MMSeg: Multi-scale Vision Mamba for Lightweight Generalizable Medical Image Segmentation.

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Medical image segmentation is vital for computer-aided diagnosis and treatment planning.
• Existing methods face challenges with cross-domain generalization and high computational costs.
• The Segment Anything Model (SAM), while powerful, has quadratic complexity hindering clinical use.

Purpose of the Study:

• To develop a lightweight and efficient medical image segmentation framework.
• To enhance cross-domain generalization and reduce computational complexity.
• To improve clinical deployment of AI in medical imaging.

Main Methods:

• Proposed MMSeg framework utilizing the Vision Mamba architecture.
• Introduced Multi-scale Lightweight Mamba Encoder (MLME) for efficient multi-scale feature processing.
• Developed Automatic Domain Matching Decoder (ADMD) for dynamic domain alignment.
• Incorporated Feature Distillation Module (FDM) for accelerated convergence using few-shot learning.

Main Results:

• MMSeg demonstrated superior performance compared to state-of-the-art methods on diverse medical datasets.
• The framework achieved significant improvements in cross-domain generalization.
• MMSeg maintained high computational efficiency, suitable for resource-constrained environments.

Conclusions:

• MMSeg provides an efficient and effective solution for medical image segmentation.
• The proposed framework addresses limitations of existing methods in generalization and computational load.
• MMSeg shows strong potential for practical clinical deployment in medical image analysis.