Iterative X-ray CT Reconstruction with Target Kernel Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional filtered back projection methods in X-ray computerized tomography face limitations such as low-frequency cone beam artifacts and the inherent trade-off between image definition and noise, while iterative reconstruction methods lack operator control over image characteristics.
Innovation Solution
A method that allows for the selection of a 'virtual' target convolutional kernel to influence image characteristics during iterative reconstruction, using a regularization convolutional kernel to adjust image data and achieve desired image quality by iteratively refining the image data with residue images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional filtered back projection is used to achieve sharp image definition, then image definition is improved, but image noise increases
Solution Approach 1:
The invention transforms the image reconstruction problem from direct space to frequency space by applying Fourier transformation. In the frequency domain, the reconstruction is performed by multiplying the measured data with a transfer function that incorporates the desired modulation transfer function (MTF). This parameter transformation allows independent control of image definition and noise characteristics without the direct trade-off present in conventional filtered back projection methods.
Solution Approach 2:
The invention introduces a transfer function as an intermediary element between the measured projection data and the reconstructed image. This transfer function, which incorporates the desired MTF, acts as a mediator that shapes the frequency content of the reconstructed image. By using this intermediary, the system can achieve the desired image definition while separately controlling noise characteristics through the regularization parameter, eliminating the direct coupling between sharpness and noise in conventional methods.
2Object-affected harmful factors
If iterative reconstruction methods are used to reduce image noise, then image noise is reduced, but operator control over image characteristics is lost
Solution Approach 1:
The invention allows the operator to pre-define the desired modulation transfer function (MTF) before the reconstruction process. This desired MTF is incorporated into the transfer function used during reconstruction. By performing this preliminary specification of image characteristics, the operator maintains full control over the final image quality, including both noise levels and definition, while the iterative reconstruction process executes the predetermined specifications.
Solution Approach 2:
The invention implements a feedback mechanism where the reconstructed image is continuously compared with the desired image characteristics defined by the transfer function containing the desired MTF. The difference (residue) is used to iteratively update the image reconstruction, ensuring that the final image matches the operator's predetermined specifications for noise and definition. This feedback loop maintains operator control while achieving noise reduction.
3Speed
If conventional filtered back projection is used for reconstruction, then reconstruction speed is improved, but low-frequency cone beam artifacts and spiral artifacts occur
Solution Approach 1:
The invention replaces the mechanical filtered back projection algorithm with a frequency-domain based reconstruction method. Instead of using the traditional filtering and back-projection mechanical steps that are prone to artifacts, the system uses Fourier transformation and multiplication in the frequency domain. This substitution fundamentally changes the reconstruction mechanism, eliminating the geometric artifacts (cone beam and spiral artifacts) that are inherent in conventional FBP while maintaining computational efficiency.
Data Source
AI summary
A method and an image reconstruction device are disclosed for reconstructing image data on the basis of input projection data obtained via an X-ray computerized tomography system. A target convolutional kernel is selected, which, when reconstructing image data from the input projection data using simple filtered back projection, would lead to target image characteristics. Image data is then reconstructed using an iterative reconstruction method of at least one embodiment. In at least one embodiment, the method includes a) reconstructing image data of a first iterative stage from the input projection data, b) generating synthetic projection data on the basis of the image data of the current iterative stage, c) forming difference projection data on the basis of the input projection data and the synthetic projection data, d) generating residue image data from the difference projection data, e) combining the residue image data with the image data of the current iterative stage to form image data of an additional iterative stage, wherein the image data of the current iterative stage is subjected to filtering before or during combination with the residue image data by using a regularization convolutional kernel which is determined on the basis of the selected target convolutional kernel, and f) repeating b) to e) until a termination condition occurs.