Multiscale Bilateral Filtering for MR Image Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional MR image processing techniques are not robust enough for wide-range scan types and neuroanatomies, leading to unsatisfactory segmentation results and challenges in attenuation correction for PET images due to imaging artifacts, anatomical variability, and poor registration.
Innovation Solution
The method involves transforming MR images into the Radon Domain, decomposing them into multiscale sinograms using bilateral filtering, and reconstructing the images to achieve robust segmentation and improved signal-to-noise ratio, enabling accurate segmentation of anatomical landmarks and attenuation correction for PET images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional MR image processing techniques are used, then the processing is simple and fast, but the robustness to noise and segmentation accuracy are insufficient
Solution Approach 1:
The patent segments the image processing into distinct domains: Radon domain transformation, multiscale decomposition, bilateral filtering, and inverse Radon transformation. Each domain handles specific processing tasks, allowing noise robustness to be achieved through systematic separation of processing steps rather than complex operations in a single domain.
Solution Approach 2:
The patent transforms images from the spatial domain to the Radon domain, adding a new dimensional perspective. This transformation enables noise filtering and segmentation operations that are difficult to perform in the original spatial domain, improving robustness through a different representational dimension.
2Measurement precision
If conventional techniques are used, then the processing is straightforward, but the segmentation accuracy and signal-to-noise ratio are insufficient
Solution Approach 1:
The patent employs dynamic multiscale processing where the bilateral filter operates at multiple scales (levels of detail). This allows the system to adaptively adjust processing parameters based on the scale of features being analyzed, improving segmentation accuracy for structures of varying sizes while maintaining computational manageability through systematic scale progression.
Solution Approach 2:
The patent changes parameters systematically through the Radon transformation and multiscale decomposition process. By transforming to the Radon domain and applying bilateral filtering at multiple scales, the system modifies image parameters (contrast, noise characteristics) in a controlled manner to enhance segmentation accuracy without requiring overly complex processing.
3Adaptability or versatility
If conventional techniques are used, then the processing is simple, but the adaptability to different scan types and neuroanatomies is insufficient
Solution Approach 1:
The patent creates a universal processing framework that can handle multiple scan types and neuroanatomies through the same core operations: Radon transformation, multiscale bilateral filtering, and inverse transformation. The methodology is designed to be broadly applicable across different imaging scenarios without requiring scan-type-specific algorithms, achieving versatility through a unified approach.
Data Source
AI summary
Systems, methods and computer-readable storage mediums relate to segmenting MR images using multiscale bilateral filtering. Before the multiscale bilateral filtering, the MR images are transformed from the Image Domain to the Radon Domain.