Enhanced Few-Shot Learning for Medical Image Segmentation via Meta-Learning with Attention-Guided Feature Augmentation

Abstract

Medical image segmentation is a crucial task in computer-aided diagnosis, enabling accurate localization and delineation of anatomical structures and pathological regions. However, deep learning-based segmentation methods typically require large amounts of annotated data, which are often scarce and expensive to acquire in the medical domain. Few-shot learning (FSL) offers a promising solution by enabling models to learn from limited labeled examples. This paper proposes an enhanced FSL framework for medical image segmentation that combines meta-learning with attention-guided feature augmentation.  Specifically, we employ a Prototypical Network-based meta-learning architecture, which learns to extract task-specific prototypes from support sets. To address the challenge of limited data, we introduce an attention mechanism that focuses on salient image regions and guides feature augmentation, thereby enhancing the diversity and representativeness of the support set features.  Experimental results on benchmark medical image segmentation datasets demonstrate that the proposed method significantly outperforms existing FSL approaches, achieving state-of-the-art performance with minimal labeled data. The proposed approach holds substantial promise for improving the efficiency and effectiveness of medical image analysis, particularly in scenarios with limited labeled data.