3D Image Linear Structure Detection via Curvature-Adaptive Normalization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting linear and planar structures in three-dimensional medical images, such as blood vessels, face challenges in accurately identifying curved, branched, or diseased structures due to data deficiencies and erroneous connections from noise points.
Innovation Solution
An image processing apparatus and method that calculates the directions of principal axes or normal lines within detection regions, normalizes candidate target regions, and uses machine learning to judge the presence of target tissues, employing Hessian matrices and cost functions to connect candidate points accurately.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If three perpendicularly intersecting sectional images are used for detection, then the total amount of data for calculation is decreased, but data regarding curved target tissues will be greatly lacking
Solution Approach 1:
The patent dynamically adjusts the number and orientation of sectional images based on the curvature of the target tissue. For straight structures, three perpendicular sections suffice, but for curved structures, the system increases the number of sections and adjusts their orientations to follow the curvature, thereby maintaining adequate data coverage while optimizing calculation efficiency.
Solution Approach 2:
The patent changes the parameters of the detection method by adapting the number, spacing, and orientation of sectional images according to the detected curvature of target tissues. This allows the system to maintain appropriate data density for curved structures while reducing unnecessary calculations for straight structures.
2Measurement precision
If Hessian matrices are used to extract linear structures, then blood vessels of ideally linear structures can be judged, but it is difficult to judge curved, branched, or diseased blood vessels
Solution Approach 1:
The patent transitions from static linear detection using Hessian matrices to a dynamic approach that adapts to the curvature of target tissues. By calculating curvature values and adjusting the detection strategy accordingly, the system can accurately detect both straight and curved blood vessels, as well as branched and diseased structures.
Solution Approach 2:
The patent introduces curvature as a new parameter to complement the Hessian matrix analysis. By combining curvature calculation with eigenvalue analysis, the system can distinguish between linear, curved, branched, and diseased structures, significantly expanding its detection capabilities beyond ideal linear structures.
3Ease of manufacture
If candidate points are connected by minimum spanning tree technique, then tree structures are reconstructed, but erroneous connections occur when noise candidate points are present
Solution Approach 1:
The patent introduces a feedback mechanism where curvature information is used to validate and correct the minimum spanning tree connections. By comparing the curvature along the connected path with the expected curvature of the target tissue, the system can identify and remove erroneous connections caused by noise points, thereby improving connection accuracy.
Solution Approach 2:
The patent uses curvature calculation as an intermediary to mediate between the minimum spanning tree algorithm and the final connection result. The curvature information acts as a filter that validates whether a connection between candidate points is anatomically plausible, preventing erroneous connections while maintaining the efficiency of the minimum spanning tree approach.
Data Source
AI summary
[Objective] To improve detection performance of target tissues formed by linear structures within three dimensional images.[Constitution] The directions of principal axes of target tissues formed by linear structures or the directions of lines normal to tissues formed by planar structures within detection regions are calculated. Normalization processes are administered with respect to candidate target regions that include candidate target tissues based on the directions of the principal axes or the directions of the normal lines. Features of the normalized candidate target regions are calculated, and judgments are performed regarding whether the target tissues are included in the candidate target regions, employing the calculated features.