SCEAF-UNet: Medical image segmentation based on spatial-channel feature enhancement and adaptive fusion

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Medical image segmentation requires balancing spatial and channel features for improved performance.
• Existing methods may struggle with detailed feature representation and accurate border delineation.

Purpose of the Study:

• To propose a novel module, the spatial-channel feature enhancement and adaptive fusion (SCEAF) module, to enhance medical image segmentation.
• To develop the SCEAF-UNet architecture by integrating the SCEAF module and an edge attention fusion (EAF) module.

Main Methods:

• The proposed SCEAF module combines a multi-scale spatial attention gated block (MSAGBlock) and a channel attention modulation block (CAMBlock) with gated fusion.
• The EAF module is integrated at skip connections to capture edge information and highlight structural contours.
• The SCEAF module is incorporated into the decoder of the RWKV-UNet backbone.

Main Results:

• The SCEAF-UNet architecture demonstrated significant performance improvements over existing models on the Synapse and ACDC datasets.
• Ablation studies confirmed the effectiveness and scalability of the SCEAF and EAF modules.
• The proposed modules enhance spatial detail recovery and channel feature discrimination.

Conclusions:

• The SCEAF-UNet architecture offers a superior approach to medical image segmentation.
• The SCEAF and EAF modules are effective, scalable, and adaptable for various medical image segmentation tasks.