Image Registration Using Intensity Map for Low-Contrast Edge Alignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image registration techniques face challenges in accurately aligning low-contrasted image edges, often resulting in deformation of bone structures or requiring manual user interaction, especially in computed tomography (CT) and other imaging modalities.
Innovation Solution
The introduction of an intensity map that increases the dynamic range of low-contrasted tissue interfaces before registration, allowing for improved alignment of low-contrasted edges with high-contrasted edges without modifying the registration algorithm, using pre-stored or on-demand generated intensity maps based on anatomy and imaging protocols.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If spatially variable regularization is used to relax the regularization term for low-contrasted image edges, then registration accuracy for low-contrasted edges is improved, but device complexity and time consumption increase due to requiring prior knowledge such as detailed segmentation
Solution Approach 1:
The patent introduces an intermediary component - the intensity map - that mediates between the input images and the registration algorithm. This intensity map selectively enhances low-contrasted edges without requiring complex spatially variable regularization or manual segmentation, thus improving registration accuracy while maintaining process simplicity
Solution Approach 2:
The patent applies preliminary action by generating the intensity map before the registration process. This pre-processing step enhances low-contrasted edges in advance, allowing the standard registration algorithm to work effectively without requiring complex modifications or prior knowledge of tissue segmentation
2Measurement precision
If landmarks are used at low-contrasted image edges to improve registration, then alignment accuracy is improved, but ease of operation deteriorates due to requiring manual user effort to generate and place landmarks
Solution Approach 1:
The patent implements self-service by enabling the system to automatically generate the intensity map and perform edge enhancement without manual user intervention. The algorithm automatically identifies and enhances low-contrasted edges based on image intensity characteristics, eliminating the need for users to manually place landmarks while maintaining high alignment accuracy
3Measurement precision
If the regularization term is weakened to emphasize the similarity term for low-contrasted edges, then registration accuracy is improved, but reliability deteriorates due to deformation of bone structures
Solution Approach 1:
The patent applies local quality by creating an intensity map that selectively enhances only the low-contrasted edge regions while leaving other areas unchanged. This localized enhancement allows the similarity term to be emphasized specifically where needed (at low-contrasted edges) without globally weakening the regularization term, thus maintaining anatomical plausibility in bone structures while improving registration at tissue interfaces
Data Source
Figure 1
Figure 2~3
Figure 4~5
AI summary
A method includes increasing a dynamic range of a first sub-range (410, 510) of pixel intensity values of at least two images based on an intensity map, thereby creating at least two modified images, determining a deformation vector field between the at least two modified images, and registering the at least two images based on the deformation vector field. An image processing system (118) includes a processor (120) and a memory (122) encoded with at least one image registration instruction (124). The processor executes the at least one image registration instruction, which causes the processor to: increase a dynamic range of a first sub-range of pixel intensity values of at least two images based on an intensity map, thereby creating at least two modified images; determine a deformation vector field between the at least two modified images, and register the at least two images based on the deformation vector field.