A topology-preserving three-stage framework for fully-connected coronary artery extraction

Area of Science:

• Medical Imaging
• Computer Vision
• Cardiovascular Disease Analysis

Background:

• Accurate coronary artery extraction is vital for diagnosing coronary artery disease.
• Current methods struggle with thin vessels, complex structures, and poor contrast, leading to segmentation errors.

Purpose of the Study:

• To develop a topology-preserving framework for complete coronary artery extraction.
• To overcome limitations of existing segmentation methods in coronary artery analysis.

Main Methods:

• A three-stage framework: vessel segmentation with a new centerline enhanced loss, centerline reconnection using a regularized walk algorithm, and missing vessel reconstruction via implicit neural representation.
• The reconnection algorithm integrates distance, centerline probabilities, and directional cosine similarity.

Main Results:

• The proposed framework achieved high performance on ASOCA and PDSCA datasets.
• Achieved Dice scores of 88.53% and 85.07%, and Hausdorff Distances of 1.07 mm and 1.63 mm, respectively.
• Demonstrated superior performance compared to existing coronary artery extraction methods.

Conclusions:

• The developed framework effectively addresses challenges in coronary artery extraction.
• It enables more accurate and complete extraction of the coronary tree for improved diagnostic capabilities.
• The topology-preserving approach enhances the reliability of computer-aided diagnosis for coronary artery disease.