Uncertainty-Guided Prototype Reliability Enhancement Network for Few-Shot Medical Image Segmentation

Area of Science:

• Medical Image Analysis
• Computer Vision
• Machine Learning

Background:

• Few-shot learning (FSL) is crucial for medical image segmentation due to limited labeled data.
• Current FSL methods struggle with effective prototype learning for accurate segmentation.
• Existing approaches often rely on basic nearest-neighbor searching for segmentation.

Purpose of the Study:

• To propose the Uncertainty-guided Prototype Reliability Enhancement Network (UPRE-Net) for few-shot medical image segmentation.
• To improve the reliability and informativeness of class prototypes in data-scarce medical imaging.
• To enhance the fusion of information from multiple support images for better segmentation outcomes.

Main Methods:

• Implemented a dual-support branch with augmentation for comprehensive information extraction from support images.
• Introduced an Uncertainty-guided Prototype Generation (UPG) module to select informative global and local prototypes using uncertainty measures.
• Developed a Reliable Dynamic Fusion (RDF) module for adaptive integration of dual-support branch predictions.
• Utilized an Uncertainty-induced Weighted Loss (UWL) to focus model training on high-uncertainty regions.

Main Results:

• UPRE-Net demonstrated significant performance improvements over state-of-the-art methods on four benchmark medical image datasets.
• The proposed UPG module effectively enhances prototype reliability by selecting informative prototypes.
• The RDF module successfully integrates information from dual-support branches for more robust segmentation.
• UWL loss function guided the model to prioritize uncertain regions, improving segmentation accuracy.

Conclusions:

• UPRE-Net offers a robust solution for few-shot medical image segmentation by enhancing prototype reliability and information fusion.
• The uncertainty-guided approach effectively addresses the challenges of limited data in medical segmentation.
• Experimental results validate the superiority of UPRE-Net, highlighting its potential for clinical applications.