Entropy-Dependent Adaptive Filtering for PET Image Noise Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional PET image filtering methods, such as Gaussian and non-local means filtering, often reduce noise but also diminish the visual delineation of structures and biochemical quantitation, while modified methods requiring multi-modal registration and segmentation increase complexity and error susceptibility.
Innovation Solution
The proposed solution involves suppressing filtering strength based on voxel entropy, where higher entropy regions, like boundaries and organs, receive lower filtering strength, using entropy to determine the filtering strength for each voxel, and applying a non-local means filtering equation with a parameter related to the inverse normalized entropy, thereby preserving detail and quantitation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If a Gaussian filter is applied to reduce noise in PET images, then noise is reduced, but edges and regions of interest are smoothed and visual delineation is decreased
Solution Approach 1:
The patent applies different filtering strengths to different regions of the image based on local entropy characteristics. High-entropy regions (edges, boundaries, regions of interest) receive lower filtering strength to preserve detail, while low-entropy regions (homogeneous areas) receive higher filtering strength for noise reduction. This is achieved by calculating entropy for each voxel and using it to modulate the filtering parameter locally.
2Object-affected harmful factors
If traditional non-local means filtering is applied to reduce noise while preserving quantitation, then noise is reduced and quantitation is preserved, but contrast of edges is reduced and level of detail is decreased
Solution Approach 1:
The patent introduces dynamic adaptation of filtering parameters based on local image characteristics. The filtering strength is not fixed but dynamically adjusted for each voxel based on its entropy value. This allows the filter to adapt its behavior locally - being more aggressive in homogeneous areas and more conservative near edges and structures of interest.
3Reliability
If modified non-local means filtering is applied where target voxel acquires weight only if surrounding region is identical to reference voxel, then some shortcomings are addressed, but the prerequisite condition is unlikely to occur in PET images limiting effectiveness
Solution Approach 1:
The patent changes the parameter used to determine filtering strength from binary identity matching to continuous entropy-based weighting. Instead of requiring exact matches of surrounding regions, the patent uses entropy values to create a gradient of filtering strengths, making the filter applicable to the continuous nature of PET image data while maintaining robustness.
4Loss of information
If anatomical information from CT image is used to locate region boundaries and suppress filter strength in regions of interest, then detail in regions of interest is preserved, but complexity, cost and susceptibility to error increase due to registration and segmentation requirements
Solution Approach 1:
The patent makes the filtering process self-adaptive by using entropy calculations directly on the PET image data itself, without requiring external anatomical information or manual segmentation. The entropy-based mechanism automatically identifies regions of interest based on local variability, eliminating the need for complex registration and segmentation workflows while achieving similar protective effects for important structures.
Data Source
AI summary
Systems and methods include determination of an entropy value associated with each of a plurality of voxels of a three-dimensional image, determination, for each of the plurality of voxels, of a respective filter based on the entropy value associated with the voxel, wherein a first filter determined for a first voxel associated with a first entropy value is different from a second filter determined for a second voxel associated with a second entropy value different from the first entropy value, application, for each of the plurality of voxels, of the respective filter to a value of the voxel to generate a replacement value for the voxel, and generation of a filtered three-dimensional image based on the generated replacement value of each of the plurality of voxels.