Label Merging Function for Partially-Annotated Image Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Manual annotation of medical images for training neural networks is time-consuming and costly, making it prohibitive for large-scale studies, and existing methods require fully-annotated images for effective training, which limits the use of partially-annotated data.
Innovation Solution
A method that uses a label merging function to map labels from fully-annotated images to partially-annotated images, allowing the use of both types of images for training neural networks, thereby increasing the size of the training set and improving segmentation performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual annotation is used to create fully-annotated training images, then segmentation performance is improved, but time consumption and cost increase significantly
Solution Approach 1:
The patent applies partial annotation by using a label merging function that combines fully-annotated images with partially-annotated images for training. Instead of requiring complete annotations for all training data, the system merges labels from fully-annotated images with partially-annotated images, allowing training to proceed with less manual annotation effort while maintaining acceptable segmentation performance.
2Measurement precision
If only fully-annotated images are used for training, then model accuracy is improved, but the quantity of usable training data is limited
Solution Approach 1:
The patent merges fully-annotated images with partially-annotated images through a label merging function. This function maps labels from fully-annotated images to partially-annotated images, combining the strengths of both data types to create an expanded training set that maintains accuracy while increasing data quantity.
Solution Approach 2:
The label merging function serves multiple purposes: it processes fully-annotated images, partially-annotated images, and generates the merged training data. This universal approach allows the system to utilize diverse annotation levels in a single training pipeline, maximizing the utility of available data.
3Measurement precision
If specialized software and expert annotators are used, then segmentation quality is improved, but device complexity and operational difficulty increase
Solution Approach 1:
The label merging function automates the process of combining fully-annotated and partially-annotated images. Instead of requiring complex manual processes for each annotation type, the system uses a self-service merging mechanism that automatically processes both data types through a unified function, simplifying the overall workflow.
Data Source
AI summary
Methods and systems for training a model labeling two or more organic structures in an image. One method includes receiving a set of training images including a first plurality of images and a second plurality of images. Each of the first plurality of images including a label for a first subset of the two or more organic structures and each of the second plurality of images including a label for a second subset of the two or more organic structures, the second subset being different than the first subset. The method also includes training the model using the first plurality of images, the second plurality of images, and a label merging function mapping a label included in the first plurality of images to a label included in the second plurality of images.