CFM-UNet: coupling local and global feature extraction networks for medical image segmentation

Area of Science:

• Medical Image Analysis
• Deep Learning Architectures
• Computer Vision

Background:

• Convolutional Neural Networks (CNNs) excel at local feature extraction in medical imaging but struggle with global context.
• Mamba networks capture long-range dependencies but may miss fine spatial details.
• A hybrid approach is needed to combine CNN and Mamba strengths for improved medical image segmentation.

Purpose of the Study:

• To develop a novel hybrid deep learning model, CNN-Fusion-Mamba-based U-Net (CFM-UNet), for enhanced medical image segmentation.
• To integrate CNNs for local feature extraction and Mamba for global feature extraction within a U-Net framework.
• To evaluate the performance and generalization ability of CFM-UNet on diverse medical imaging datasets.

Main Methods:

• The proposed CFM-UNet integrates CNN-based Bottle2neck blocks for local feature learning.
• Mamba-based visual state space blocks are incorporated for global feature extraction.
• A novel Spatial-channel Excitation Fusion (SEF) block is designed for effective feature fusion between parallel CNN and Mamba pathways.

Main Results:

• CFM-UNet demonstrated superior performance compared to existing advanced methods on multiple medical image segmentation tasks.
• The model achieved high accuracy in segmenting liver organs, liver tumors, spine, and colon polyps.
• CFM-UNet exhibited strong generalization capabilities, particularly in liver organ segmentation.

Conclusions:

• The proposed CFM-UNet effectively combines the complementary strengths of CNNs and Mamba for medical image segmentation.
• This hybrid architecture significantly improves segmentation accuracy and robustness.
• CFM-UNet represents a promising advancement in automated medical image analysis.