Multi-Energy CT Denoising via Deep Learning Spectral-Spatial Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Multi-energy CT imaging faces challenges in noise reduction and artifact correction due to statistically correlated noise in material images, which can lead to undesirable texture artifacts and material misidentification.
Innovation Solution
A deep learning-based denoising framework that jointly denoises multiple basis material images in both spectral and spatial domains, using neural networks to transform and process input images, reducing correlated noise and artifacts while preserving image texture.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multi-material decomposition is performed on multi-energy CT datasets, then material images representing multiple materials are obtained, but correlated noise is introduced that causes texture artifacts and material misidentification
Solution Approach 1:
The patent segments the noise reduction problem into two distinct domains: spectral domain processing to address correlated noise between different material images, and spatial domain processing to address noise within each individual image. This segmentation allows each domain to be optimized independently with appropriate techniques for that specific type of noise correction.
Solution Approach 2:
The patent introduces a deep learning-based denoising network as an intermediary component between the multi-material decomposition process and the final material images. This neural network acts as a mediator that learns the complex noise patterns and artifacts introduced by decomposition and systematically removes them while preserving genuine material features.
2Object-affected harmful factors
If conventional image denoising algorithms are applied to material images, then noise is reduced, but texture artifacts and material inaccuracy occur due to non-stationary correlated noise
Solution Approach 1:
The patent extends the denoising approach from traditional single-image spatial domain processing to a multi-dimensional framework that simultaneously processes multiple material images across both spectral dimensions (different materials/energies) and spatial dimensions. This allows the system to exploit correlations across all dimensions to distinguish noise from genuine material features more effectively.
Solution Approach 2:
The patent transforms the material images into a different parameter space using principal component analysis (PCA) or other linear transformations. In this transformed space, the correlated noise structure becomes more amenable to denoising operations, and the denoised images are then transformed back to the original material image space, preserving material accuracy while reducing noise.
3Object-affected harmful factors
If deep learning-based denoising is applied in both spectral and spatial domains, then correlated noise and artifacts are reduced, but computational complexity increases
Solution Approach 1:
The patent performs preliminary dimensionality reduction and noise characterization before applying the full deep learning denoising framework. By pre-processing the data to identify dominant noise patterns and reduce data dimensionality, the system prepares the input in a way that reduces the computational burden of the subsequent neural network processing while maintaining denoising effectiveness.
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 solution effectively reduces noise and artifacts in multi-energy CT images, producing clinically desirable, artifact-free material images and monochromatic representations with improved accuracy and texture preservation.
Implementation Method 1
jointly denoising the multiple basis material images in at least a spectral domain utilizing a deep learning-based denoising network
Data Source
AI summary
A computer-implemented method for image processing is provided. The method includes acquiring multiple multi-energy spectral scan datasets and computing basis material images representative of multiple basis materials from the multi-energy spectral scan datasets, wherein the multiple basis material images include correlated noise. The method also includes jointly denoising the multiple basis material images in at least a spectral domain utilizing a deep learning-based denoising network to generate multiple de-noised basis material images.