. This composite approach integrates the statistical benefits of merged data with the resolution benefits of original data, achieving both uniformity improvement and resolution preservation
Data Source
Figure 1
Figure 2
Figure 3(a)~4
AI summary
Provided are a method and a system for image reconstruction. The method includes: obtaining initial scanning data of a target object; performing dimension downsampling on the initial scanning data to obtain intermediate scanning data; wherein the dimension downsampling is configured to reduce the dimension of an image by merging the pixel values of multiple pixels of the image; extracting low-frequency information based on the intermediate scanning data to obtain intermediate low-frequency information corresponding to the intermediate scanning data; wherein the low-frequency information includes low-frequency images or low-frequency data; reconstructing a target reconstruction image of the target object based on the intermediate low-frequency information and the initial scanning data. In this method, the statistics of the detector units after data merging are increased by dimensionally downsampling the projection domain data collected by the detector; It does not rely on external information such as interpolation predictions or model predictions to ensure the accuracy of subsequent corrections and reconstructions, thereby improving the correction effect of the unevenness in the image and enhancing the uniformity of the image.