EFEH-Net: Edge Feature Enhancement-Based Cardiac MRI Image Segmentation

Cardiac image segmentation is crucial in medical image analysis. However, the complex structure of the heart and variations in image quality often result in unclear boundaries, making traditional methods less effective. In this paper, we propose an edge feature enhancement-based cardiac image segmentation method to improve segmentation accuracy and robustness. We introduce an innovative encoder-decoder network architecture to enhance edge feature representation and overall segmentation performance. We integrate the Pyramid Edge Extraction (PEE) module in the encoder to extract and enhance edge features, improving the model's ability to perceive boundary regions. We also introduce the Multi-Scale Dual-Guided (MSDG) module in the skip connections, which strengthens the model's ability to recognize target boundaries and details through multiscale convolutions and the collaboration of edge and mask branches. Finally, the Confidence Prediction Refinement (CPR) module combines low-level fine-grained features with high-level semantic features, utilizing uncertainty maps and foreground confidence maps with attention mechanisms to enhance the model's perception of boundaries and foregrounds, thereby improving segmentation precision. Experiments on cardiac datasets validate the effectiveness of our method, showing significant improvements in edge accuracy and segmentation quality. These results highlight the synergistic effect of edge-specific information and multi-scale features, setting a new standard for image analysis applications.