Neural Network CT Reconstruction for Dynamic Scenes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computed tomography (CT) methods face challenges in dynamic scenes due to insufficient acquisition ability and the need for strong prior assumptions for reconstruction, limiting their applicability to specific scenarios.
Innovation Solution
A learning-based CT imaging and reconstruction method that measures scene density distribution using a neural network, employing an illumination multiplexing approach with a linear fully connected layer and nonlinear layers to improve sampling efficiency and reduce prior assumptions, allowing for high-quality reconstruction in general dynamic scenes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional CT methods use intensive sampling in different directions to achieve high-quality reconstruction, then reconstruction quality is improved, but acquisition time increases and the method cannot keep up with rapid scene changes in dynamic scenes
Solution Approach 1:
The patent applies preliminary action by pre-training a neural network model offline using intensive sampling data. The trained model stores reconstruction knowledge that can be rapidly applied during dynamic scene capture, eliminating the need for intensive sampling during actual acquisition while maintaining high reconstruction quality
Solution Approach 2:
The patent replaces the mechanical sampling system with a learning-based system. Instead of physically acquiring numerous samples through intensive sampling, the system uses a neural network that has learned from training data to reconstruct images from sparse measurements, substituting physical sampling with intelligent computation
2Productivity
If the number of light source samples is reduced to improve sampling speed, then acquisition speed is improved, but strong prior assumptions are required for reconstruction which limits applicability to certain scenarios
Solution Approach 1:
The patent achieves universality by training the neural network on diverse training data that encompasses multiple scene types and characteristics. The resulting model can handle various dynamic scenes without requiring scene-specific prior assumptions, making it applicable to general dynamic scenes including mechanical inspection, medical diagnosis, and other time-resolved applications
3Measurement precision
If existing methods use scene-specific properties to achieve reconstruction, then reconstruction accuracy for specific scenes is improved, but the method lacks universality and cannot be applied to general dynamic scenes
Solution Approach 1:
The patent creates a universal reconstruction model by training the neural network on comprehensive training data that includes various scene types. The model learns general reconstruction patterns that apply across different scenarios, eliminating the need for scene-specific algorithms while maintaining high accuracy for diverse dynamic scenes
Data Source
AI summary
A method for computed tomography imaging and reconstruction based on learning, which measures the scene density distribution in an illumination multiplexing manner. The light source(s) for imaging emit(s) light according to the intensity obtained by pre-learning, and the light from different directions is absorbed and attenuated by the scene and reaches a sensor. The measured values are calculated and reconstructed to obtain the density information of the scene. The illumination intensity and reconstruction algorithm are learned by a neural network. In this method, the CT imaging process is modeled as a linear fully connected layer, and the weight corresponds to the illumination intensity of the light source for imaging; the reconstruction algorithm is modeled as a nonlinear neural network, which can be optimized according to the characteristics of scanning geometry.