Automatic coronary artery segmentation based on multi-domains remapping and quantile regression in angiographies

Area of Science:

• Medical Imaging
• Cardiovascular Diseases
• Computational Anatomy

Background:

• Coronary artery disease (CAD) is a leading cause of mortality.
• Accurate coronary artery segmentation is crucial for computer-aided diagnosis (CADx) and analysis.
• Existing segmentation techniques struggle with complex vascular textures and manual annotation limitations in coronary angiography.

Purpose of the Study:

• To develop a fully automatic coronary artery segmentation method for angiography images.
• To address challenges posed by complex vascular shapes, overlapping structures, and low contrast regions.
• To provide a robust and clinically practical solution for coronary artery segmentation.

Main Methods:

• A novel method employing multi-domains remapping for reliable boundary identification.
• Robust discrepancy correction utilizing distance balance and quantile regression.
• Application to automatic coronary artery segmentation of angiography images.

Main Results:

• The proposed method demonstrates robust segmentation of overlapping vascular structures.
• Achieves good performance in low contrast regions of coronary angiography.
• Achieved overall segmentation performances: si 95.135%, fnvf 3.733%, fvpf 6.113%, and tpvf 96.268%.

Conclusions:

• The developed automatic segmentation method offers a significant advancement for clinical practice.
• It overcomes limitations of existing methods in handling complex vascular textures and low contrast.
• The approach shows high effectiveness and accuracy in segmenting coronary blood vessels.