X-ray Scatter Correction in CT Using Neural Network Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for X-ray scatter correction in computed tomography (CT) are either computationally intensive or lack accuracy, particularly in estimating scatter for complex objects and large detectors, leading to poor image quality and artifacts.
Innovation Solution
A medical processing apparatus using a 3D convolutional neural network trained on spectral CT data to estimate and correct X-ray scatter by generating energy-resolved projection images for material components, allowing for fast and accurate scatter simulation and correction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If Monte Carlo methods are used to estimate X-ray scatter, then accuracy is improved, but computation time and complexity increase significantly
Solution Approach 1:
The patent creates a computational model (neural network) that copies and learns from the complex physics of X-ray scatter without performing actual Monte Carlo simulations. The neural network is trained on simulated scatter data and then applied to real CT data, providing accurate scatter estimation without the computational intensity of direct Monte Carlo methods.
Solution Approach 2:
The patent replaces the mechanical/computational physics simulation (Monte Carlo) with an information-processing system (neural network). Instead of physically simulating photon interactions through complex calculations, the system uses a trained neural network to predict scatter patterns from CT images, substituting computational physics with machine learning inference.
2Measurement precision
If conventional scatter correction methods are used, then image quality is improved, but processing time increases
Solution Approach 1:
The neural network model is pre-trained using simulated CT data with known scatter patterns before being applied to actual patient scans. This preliminary training allows the system to make accurate scatter estimates during real-time processing without performing time-consuming simulations during the actual CT acquisition or reconstruction.
Solution Approach 2:
The system creates a simplified computational representation (neural network) that captures the essential relationships between CT images and scatter patterns. This copy allows for rapid inference during actual processing, avoiding the need to recompute complex scatter physics for each image.
3Productivity
If simple scatter estimation methods are used, then processing speed is improved, but accuracy deteriorates
Solution Approach 1:
The patent changes the parameters of the estimation system by using a neural network with specific architectural parameters (number of layers, filter sizes, activation functions) that have been optimized to balance speed and accuracy. The network processes input images through multiple convolutional layers with carefully designed parameters to achieve both rapid processing and accurate scatter estimation.
Data Source
Figure 1
Figure 2A
Figure 2B
AI summary
According to one embodiment, a medical processing apparatus includes a processor. The processor obtains projection data representing an intensity of an X-ray flux detected at a plurality of detector elements. The processor generates, based on the projection data, (i) a plurality of first projection images corresponding to a first material component and a plurality of energies and (ii) a plurality of second projection images corresponding to a second material component and a plurality of energies. The processor estimates an X-ray scatter flux included in the X-ray flux based on the first projection images and the second projection images. The processor corrects the projection data based on the estimated X-ray scatter flux.