Semi-Supervised Medical Image Segmentation With Voxel Stability and Reliability Constraints

Area of Science:

• Medical Image Analysis
• Artificial Intelligence
• Computer Vision

Background:

• Semi-supervised learning is crucial for medical image segmentation due to high annotation costs.
• Teacher-student models with consistency regularization and uncertainty estimation show promise but face optimization challenges.
• Existing methods lack region-level uncertainty estimation, critical for segmenting blurry medical image regions.

Purpose of the Study:

• To propose the Voxel Stability and Reliability Constraint (VSRC) model to address limitations in semi-supervised medical image segmentation.
• To enhance parameter optimization and knowledge exchange between models.
• To introduce local region-level uncertainty estimation for improved segmentation accuracy.

Main Methods:

• Developed the Voxel Stability Constraint (VSC) for parameter optimization and knowledge exchange between independent models.
• Introduced the Voxel Reliability Constraint (VRC) for local region-level uncertainty estimation.
• Extended the model with task-level consistency regularization and uncertainty estimation for auxiliary tasks.

Main Results:

• The VSRC model successfully breaks through performance bottlenecks and avoids model collapse.
• Voxel Reliability Constraint (VRC) effectively captures local uncertainty in medical images.
• Experiments on 3D medical image datasets show superior performance compared to state-of-the-art methods under limited supervision.

Conclusions:

• The VSRC model offers a robust solution for semi-supervised medical image segmentation with limited annotated data.
• The proposed VSC and VRC strategies significantly improve model optimization and segmentation accuracy.
• This approach advances the field by providing a more effective way to handle uncertainty in medical image analysis.