Patch-Based Spectral CT Decomposition for Four or More Materials
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing material decomposition methods in spectral imaging, such as dual energy CT, struggle to accurately decompose materials beyond three, as it is an ill-posed deterministic problem.
Innovation Solution
A computer system and method using statistical mixture models, particularly Gaussian Mixture Models (GMM), apply a patch-based approach to fit probability distributions to spectral diagrams, combining them into a probability map for robust material type classification, enabling decomposition into four or more materials.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If deterministic material decomposition methods are used for more than three materials, then material decomposition is attempted, but the problem becomes ill-posed and accuracy deteriorates
Solution Approach 1:
The patent changes the mathematical approach from deterministic linear system solving to probabilistic statistical modeling. By representing material compositions as probability distributions rather than fixed values, and using maximum likelihood estimation to fit these distributions to spectral data, the method can handle ill-posed problems with more than three materials while maintaining decomposition accuracy through statistical inference
Solution Approach 2:
The patent introduces probability distributions as an intermediary layer between the spectral measurements and material decomposition results. Instead of directly solving for material concentrations, the method fits probability distributions to spectral diagrams and uses these as intermediaries to infer material compositions, thereby resolving the ill-posed nature of decomposing more than three materials
2Adaptability or versatility
If existing decomposition methods are applied to medical applications requiring four or more materials, then material identification is attempted, but the ill-posed nature of the problem reduces reliability
Solution Approach 1:
The patent transforms the decomposition problem from a deterministic parameter estimation task to a probabilistic parameter fitting task. By using maximum likelihood estimation to fit probability distribution parameters to spectral data, the method achieves reliable identification of four or more material types even when the deterministic problem would be ill-posed
Solution Approach 2:
The patent replaces the mechanical/deterministic system of linear equations with a statistical/probabilistic system. Instead of solving Ax=b deterministically, the method uses probabilistic models and maximum likelihood estimation, substituting the rigid deterministic framework with a flexible statistical approach that handles medical imaging applications requiring identification of multiple material types
Data Source
AI summary
A computer system (MD) and relates method for spectral-data based material decomposition. The system comprises a statistical module (SM) configured to fit, per patch in input spectral imagery, a set of probability distributions (Pk) to a respective vector spectral diagram (5) for the respective patch (A). The patch is one of a plurality of patches in the input spectral imagery obtained by operation of a spectral imaging apparatus. The probability distributions are combinable into a probability map indicative of material type probabilities per image location in the input spectral imagery.