Medical Image Voxel Labeling With Spatial Regularization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Dictionary matching in medical imaging leads to frequent local mislabeling due to noise, resulting in high noise floors and reduced image quality in parameter maps, despite postprocessing techniques that compromise resolution and sharpness.
Innovation Solution
A computer-implemented method using a discrete optimization algorithm with a cost function that incorporates spatial regularization through a second partial cost function, optimizing label selection for voxels or pixels based on reference information and material parameters, employing discrete Markov random field optimization algorithms to suppress high-frequency noise.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If dictionary matching is performed for each voxel or pixel separately, then the labeling process is computationally efficient, but high-frequency noise and local mislabeling occur frequently
Solution Approach 1:
The patent combines individual voxel labeling with spatial context by integrating a second partial cost function that considers neighboring voxels. This merging of local and spatial information reduces high-frequency noise and local mislabeling while maintaining computational efficiency through the optimized cost function formulation.
Solution Approach 2:
The patent implements feedback by using the labeling results of neighboring voxels to influence the labeling of the current voxel through the second partial cost function. This spatial regularization feedback mechanism suppresses noise and improves labeling reliability without significantly increasing computational complexity.
2Reliability
If postprocessing filtering is applied to reduce noise, then the noise floor decreases, but image resolution and sharpness are compromised
Solution Approach 1:
The patent applies spatial regularization as a preliminary action during the labeling process itself, rather than as a postprocessing step. By incorporating the second partial cost function that considers spatial relationships, the method prevents noise formation at the source, thereby maintaining image resolution and sharpness while reducing the noise floor.
3Reliability
If spatial regularization is incorporated into the cost function, then high-frequency noise is suppressed and image quality improves, but computational complexity increases
Solution Approach 1:
The patent manages computational complexity by optimizing the parameters of the cost function, including the weighting between the first and second partial cost functions. By carefully selecting these parameters and using efficient optimization algorithms, the method achieves spatial regularization with acceptable computational complexity while maintaining improved image quality.
Data Source
AI summary
Techniques are described for processing measurement information to select a respective label for multiple voxels or pixels of a medical image dataset. Reference information is described comprising a respective reference dataset for each one of the possible labels, and each label is associated with a material parameter. A respective label for the respective voxel or pixel is selected by using a discrete optimization algorithm to optimize a cost function comprising a first partial cost function and a second partial cost function, and the incorporation of the second partial cost function into the cost function introduces a coupling between the labels at different voxels or pixels.