Hierarchical Auto-labeling of Coronary Arteries on CT Coronary Angiography Images

Area of Science:

• Cardiovascular Imaging
• Medical Image Analysis
• Computational Anatomy

Background:

• Accurate labeling of coronary artery segments is crucial for diagnosing cardiovascular diseases.
• The complexity and diversity of coronary artery structures present significant challenges for automated analysis.
• Existing methods struggle with the intricate nature of coronary anatomy.

Purpose of the Study:

• To develop a robust hierarchical scheme for automatic labeling of coronary artery segments.
• To improve the accuracy and efficiency of coronary artery analysis from medical imaging data.
• To provide a reliable tool for cardiovascular disease diagnosis support.

Main Methods:

• A hierarchical scheme utilizing PointNet++ models and novel topological structural features.
• Input consists of 3D coronary artery centerline points from CTCA images.
• A two-stage labeling process: identifying main branches (LAD/LM, LCX, RCA) followed by sub-branch indexing based on connectivity.

Main Results:

• The proposed method demonstrated satisfactory accuracy on a private clinical dataset.
• Successful identification of main coronary artery branches and their sub-branches.
• The hierarchical approach effectively handles the complexity of coronary artery structures.

Conclusions:

• The developed auto-labeling scheme shows significant promise for clinical application in cardiovascular diagnostics.
• The integration of topological features enhances the accuracy of coronary artery segmentation.
• This method offers a potential advancement in automated analysis of coronary CT angiography (CTCA) data.