Dual Prototypical Self-Supervised Learning for One-shot Medical Image Segmentation.

Area of Science:

• Medical imaging
• Artificial intelligence
• Computer vision

Background:

• Deep learning for medical image segmentation demands extensive annotated data, which is costly and time-consuming to acquire.
• Existing prototypical learning methods often overlook fine-grained details by averaging class-level prototypes.

Purpose of the Study:

• To develop a novel dual prototype network for one-shot medical image segmentation.
• To enhance feature extraction and model performance by incorporating part prototypes.

Main Methods:

• Proposed a dual prototype network incorporating global and local (part) prototypes.
• Utilized few-shot learning principles for medical image segmentation with minimal annotated data.
• Conducted experiments on the CHAOS dataset for validation.

Main Results:

• The proposed dual prototype network significantly outperformed existing methods in one-shot medical image segmentation.
• Part prototypes effectively captured fine-grained features, leading to enhanced segmentation accuracy.
• Demonstrated the method's efficacy on the CHAOS dataset.

Conclusions:

• The novel dual prototype network addresses limitations in prototypical learning for medical image segmentation.
• This approach offers a cost-effective solution for clinical practice by reducing annotation requirements.
• The method shows strong potential for improving one-shot medical image segmentation tasks.