MDEU-Net: Medical Image Segmentation Network Based on Multi-Head Multi-Scale Cross-Axis.

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Attention mechanisms, particularly cross-axis attention, have advanced medical image segmentation.
• Challenges persist in multi-scale feature extraction and fine-detail capture for complex medical images.

Purpose of the Study:

• To introduce a novel network architecture, MDEU-Net, for improved medical image segmentation.
• To address limitations in handling complex images, multi-scale feature extraction, and fine-detail capture.

Main Methods:

• Proposed a multi-head multi-scale cross-axis attention MDEU-Net architecture.
• Employed a multi-head attention mechanism for parallel feature processing.
• Integrated a gated attention mechanism for efficient feature fusion and selective emphasis.
• Incorporated residual connections to mitigate gradient vanishing and enhance complex structure capture.

Main Results:

• The MDEU-Net architecture effectively captures both local and global information.
• The model excels at extracting features across various spatial scales.
• Gated attention and residual connections improve the capture of critical details and complex structures.
• Experimental results show superior performance compared to traditional architectures.

Conclusions:

• MDEU-Net offers an effective solution for medical image segmentation challenges.
• The proposed architecture enhances computational efficiency and processing speed.
• MDEU-Net demonstrates significant performance improvements in medical image segmentation tasks.