VEA-SegUNet: Edge-Enhanced Multi-Scale Network with F2-Optimization for Robust Coronary Artery Segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Disease Research

Background:

• Accurate coronary artery disease (CAD) diagnosis relies on precise vessel segmentation in coronary angiography.
• Challenges like low contrast, artifacts, and vessel overlap impede segmentation accuracy and small vessel identification.

Purpose of the Study:

• To introduce VEA-SegUNet, a novel method for improved vessel segmentation in X-ray coronary angiography.
• To address segmentation discontinuities and enhance the detection of small vessels in coronary angiograms.

Main Methods:

• VEA-SegUNet incorporates a vessel enhancement module (VEA) using unsupervised edge detection priors.
• A multi-scale deformable convolutional attention module within the U-Net encoder captures complex vascular structures.
• F2-score optimization and F2 loss are utilized to prioritize vessel connectivity.

Main Results:

• VEA-SegUNet demonstrated superior performance compared to six state-of-the-art U-shaped architectures across multiple datasets (DCA1, CHUAC, XCA).
• Achieved high scores: F1-score 79.1%, F2-score 85.1%, recall 89.3%, IoU 65.1%, accuracy 97.7%, AUC 98.8%.
• Maintained competitive computational efficiency.

Conclusions:

• VEA-SegUNet effectively enhances coronary artery segmentation in X-ray angiography.
• The method shows significant improvements in vessel continuity and small vessel detection.
• VEA-SegUNet is a practical and effective tool for coronary artery segmentation, aiding in CAD diagnosis.