Medical Image Grouping for Efficient Training Data Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing medical image generating devices often fail to apply label information to a sufficient number of images due to low similarity between images captured at different angles, leading to inefficient training data generation for machine learning.
Innovation Solution
A training data generating system that associates medical images based on similarities of imaging targets, selects an application target image, and applies representative training information to a larger number of images through geometric transformations and similarity-based grouping, reducing the need for user input.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If medical images are processed individually without association, then user input for each image is required, but this increases user workload and reduces efficiency
Solution Approach 1:
The patent merges multiple medical images into associated image groups based on similarity, allowing training information to be applied collectively rather than individually to each image, thereby reducing user workload and improving efficiency
Solution Approach 2:
The system performs preliminary association of medical images based on similarity before requiring user input, so that when training information is applied to one image in the group, it is automatically applied to all associated images, reducing the need for repeated user input
2Quantity of substance
If label information is applied to only a few images due to low similarity, then user input is minimized, but training data quantity becomes insufficient for machine learning
Solution Approach 1:
The patent groups similar medical images into associated image groups, enabling the system to apply training information to multiple images simultaneously based on a single user input, thereby increasing training data quantity while maintaining efficiency
Solution Approach 2:
A single user input for training information is made universal by applying it to all images in the associated group, allowing one action to serve multiple images and increasing the effective quantity of training data generated
3Adaptability or versatility
If images from different angles are processed separately, then each image can be processed independently, but similarity-based association fails to capture relationships
Solution Approach 1:
The patent segments medical images into associated image groups based on similarity, allowing independent processing within each group while maintaining the ability to capture relationships between images from different angles through the association mechanism
Data Source
AI summary
A training data generating system includes a processor. The processor acquires a plurality of medical images. The processor associates medical images with each other which are included in the plurality of medical images based on similarities of an imaging target to generate an associated image group including medical images associated with each other. The processor outputs, to a display, an application target image to be an image as an application target of representative training information based on the associated image group. The processor accepts input of representative contour information indicative of a contour of a specific region in the application target image as the representative training information. The processor applies contour information, as training information, to each medical image included in the associated image group based on the representative training information input to the application target image.