Medical Image Segmentation Model Using Weakly Supervised Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for segmenting pulmonary lobes into pulmonary segments in medical imaging face challenges due to complex lung structures without explicit physical boundaries, leading to serrated boundaries and variability in labeling, which affects learning accuracy and efficiency.
Innovation Solution
A segmentation model learning method based on weakly supervised learning using first-type and second-type labeling information, where direct supervised learning is applied to voxels with clear pulmonary segment labeling and indirect supervised learning is used for voxels with pulmonary lobe labeling, optimizing network parameters to enhance learning efficiency and accuracy without requiring total pulmonary segment labeling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional distance-based segmentation method is used, then pulmonary segment labeling can be generated automatically, but the segmentation boundaries become serrated and unfavorable for observation
Solution Approach 1:
The patent introduces a smoothing filter as an intermediary processing step between the distance-based segmentation and final labeling. This mediator processes the initially generated segmentation results to eliminate serrated boundaries while preserving the overall segmentation structure, thus resolving the contradiction between automatic segmentation efficiency and boundary smoothness.
2Measurement precision
If fully supervised learning with manual labeling is used, then segmentation accuracy can be improved, but the labeling process becomes difficult and time-consuming
Solution Approach 1:
The patent employs weakly supervised learning where only partial manual labeling (pulmonary lobe labels) is performed instead of complete pulmonary segment labeling. The model learns from this partial supervision and automatically generates the remaining segmentations, significantly reducing manual labeling time while maintaining acceptable accuracy through the two-stage learning approach.
Solution Approach 2:
The learning process is segmented into two distinct stages: first learning pulmonary lobe segmentation from partial labels, then learning pulmonary segment segmentation within each lobe. This segmentation of the learning process allows the model to progressively acquire segmentation capabilities without requiring complete manual annotations upfront.
3Area of stationary object
If different doctors perform manual labeling, then comprehensive coverage can be achieved, but labeling variability significantly affects learning results
Solution Approach 1:
The patent introduces a smoothing constraint as an intermediary mechanism that mediates between different doctors' labeling variations. This constraint acts as a regularizer during model training, reducing the impact of labeling inconsistencies while preserving the essential segmentation patterns, thus improving reliability without sacrificing coverage.
Data Source
AI summary
A segmentation model learning method according to an embodiment includes learning that, based on a loss function value, includes performing supervised learning of the voxels in medical image data according to the region to which the voxels belong. The learning of the medical image data includes: using first-type labeling information, which is meant for segmenting a predetermined structure into a plurality of categories, about the voxels of a predetermined structure and causing a segmentation model to perform direct supervised learning that represents learning for segmentation of the predetermined structure into a plurality of categories; using second-type labeling information, which is meant for segmenting a massive region covering the predetermined structure into a plurality of blocks, about the voxels of a massive region and causing the segmentation model to perform indirect supervised learning that represents learning for segmentation of the massive region into a plurality of categories; and optimizing the network parameters of the segmentation model.