Weak supervision medical image segmentation method and system based on causal intervention
By constructing a causal intervention module in medical image segmentation, decoupling causal features and mixed features, the problems of labeling difficulties and data confusion in medical image segmentation are solved, and the accuracy and diagnostic effect of segmentation are improved.
CN119693642AActive Publication Date: 2025-03-25XI AN JIAOTONG UNIV
Patent Information
- Application Number
- CN202411744137.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-30
- Publication Date
- 2025-03-25
- Estimated Expiration
- 2044-11-30
AI Technical Summary
Technical Problem
The prior art faces the problem of lack of pixel-level fine labeling and data confounding factors in medical image segmentation, resulting in poor segmentation performance.
Method used
Weakly supervised medical image segmentation method based on causal intervention is adopted to decouple causal characteristics and confounding characteristics by constructing a causal intervention module, and use causal intervention to remove the impact of confounding characteristics on the predicted results.
Benefits of technology
It improves the accuracy of medical image segmentation, especially when there are interferences of confounding factors, and significantly improves the diagnostic effect of brain images.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN119693642A_ABST
Abstract
The invention discloses a weak supervision medical image segmentation method and system based on causal intervention, and the method comprises the steps: constructing a causal intervention module after preprocessing an original image sample, and designing causal feature loss, hybrid feature loss and intervention loss; a causal category score and a mixed category score are respectively obtained through average pooling and are respectively used as inputs of causal loss and mixed loss; the causal features and random transformation hybrid features are combined to obtain hybrid scores as input of intervention loss; according to the method, an improved weak supervision medical image segmentation model is established, a target area can be more effectively positioned, a weak supervision brain tumor image segmentation model with better fitting can still be obtained under the interference of confounding factors, the accuracy of a brain image false segmentation mask is improved, and the diagnosis effect of a brain image is improved.
Need to check novelty before this filing date? Find Prior Art
Citation Information
Patent Citations
Medical image segmentation method and device, computer equipment and storage medium
CN109872333A
Vision Transform network-based weak supervision instance segmentation method and system, and medium
CN115359254A
Weak supervision medical image segmentation method based on causal invariance
CN116843903A
Out-of-distribution image classification method and system based on sample balance causal reasoning
CN116883728A
Gait recognition method and device based on causal intervention mechanism
CN117173790A
Cited By
Image recognition method and device, model training method and device, equipment and medium
CN120375098A