Deep Learning Sinogram Completion for Sparse CT Scans
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current dual-energy CT systems face challenges with sinogram completion in kV-switching, leading to artifacts and degraded image quality due to missing projection angles, particularly in sparse kV-switching systems, which require efficient reconstruction algorithms to overcome streak artifacts and beam hardening effects.
Innovation Solution
The implementation of a two-channel deep learning artificial neural network (DL ANN) for sinogram completion, utilizing complementary information from high-kV and low-kV projection data to correct imperfections and enhance image quality, employing a 2.5-dimensional approach and frequency splitting to reduce computational burden and hardware costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If sparse kV-switching is used to reduce hardware costs and computational burden, then device complexity and processing requirements are reduced, but image quality deteriorates due to streak artifacts and beam hardening effects from missing projection angles
Solution Approach 1:
A deep learning network is introduced as an intermediary component between the sparse projection data and the final image reconstruction. The network processes the incomplete sinogram data, learning to predict and fill in missing projection angles while correcting for artifacts, thereby enabling high-quality reconstruction from reduced hardware configurations
Solution Approach 2:
The deep learning model is trained in advance on complete sinogram data to learn the mapping between full and sparse projections. This preliminary training enables the model to perform sinogram completion during actual scanning, correcting for beam hardening and streak artifacts before image reconstruction occurs
2Manufacturing precision
If fast kV-switching is used to acquire complete projection data at all views, then image quality is maintained, but hardware costs and system complexity increase due to ultra-high frequency generator and parallel data acquisition systems
Solution Approach 1:
The patent replaces expensive, complex fast kV-switching hardware with a more economical sparse switching approach combined with computational correction. The deep learning model acts as a computational substitute for the expensive hardware, enabling quality reconstruction without requiring ultra-high frequency generators or parallel data acquisition systems
Solution Approach 2:
The patent substitutes the mechanical and electrical complexity of fast kV-switching hardware with a computational solution. Instead of using complex hardware to acquire complete data, the system uses a deep learning-based sinogram completion algorithm to reconstruct the missing information, replacing physical system complexity with computational processing
3Productivity
If traditional reconstruction algorithms are used with sparse projection data, then processing is simpler, but image quality deteriorates due to streak artifacts and beam hardening effects
Solution Approach 1:
The patent transforms the problem from direct image reconstruction to sinogram domain completion. By changing the processing parameter space to work in the sinogram domain rather than directly in the image domain, the deep learning model can effectively fill missing projections and correct artifacts, achieving both quality reconstruction and processing efficiency
Data Source
AI summary
A deep learning (DL) network corrects/performs sinogram completion in computed tomography (CT) images based on complementary high- and low-kV projection data generated from a sparse (or fast) kilo-voltage (kV)-switching CT scan. The DL network is trained using inputs and targets, which respectively generated with and without kV switching. Another DL network can be trained to correct sinogram-completion errors in the projection data after a basis/material decomposition. A third DL network can be trained to correct sinogram-completion errors in reconstructed images based on the kV-switching projection data. Performance of the DL network can be improved by dividing a 3D convolutional neural network (CNN) into two steps performed by respective 2D CNNs. Further, the projection data and DLL can be divided into high- and low-frequency components to improve performance.