PFMNet: Prototype-based feature mapping network for few-shot domain adaptation in medical image segmentation

Area of Science:

• Medical image analysis
• Deep learning
• Computer vision

Background:

• Deep learning models require large annotated datasets, which are scarce for rare diseases.
• Training robust models for rare disease image segmentation is challenging due to limited data.
• Few-shot domain adaptation (FSDA) offers a solution by leveraging limited target domain data.

Purpose of the Study:

• To introduce a novel prototype-based feature mapping network (PFMNet) for few-shot domain adaptation in medical image segmentation.
• To address the challenge of limited annotated data in rare disease research.

Main Methods:

• PFMNet utilizes an encoder-decoder architecture with a prototype-based feature mapping (PFM) module.
• The PFM module translates target domain features into source domain-like features for better decoder comprehension.
• The network is designed to perform effective few-shot segmentation with limited data.

Main Results:

• PFMNet demonstrated efficacy in few-shot medical image segmentation tasks.
• Experiments were conducted on cross-center optic disc/cup, polyp segmentation, and cross-modality cardiac structure segmentation.
• Performance was evaluated across 5-shot, 10-shot, 15-shot, and 20-shot settings.

Conclusions:

• The proposed PFMNet effectively addresses data scarcity in medical image segmentation via few-shot domain adaptation.
• PFMNet shows promise for improving rare disease research and other medical imaging applications with limited data.