Weakly Supervised Pathological Image Segmentation via Online Noise Suppression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for pathological image tissue segmentation require intensive pixel-level labeling, which is time-consuming and costly, especially for diverse and complex pathological images, necessitating professional pathologists for accurate labeling.
Innovation Solution
A weakly supervised pathological image tissue segmentation method using an online noise suppression strategy, which involves acquiring H&E stained images, processing them to generate patch-level labels, training a classification network to produce pseudo-labels, and then training a semantic segmentation network to achieve pixel-level segmentation results, reducing the need for extensive pixel-level labeling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If pixel-level labels are used for training segmentation networks, then segmentation accuracy is improved, but labeling time and cost increase significantly
Solution Approach 1:
The method performs preliminary classification at the patch level before segmentation, using a classification network to generate pseudo-labels that guide subsequent segmentation. This preliminary action reduces the complexity of direct pixel-level labeling while maintaining segmentation accuracy, as the classification network pre-processes the data to identify relevant regions and characteristics.
Solution Approach 2:
The patent introduces pseudo-labels as an intermediary between patch-level labels and pixel-level segmentation. The pseudo-labels serve as a bridge that translates coarse patch-level annotations into fine-grained segmentation guidance, enabling the segmentation network to learn from weaker supervision while achieving pixel-level accuracy without requiring extensive manual pixel-level labeling.
2Ease of manufacture
If patch-level labels are used for training, then labeling cost is reduced, but segmentation precision deteriorates
Solution Approach 1:
The patent divides the pathological image into patches and applies classification at the patch level, then uses the generated pseudo-labels to guide pixel-level segmentation. This segmentation approach allows independent processing of different regions, reducing labeling cost while maintaining overall segmentation precision through the two-stage framework.
Solution Approach 2:
The method transitions from direct pixel-level supervision to patch-level supervision, changing the dimensional level of supervision from fine-grained (pixels) to coarse-grained (patches). This dimensionality change reduces labeling complexity while the pseudo-label generation process recovers the fine-grained segmentation information needed for precise tissue boundary detection.
3Productivity
If pseudo-labels are generated without noise suppression, then training speed is improved, but segmentation reliability decreases
Solution Approach 1:
The patent converts the harmful effect of noise in pseudo-labels into a beneficial filtering process. By applying online noise suppression, the method identifies and removes incorrect pseudo-labels while retaining valid ones, thereby improving segmentation reliability. The noise suppression mechanism transforms the potential harm of noisy pseudo-labels into an opportunity for quality control and validation.
Solution Approach 2:
The online noise suppression strategy implements a feedback mechanism where the segmentation network's predictions are continuously evaluated and compared against the pseudo-labels. Incorrect predictions trigger noise suppression actions, and the refined pseudo-labels are fed back into training, creating a closed-loop system that progressively improves segmentation reliability while maintaining training efficiency.
Data Source
AI summary
Provided is a weakly supervised pathological image tissue segmentation method based on an online noise suppression strategy, including: acquiring a hematoxylin-eosin (H&E) stained graph, processing the H&E stained graph to obtain a data set, dividing the data set, training a classification network based on a divided data set, and generating a pseudo-label; suppressing a noise existing in the pseudo-label based on the online noise suppression strategy, and training a semantic segmentation network through the pseudo-label after noise suppression and a training set corresponding to the pseudo-label to obtain a prediction result of the semantic segmentation network after the training, and taking the prediction result as a final segmentation result.