Partial Annotation Strategy for Medical Image Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current digital image segmentation techniques face inefficiencies due to structural redundancy in medical images, where fully annotated images are often required for training automatic segmentation methods, leading to slow and expensive manual annotation processes.
Innovation Solution
A distributed annotation strategy is employed, where only one structure of interest is annotated in each digital image, using a predictive algorithm to determine the probability of other structures and background, reducing redundancy and enabling effective automatic segmentation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If fully annotated images are used for training automatic segmentation, then segmentation accuracy is improved, but annotation time and cost increase significantly
Solution Approach 1:
Instead of annotating all structures in each image (full annotation), the patent applies partial annotation by only annotating one structure per image. The predictive algorithm then infers the annotations for other structures, achieving sufficient training data without the time cost of complete manual annotation.
Solution Approach 2:
The system uses the predictive algorithm to automatically generate annotations for unannotated structures based on the annotated structure and structural correlations. This self-service mechanism eliminates the need for manual annotation of all structures, reducing time and cost while maintaining training quality.
2Measurement precision
If fully annotated images are used for training automatic segmentation, then segmentation accuracy is improved, but annotation cost increases significantly
Solution Approach 1:
The patent reduces the quantity of annotation work from full annotation (all structures) to partial annotation (one structure per image). This partial action is sufficient for training when combined with the predictive algorithm's inferences, significantly reducing annotation cost while maintaining model accuracy.
Solution Approach 2:
The predictive algorithm performs self-service by automatically generating annotations for multiple structures based on a single annotated structure and learned structural correlations. This eliminates the need for expensive manual annotation of all structures, reducing overall annotation cost.
3Productivity
If structural redundancy is reduced by annotating only one structure per image, then annotation efficiency is improved, but structural information is lost
Solution Approach 1:
The predictive algorithm uses feedback from structural correlations and spatial relationships to infer annotations for unannotated structures. The algorithm learns from the annotated structure and the image context to generate consistent predictions, maintaining structural information despite reduced annotation coverage.
Solution Approach 2:
The predictive algorithm acts as an intermediary that bridges the gap between limited annotated data and complete structural information. It uses structural correlations and spatial relationships as intermediaries to infer missing annotations, preserving structural information while enabling efficient partial annotation.
Data Source
AI summary
A method for automatically training and applying automatic segmentation in digital image processing is provided. The method may include, in response to receiving a plurality of digital images wherein each digital image associated with the plurality of digital images comprises only one annotated structure out of a plurality of structures included in each digital image, applying a predictive algorithm to each digital image that determines a predicted probability of each annotation in each digital image, determines a predicted background for each digital image, and merges the predicted probability of each annotation with the predicted background in each digital image. The method may further include, in response to applying the predictive algorithm, using the received plurality of digital images to train and apply an application for automatically segmenting unlabeled digital images.