Rethinking Dual-Stream Super-Resolution Semantic Learning in Medical Image Segmentation.

Area of Science:

• Medical Image Analysis
• Computer Vision
• Deep Learning

Background:

• Accurate medical image segmentation is crucial for diagnosis and treatment planning.
• High-resolution segmentation often requires significant computational resources, limiting practical application.
• Existing dual-stream frameworks with super-resolution tasks show insufficient feature similarity for small medical targets.

Purpose of the Study:

• To propose a novel Dual-Stream Shared Feature (DS2F) framework to enhance medical image segmentation.
• To address the limitations of insufficient feature similarity in current dual-stream learning approaches for medical imaging.
• To improve the accuracy and efficiency of segmenting small structures like vessels and lesions.

Main Methods:

• Developed the Dual-Stream Shared Feature (DS2F) framework with a Shared Feature Extraction Module (SFEM).
• Introduced the Multi-Scale Cross Gate (MSCG) as a novel SFEM utilizing multi-scale features.
• Defined a proxy task and proxy loss to guide feature learning towards relevant targets.

Main Results:

• The DS2F framework demonstrated effectiveness across six diverse public datasets and three medical imaging scenarios.
• Ablation studies confirmed the significant contribution of the proposed DS2F framework and its components.
• The novel MSCG module effectively utilized multi-scale features for improved segmentation.

Conclusions:

• The proposed DS2F framework significantly improves medical image segmentation performance, particularly for challenging small targets.
• The integration of a shared feature extraction module with a specifically designed proxy task enhances feature representation.
• This approach offers a computationally efficient and accurate solution for medical image analysis tasks.