Mesh Collision Avoidance in Medical Image Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing model-based image segmentation methods struggle to simultaneously segment multiple objects in medical images without intersecting meshes, especially when organs slide relative to each other, as they fail to account for spatial relationships and often result in incorrect spatial adaptations.
Innovation Solution
A system that penalizes candidate features inside regions defined by other meshes, using a pre-positioning unit, transformation unit, adaptation unit, and control unit to iteratively adapt meshes and prevent overlap by adjusting the strength of features based on their distance from the region boundary, thereby maintaining accurate spatial relationships between objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If separate adaptation of multiple meshes is performed, then each mesh can be adapted independently to its corresponding object, but spatial relationships between objects are not accounted for resulting in wrong adaptation results such as intersecting meshes
Solution Approach 1:
The patent merges multiple separate mesh adaptation processes into a unified simultaneous adaptation framework. Multiple meshes are adapted at the same time while considering spatial relationships between them, preventing intersections and maintaining correct anatomical relationships. This is achieved by formulating a joint optimization problem that coordinates the adaptation of all meshes together rather than independently.
Solution Approach 2:
The patent implements feedback mechanisms where the adaptation of one mesh influences the adaptation of other meshes. Spatial constraints from previously adapted meshes are fed back into the optimization process for subsequent meshes, allowing iterative refinement that respects established spatial relationships while adapting new structures.
2Reliability
If additional edges connecting meshes are introduced to maintain spatial relationships, then mesh intersections can be avoided, but the method only works when spatial relationships can be described by pre-positioning meshes, which fails for sliding organs
Solution Approach 1:
The patent transitions from static pre-positioning constraints to dynamic adaptation. Instead of fixing spatial relationships through pre-positioned additional edges, the system allows meshes to adapt dynamically to their surroundings. The optimization process automatically adjusts mesh positions and shapes in response to image data and spatial constraints, enabling accurate segmentation of sliding organs like the heart or intestines without requiring predetermined spatial models.
Solution Approach 2:
The patent changes the parameters of the adaptation process to accommodate sliding organs. Rather than using fixed additional edges with predetermined lengths and positions, the system uses variable spatial constraints that can adapt to changing anatomical relationships. The optimization algorithm adjusts mesh vertex positions and mesh topology parameters dynamically to maintain valid spatial relationships throughout the adaptation process.
3Reliability
If candidate features inside regions defined by other meshes are penalized, then mesh overlap is prevented, but prominent features inside regions may still be incorrectly selected
Solution Approach 1:
The patent applies different quality weights to different candidate features based on their spatial location. Features inside regions defined by other meshes receive reduced weights or penalties, while features near boundaries or in unoccupied spaces receive higher weights. This local differentiation of feature quality allows the system to preferentially select features that maintain proper spatial relationships while still considering prominent anatomical structures.
Solution Approach 2:
The patent modifies the parameter space of feature evaluation by introducing spatial penalty terms to the feature strength calculation. The evaluation function incorporates distance-to-boundary metrics and region-based penalties that dynamically adjust the perceived strength of candidate features. This parameter transformation allows prominent features inside restricted regions to be downweighted while preserving the ability to detect and adapt to important anatomical structures through the optimization process.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The invention relates to a system(100) for segmenting an object in image data using model-based image segmentation, the system comprising a feature unit (120) for identifying features in the image data for computing an external energy of a mesh on the basis of a current position of the mesh, wherein the feature unit (120) further comprises a candidate feature unit (122) for selecting a plurality of candidate features in the image data, for identifying a feature to be included in the features identified in the image data, a position unit (124) for determining a position of each candidate feature of the plurality of the candidate features relative to a region of the image data, a feature function unit (126) for computing a strength of each candidate feature, wherein the strength of each candidate feature depends on the position of each candidate feature relative to the region, and an evaluation unit (128) for evaluating each candidate feature of the plurality of candidate features and for identifying the feature among the plurality of candidate features based on this evaluation. Determining whether a selected candidate feature is located inside the region which should be avoided, e.g., inside another mesh adapted to another object in the image data, allows penalizing this candidate feature during the computation of the strength of each feature and thus during the evaluation of the plurality of candidate features.