Image Registration Using Template Patch Similarity for Surgical Target Localization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image registration methods in image-guided radiation therapy face challenges in accurately and efficiently locating moving targets within the body by aligning pre-operative images with live images, due to differences in coordinate systems, imaging modalities, and anatomical changes, which affects the precision and speed of radiation delivery.
Innovation Solution
The method involves selecting template patches from a digitally reconstructed radiograph (DRR) based on their ability to distinguish target characteristics, computing similarity values at candidate locations in live x-ray images, and generating a global similarity map to determine the target's location, using weighted sums of patch-level similarity values to enhance accuracy and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional image registration methods are used to align pre-operative images with live images, then the target location can be determined, but the computational efficiency and accuracy are insufficient due to coordinate system differences, imaging modality differences, and anatomical changes
Solution Approach 1:
The patent divides the target object into multiple patches and processes each patch separately to compute local similarity values. This segmentation approach improves computational efficiency by breaking down the complex global registration problem into smaller, more manageable local comparisons, while maintaining accuracy through aggregated results from multiple patches.
Solution Approach 2:
The patent transforms the 3D volume data into 2D multi-planar reformatted (MPR) images for comparison. This dimensionality change simplifies the computational complexity of comparing volumetric data while preserving the essential anatomical relationships, thereby improving computational efficiency without significantly compromising localization accuracy.
2Measurement precision
If multiple transformations are applied to account for imaging system differences and anatomical changes, then the registration accuracy can be improved, but the computational complexity and time required increase
Solution Approach 1:
The patent extracts and separately processes different transformation components (rigid body transformation and deformable transformation) rather than applying them simultaneously. This extraction approach allows each transformation type to be optimized independently, reducing overall computational complexity while maintaining the accuracy benefits of multiple transformations.
Solution Approach 2:
The patent applies rigid body transformation first to achieve coarse alignment between images, then applies deformable transformation for fine-tuned local adjustments. This preliminary action sequence reduces the search space for the second transformation, thereby reducing computational complexity while maintaining high alignment accuracy.
3Measurement precision
If the target location is determined by comparing image content between two images, then the target can be located, but the precision is insufficient when the target moves between image acquisitions
Solution Approach 1:
The patent employs deformable transformation that can dynamically adapt to anatomical changes and target motion between image acquisitions. This dynamic approach allows the registration to accommodate non-rigid movements and deformations, improving target location precision while maintaining the ability to handle various types of motion through flexible transformation models.
Data Source
AI summary
An image registration process for detecting a change in position of a surgical target, such as a tumor, within a patient is disclosed. A pre-operative model of the target and surrounding area is generated, then registered to live patient images to determine or confirm a location of the target during the course of surgery. Image registration is based on a non-iterative image processing logic that compares a similarity measure for a template of the target (generated from the pre-operative model) at one location within the live image to other locations within the live image.