CT Registration via Thin Plate Splines for Lung Deformation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current bronchoscopic navigation systems face challenges in accurately registering initial 3D models of patient airways from CT images with subsequent models due to lung flexibility and shape changes during procedures, leading to reduced navigation accuracy.
Innovation Solution
The system employs rigid registration, multi-rigid registration, and Thin Plate Splines (TPS) Transformation, combined with geometric and topological filtering, to align and match reference points between initial and subsequent 3D models, enhancing registration accuracy and adaptability to lung dynamics.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If manual registration is used to align initial and subsequent 3D models, then the system can accommodate lung flexibility and shape changes, but the registration accuracy deteriorates due to difficulty in visually identifying and associating airway branching points
Solution Approach 1:
The patent replaces manual visual identification and mechanical association of airway branching points with an automated image processing system. The system uses centerline extraction, skeletonization, and tree matching algorithms to automatically align the initial and subsequent 3D models, eliminating the imprecision of manual visual registration while maintaining adaptability to lung shape changes.
Solution Approach 2:
The patent introduces intermediate processing steps including centerline extraction, skeletonization, and tree matching as mediators between the initial and subsequent 3D models. These intermediaries enable automated comparison and alignment of airway structures, providing both accuracy through algorithmic precision and adaptability through flexible tree matching that accommodates lung deformation.
2Measurement precision
If automatic registration using point cloud survey is used to improve registration accuracy, then the matching precision between models improves, but the system complexity increases due to the need for additional survey procedures and segmentation algorithms
Solution Approach 1:
The patent performs preliminary actions by pre-processing the CT images to generate the initial 3D model with extracted airway centerlines and skeletonized representations before the bronchoscopic procedure. This preliminary preparation stores the airway geometry in a format optimized for rapid automated matching, reducing the complexity and time required during the actual procedure while maintaining high registration accuracy.
3Device complexity
If rigid registration methods are used to simplify the registration process, then the system complexity is reduced, but the navigation accuracy deteriorates because rigid transformation cannot account for lung shape changes during procedures
Solution Approach 1:
The patent implements a dynamic registration approach that combines rigid transformation for global alignment with flexible tree matching for local adaptation. The system first applies rigid registration to match the overall orientation and position of the initial and subsequent 3D models, then uses skeleton-based tree matching to accommodate local shape changes and deformations, achieving both simplicity and accuracy.
Solution Approach 2:
The patent segments the airway tree into hierarchical levels, applying different registration strategies to different segments. Rigid transformation is applied to proximal airways that undergo minimal deformation, while flexible tree matching is applied to distal airways that experience greater shape changes, optimizing both computational efficiency and registration accuracy across the entire airway tree.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Systems, methods, and computer-readable media for registering initial computed tomography (CT) images of a luminal network with subsequent CT images of the luminal network include obtaining initial CT images of the luminal network and subsequent CT images of the luminal network, generating an initial three-dimensional (3D) model of the luminal network based on the initial CT images of the luminal network, generating a subsequent 3D model of the luminal network based on the subsequent CT images of the luminal network, matching the initial 3D model with the subsequent 3D model based on a registration, and performing geometric filtering and/or topological filtering on the matching of the initial 3D model and the subsequent 3D model. In various embodiments, a thin plate splines transformation from the initial 3D model to the subsequent 3D model is derived based on portions of the initial 3D model and the subsequent 3D model that were matched by the matching and that remain matched after the geometric filtering or the topological filtering.