Graph Cut Data Sorting for Tomographic Image Labeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current techniques for sorting tomographic images based on body portions have limited accuracy, especially when dealing with features like the length of body parts, and existing graph cut algorithms do not effectively maintain the order and length of elements during the sorting process.
Innovation Solution
A data sorting apparatus and method using a graph cut process to allocate labels to tomographic images, where weights are set to prioritize links corresponding to elements with high scores, and additional links are introduced to regulate cutting based on upper and lower limit values for label allocation, ensuring accurate sorting while maintaining the order and length of elements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If recognition is performed in the unit of tomographic images using machine learning, then the recognition process can be performed independently for each image, but the recognition rate is limited and sorting accuracy deteriorates
Solution Approach 1:
The patent combines individual image recognition results with sequential ordering information to create a unified sorting system. The graph cut method merges recognition scores from multiple images with the constraint that portions must appear in a specific sequence, resolving the contradiction between independent processing efficiency and accurate recognition by integrating both aspects into a single optimization framework
2Productivity
If graph cut process is applied to sort tomographic images, then sorting can be performed efficiently, but existing methods do not effectively maintain the order and length of elements
Solution Approach 1:
The patent applies different weight characteristics to different links in the graph structure. s-links and t-links use weights based on recognition scores, while n-links use weights based on interval information. This local differentiation of link properties allows the graph cut algorithm to simultaneously achieve efficient sorting while maintaining accurate element order and length relationships
3Measurement precision
If anatomical features and hierarchical relationships are used to correct recognition results, then sorting accuracy is improved, but the method is limited to axial sections and does not handle length features
Solution Approach 1:
The patent creates a universal sorting method that works for any imaging modality and any arrangement of portions. By using graph cut with configurable s-links, t-links, and n-links, the system can handle axial sections, coronal sections, sagittal sections, and even non-anatomical data like gene sequences. The method incorporates both hierarchical relationships and length features through the interval information, making it broadly applicable beyond the limitations of previous anatomical-feature-based methods
Data Source
AI summary
Plural pieces of data where N (N>2) elements are arranged in a predetermined-direction in a specific-order are sorted into any one of N labels using a graph-cut-process. Each of the plural pieces of data has scores indicating element-likenesses for the plural respective elements. For each piece of data, weights are set about links along a first-direction directing from a node s to a node t so that a small weight is given to a link corresponding to an element having a maximum-score in the data. A weight for regulating cutting is set about links along a second-direction opposite to the first-direction and links along a direction in which the order of the respective pieces of data progresses. A graph-cut-process is executed on a graph for which the weights are set to determine links to be cut, and the N labels are allocated to the plural pieces of data.