Medical Image Registration via Sub-Image Segmentation and Stochastic Masking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The registration of medical image data sets is complex, especially when scene changes occur due to foreign objects or movement, leading to inaccurate coordinate transformations, particularly in intraoperative imaging where different imaging methods and devices are used, resulting in noise and contrast issues that impair registration accuracy.
Innovation Solution
A method involving subdividing medical image data sets into sub-images, performing individual registrations with stochastic masking to identify and mask out sub-images with scene changes, and then performing a final registration by excluding these sub-images to improve similarity measures and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional registration methods are used between medical images from different devices or methods, then registration can be performed, but scene changes (foreign objects, movement) cause incorrect or unusable coordinate transformations
Solution Approach 1:
The image is divided into multiple sub-images, and each sub-image is registered independently with its own similarity measure optimization. This segmentation allows the system to identify and exclude sub-images affected by scene changes while maintaining registration accuracy for unaffected regions.
Solution Approach 2:
Sub-images that are affected by scene changes are extracted and masked out from the registration process. By identifying sub-images with poor similarity measures and excluding them, the harmful effect of scene changes is removed from the overall registration calculation.
2Reliability
If multiple degrees of freedom are available for registration (translation, rotation, scaling, shearing, elastic/inelastic transformation), then the probability of successful registration increases, but the computing time increases
Solution Approach 1:
By segmenting the image into sub-images and performing independent registrations, the system can explore multiple degrees of freedom for each sub-image separately. This parallelized approach increases the probability of successful registration while managing computing time through selective processing of only relevant sub-images.
Solution Approach 2:
The system performs registration on only those sub-images that are not affected by scene changes, rather than processing the entire image. This partial action approach reduces computing time while maintaining high registration success probability for the relevant image regions.
3Adaptability or versatility
If images from different imaging methods (X-ray C-arm, CT) are registered, then comprehensive medical imaging is achieved, but noise and contrast differences impair registration accuracy
Solution Approach 1:
The system divides the image into sub-images and calculates separate similarity measures for each. This segmentation allows the system to handle noise and contrast differences locally, identifying sub-images with reliable similarity measures even when registering images from different imaging methods with varying noise and contrast characteristics.
Solution Approach 2:
Different sub-images are treated with different quality assessments based on their local characteristics. Sub-images with good similarity measures (indicating consistent noise and contrast properties) are weighted more heavily, while sub-images affected by imaging method differences are masked out, achieving local optimization for multi-device registration.
Data Source
AI summary
A method of registering two sets of medical image data taking into account scene changes can include providing a first and a second medical image data set by means of a medical device, subdividing the first and second medical image data sets into an equal number of sub-images, performing a number of individual registrations between the first and second medical image data sets with respective optimization of a similarity measure, identifying the sub-images that have a scene change, and performing a final registration between the first and the second medical image data set by masking out the identified sub-images or a masked out sub-image combination. For each individual registration, at least one sub-image of the first and/or second medical image data set can be masked out by means of a random process when determining the respective measure of similarity.