Tracker-Less Image Registration with Semantic Landmark Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image fusion methods between dynamic and static imaging modalities require complex tracking hardware and processing, leading to high computational burden and time consumption.
Innovation Solution
A tracker-less method for registering and combining images using semantic and geometric descriptors to match landmarks between different image datasets, allowing for rapid image fusion without additional hardware.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If tracking hardware and complex processing methods are used for image fusion between dynamic and static modalities, then registration accuracy is improved, but device complexity and computational burden increase
Solution Approach 1:
The patent extracts and removes the tracking hardware component from the image fusion system. Instead of using external trackers to monitor probe position, the system uses landmark detection and matching algorithms that work directly with the image data itself, eliminating the need for separate tracking devices while maintaining registration capability
Solution Approach 2:
The patent replaces the mechanical/optical tracking system with a computational image processing system. Rather than physically tracking probe movement through external hardware, the system uses semantic and geometric descriptors to computationally match landmarks between dynamic and static images, substituting mechanical tracking with algorithmic registration
2Measurement precision
If tracking hardware and complex processing methods are used for image fusion, then registration accuracy is improved, but processing time and computational requirements increase
Solution Approach 1:
The patent performs preliminary action by pre-computing and storing semantic descriptors and geometric relationships for static reference images. When dynamic images need to be registered, the system can quickly compare against pre-processed reference data rather than performing full registration computations in real-time, significantly reducing processing time while maintaining accuracy
Solution Approach 2:
The patent segments the registration process into distinct computational stages: landmark detection, semantic descriptor extraction, geometric relationship computation, and matching. This segmentation allows each stage to be optimized independently and enables parallel processing of multiple landmarks, reducing overall computational burden and processing time
3Speed
If real-time image acquisition is used to follow moving tissues and intervention tools, then temporal resolution is improved, but image quality and signal-to-noise ratio deteriorate
Solution Approach 1:
The patent merges dynamic real-time images with static high-quality reference images through registration and fusion. The dynamic images provide temporal information and current anatomical position, while the static reference images provide high signal-to-noise ratio and detailed anatomical structures. By combining these complementary data sources, the system achieves both real-time monitoring capability and high image quality
Solution Approach 2:
The patent applies local quality by using the high-quality static reference images to enhance specific anatomical regions in the dynamic images. Through landmark-based registration, the system selectively transfers high-quality anatomical details from the static reference to the corresponding regions in dynamic images, improving local signal-to-noise ratio where it matters most while maintaining real-time temporal resolution
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method for tracker-less registering images of the same object generated from two different image datasets, and particularly from two different image datasets acquired by means of two different image acquisition methods and/or techniques comprising: identifying landmarks in the image or images of the two dataset; associating to each landmark a univocal semantic description comprising semantic labels describing at least one feature of the landmark and/or geometrical labels describing at least one or some of the geometric relationships between the landmarks identified and/or geometrical labels relating to the shape and/or orientation of the said identified landmarks; considering two images of the said two data sets as being images along image slicers or image planes registered one with the other when the semantic descriptions related to the landmarks present in the said images are matching one with the other.