Prototype-Guided Graph Reasoning Network for Few-Shot Medical Image Segmentation.

Area of Science:

• Computer Vision and Machine Learning
• Medical Image Analysis

Background:

• Few-shot semantic segmentation (FSS) is crucial for medical tasks with scarce labeled data.
• Existing FSS methods struggle with significant appearance and scale variations between support and query objects in clinical settings.
• These variations lead to suboptimal segmentation masks in medical image analysis.

Purpose of the Study:

• To develop a novel prototype-guided graph reasoning network (PGRNet) for improved few-shot semantic segmentation.
• To explicitly explore contextual relationships within query images to overcome intra-class variances.
• To enhance the resilience of segmentation models to object variations in medical data.

Main Methods:

• Proposed a prototype-guided graph reasoning module for information interaction on query graphs, guided by support prototypes.
• Introduced a dynamic prototype generation mechanism to create multiple support prototypes from support images.
• Integrated these components to leverage structural properties and contextual information for robust segmentation.

Main Results:

• PGRNet demonstrated superior performance compared to existing FSS methods on three medical segmentation datasets (CHAOS-T2, MS-CMRSeg, Synapse).
• The method effectively overcomes intra-class variances by exploiting structural properties of query images.
• Achieved new state-of-the-art results in few-shot medical image segmentation.

Conclusions:

• PGRNet offers a robust solution for few-shot semantic segmentation in data-scarce medical scenarios.
• The proposed network architecture effectively handles object variations through enhanced contextual reasoning.
• This work sets a new benchmark for FSS in medical image analysis.