Likelihood Map Generation With Region Constraints for Target Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The preparation of ground-truth image data is laborious and existing methods for generating such data, like those based on bounding boxes, can reduce the accuracy of approximating the actual region of a target, necessitating a more efficient and accurate technique.
Innovation Solution
An information processing apparatus that acquires first information indicating the position and size of a target, and second information providing constraints on the target's region, to generate a likelihood map as ground-truth image data using a 3D U-Net model, ensuring the target's region is accurately constrained within the liver mask image data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If ground-truth image data is generated using bounding box methods, then the generation process becomes simpler and faster, but the accuracy of approximating the actual target region deteriorates
Solution Approach 1:
The patent segments the target region approximation into multiple components: using bounding boxes for initial localization and then applying constraint information to refine the region. This segmentation allows the system to maintain computational efficiency while improving accuracy through iterative refinement of the ground-truth image data generation process.
Solution Approach 2:
The patent changes the parameters used in ground-truth image data generation by incorporating constraint information (such as anatomical constraints in medical imaging) alongside traditional bounding box parameters. This parameter change enables the system to generate more accurate target region approximations while maintaining generation efficiency through automated constraint application.
2Measurement precision
If manual preparation of ground-truth image data is performed, then the accuracy and quality of the data is improved, but the time and labor required increases
Solution Approach 1:
The patent implements self-service by enabling the system to automatically generate ground-truth image data using algorithms that incorporate constraint information. The system serves itself by automatically processing input images, applying constraints, and generating accurate ground-truth data without requiring manual annotation, thus eliminating the time-consuming manual preparation process while maintaining high data quality.
Solution Approach 2:
The patent replaces the mechanical manual annotation process with an automated computational system. Instead of manually segmenting and labeling target regions, the system uses algorithmic approaches that process images automatically, applying constraint information to generate ground-truth data. This substitution eliminates human labor while maintaining or improving data quality through consistent automated processing.
3Device complexity
If the ground-truth image data is generated based on bounding box dimensions, then the generation process is simpler, but the accuracy of representing the actual target region deteriorates
Solution Approach 1:
The patent applies the nested doll principle by placing constraint information inside the bounding box framework. The bounding box provides the outer structure for simplicity, while constraint information is nested within to refine the target region representation. This nested approach maintains the simplicity of bounding box-based generation while incorporating the precision of constraint-based refinement to accurately represent the actual target region.
Data Source
AI summary
An information processing apparatus includes an acquisition unit configured to acquire first information including information indicating a position and a size of a target in training image data and second information including information indicating a constraint regarding a region of the target, and a generation unit configured to generate a likelihood map indicating likeliness of being the target as ground-truth image data corresponding to the training image data based on the first information and the second information.