2D/3D Image Segmentation Refinement for Uncertain Boundaries
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image segmentation methods, both manual and automatic, are inadequate for achieving precise and accurate segmentation of objects in medical images such as CT or CBCT, suffering from human bias, complexity, and lower robustness.
Innovation Solution
A computer-implemented method that includes obtaining an initial segmentation, determining an uncertain area, processing it to find the most visible boundary using graph cuts and geodesic curves, and classifying discrepancies to refine the segmentation using trained models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If manual segmentation methods are used, then segmentation can be performed with interactive tools, but the results are prone to human bias and lack precision
Solution Approach 1:
The patent introduces an intermediary system that combines manual annotation with automatic processing. The semi-automatic method uses user-provided seed points or rough annotations as input, then applies automated algorithms (region growing, graph cuts, random walks) to complete the segmentation, eliminating direct human drawing while preserving user control and intent.
Solution Approach 2:
The patent replaces the mechanical act of manual pixel-by-pixel or contour drawing with automated computational processes. Instead of humans directly manipulating segmentation boundaries, the system uses algorithms that automatically propagate user intent through the image data, substituting mechanical human action with computational automation.
2Measurement precision
If semi-manual segmentation methods are used, then precision can be improved, but the process becomes complex and time-consuming
Solution Approach 1:
The patent applies preliminary actions by pre-processing the image data before segmentation. This includes computing intensity profiles, gradients, and other feature maps in advance, so that when segmentation is triggered, the computationally intensive work has already been prepared, enabling fast execution of the actual segmentation algorithm.
Solution Approach 2:
The patent segments the segmentation process itself into distinct phases: user input phase (providing seed points or rough annotations), automatic propagation phase (using algorithms like region growing or graph cuts), and refinement phase (post-processing to smooth boundaries or remove artifacts). This segmentation of the process allows each phase to be optimized independently.
3Productivity
If automatic segmentation methods are used, then speed and convenience are improved, but robustness and precision are reduced
Solution Approach 1:
The patent implements feedback loops where the segmentation results are evaluated against the original image data and user inputs. The system provides feedback to users about confidence levels, ambiguous regions, or potential errors, allowing users to review and correct problematic areas. This feedback mechanism maintains robustness by catching automatic algorithm failures while preserving speed through automated processing of clear-cut cases.
4Measurement precision
If precise segmentation following the most visible contour is required, then segmentation accuracy improves, but the method becomes more complex
Solution Approach 1:
The patent moves the segmentation problem from 2D image space to a higher-dimensional parameter space by incorporating multiple features (intensity, gradient magnitude, texture, spatial relationships) and multiple constraints (user inputs, anatomical priors, topological constraints). This dimensional elevation allows complex contour-following requirements to be satisfied through systematic optimization rather than complex geometric algorithms.
Data Source
AI summary
A computer-implemented method for segmenting an image, including (S1) obtaining a 2D/3D image I of a region of interest; (S2) performing an initial segmentation of the 2D/3D image I; (S3) processing each image Pj, from one or several error margins of the initial segmentation of the image I so as to obtain a corresponding image Uj; (S4) processing the pixels/voxels of the 2D/3D image I corresponding to the uncertain area of the image Uj; (S5) determining an image Ej of discrepancy between the pixels/voxels Fjn of the mask Fj and the pixels/voxels Pjn of the image Pj; (S7) performing an assignment processing of each pixel/voxel Ejn with discrepancy to assign it or not to the object Aj and thus obtain a final segmentation mask Mj including a refined segmentation of the object Aj.