Deep Learning Sinogram Completion for Dual-Energy CT
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current dual-energy X-ray computed tomography (CT) systems face deficiencies in spectral imaging, including low-energy noise, energy separation degradation, high costs, and cross-scatter effects in dual-layer and dual-source systems, and substantial costs for fast kV-switching systems.
Innovation Solution
An X-ray CT system employing a deep learning (DL) network to acquire and process projection datasets with different energy spectra, using a learned model to generate complete projection datasets for improved sinogram completion and material decomposition, reducing hardware costs and enhancing image quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If dual-layer detector systems are used for spectral CT, then energy separation is achieved, but low-energy noise and poor out-of-band energy suppression occur
Solution Approach 1:
A learning-based sinogram completion model is introduced as an intermediary to process and refine the projection data from dual-layer detectors. The model learns to complete sinograms by leveraging correlations between high-energy and low-energy data, effectively separating energy bands while suppressing low-energy noise and improving out-of-band energy rejection.
2Productivity
If dual-source systems are used for spectral CT, then simultaneous dual-energy acquisition is achieved, but system cost and cross-scatter effects increase
Solution Approach 1:
Instead of using a second physical X-ray source, the system creates a virtual copy of the missing energy spectrum data through learned sinogram completion. The model generates the complete sinogram for one energy spectrum by learning from the other spectrum's data, eliminating the need for expensive dual-source hardware while achieving simultaneous dual-energy acquisition.
3Speed
If fast kV-switching systems are used for spectral CT, then dual-energy data acquisition is achieved, but substantial costs for ultra-high frequency generator and parallel DAS are required
Solution Approach 1:
The patent replaces the mechanical/electrical fast kV-switching system with a computational approach. Instead of physically switching kV at ultra-high frequencies, the system acquires data at standard switching speeds and uses a learning-based model to synthesize the complete dual-energy sinograms, substituting hardware complexity with intelligent algorithms.
4Device complexity
If sparse-kV switching is used to reduce hardware costs, then incomplete sinogram data is acquired, but image quality deteriorates
Solution Approach 1:
The learning-based sinogram completion model performs preliminary action by pre-learning the correlations between high-energy and low-energy projection data during training. This pre-learned knowledge enables accurate reconstruction of complete sinograms from sparse measurements during actual scanning, maintaining image quality while allowing cost-effective sparse-kV switching hardware.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The approach improves image quality and reduces hardware costs by enabling sparse-kV switching, overcoming the limitations of existing DECT systems through enhanced sinogram completion and material decomposition, resulting in high-quality images with less expensive hardware.
Implementation Method 1
The attenuation of the radiation that has passed through the body is measured by processing electrical signals received from the detector
Data Source
Figure 1
Figure 2A~2B
Figure 3A~3B
AI summary
According to one embodiment, an X-ray computed tomography (CT) system acquires a first projection dataset and a second projection dataset. The first projection dataset corresponds to a group of discrete first views, and is acquired with the emission of X-rays of a first energy. The second projection dataset corresponds to a group of discrete second views different from the first views, and is acquired with the emission of X-rays of a second energy different from the first energy. The X-ray CT system generates a third projection dataset corresponding to a group of third views and the X-rays of the first energy, and a fourth projection dataset corresponding to a group of fourth views and the X-rays of the second energy, by inputting the first projection dataset and the second projection dataset to a learned model.