Image Segmentation With Rater-Accuracy Vectors for Medical Precision
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image segmentation methods lack accuracy in identifying target objects within images, particularly in medical scenarios where precise segmentation of body parts is crucial.
Innovation Solution
An image segmentation method that incorporates prior knowledge vectors representing the labeling accuracy of multiple raters to fuse their labeling results, reconstruct labeled segmented images, and process the feature maps to enhance the accuracy of the final segmented image.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional image segmentation methods are used, then the segmentation process is simple and fast, but the accuracy of identifying target objects is insufficient
Solution Approach 1:
The patent applies preliminary action by integrating prior knowledge vectors into the neural network model before the actual segmentation process. These vectors encode labeling accuracy information from multiple raters, allowing the model to pre-adjust its decision boundaries and weighting schemes based on expected annotation quality, thereby improving segmentation accuracy before processing the actual medical images
Solution Approach 2:
The patent uses prior knowledge vectors as an intermediary element between the raw image data and the segmentation output. These vectors serve as a mediator that carries information about rater reliability and labeling accuracy, allowing the model to indirectly incorporate expert consensus information without directly processing multiple annotated versions of each image
2Measurement precision
If multiple rater labeling results are integrated, then the accuracy of image segmentation is improved, but the processing time and computational complexity increase
Solution Approach 1:
The patent extracts only the essential information from multiple rater labelings - specifically the prior knowledge vectors that represent labeling accuracy and rater reliability - rather than processing all the detailed annotation data. This extraction approach allows the model to benefit from multi-rater consensus while avoiding the computational burden of processing complete annotation sets from multiple raters
Solution Approach 2:
The patent changes the parameter representation by converting multiple detailed labeling results into compact prior knowledge vectors that capture the essential statistical properties of rater performance. This parameter transformation reduces the dimensionality and complexity of the input data while preserving the information needed for accurate segmentation
Data Source
AI summary
An image segmentation method includes: encoding an original image containing a target object based on a prior knowledge vector, to obtain a target feature map, the prior knowledge vector comprising a plurality of prior knowledge weights each representing accuracy of a corresponding rater labeling a region of an object in an image; decoding the target feature map, to obtain a first segmented image of the original image, the first segmented image indicating a target region in which the target object is located in the original image; performing image reconstruction on the first segmented image based on the prior knowledge vector, to obtain labeled segmented images, wherein one labeled segmented image corresponds to one prior knowledge weight and indicates a target region labeled by a corresponding rater; and processing the target feature map based on the labeled segmented images, to obtain a second segmented image of the original image.