A model-guided method for improving coronary artery tree extractions from CCTA images.

Area of Science:

• Medical Imaging
• Cardiovascular Imaging
• Image Analysis

Background:

• Automatic extraction of coronary artery trees (CATs) from coronary computed tomography angiography (CCTA) images is crucial for clinical applications.
• Current automatic extraction methods often yield inaccuracies requiring manual correction, hindering clinical workflow.

Purpose of the Study:

• To develop and evaluate a model-guided method for automatically detecting and improving incorrect extractions in automatically generated CATs.
• To enhance the accuracy and clinical usability of CCTA-derived coronary artery models.

Main Methods:

• A coarse-to-fine approach was employed, applying initial improvements to all vessels and subsequent fine improvements to clinically significant vessels.
• A decision tree guided iterative improvement process until predefined stop criteria were met.
• The method was validated on 122 CCTA datasets after parameter optimization using 18 datasets.

Main Results:

• The model-guided method improved the average quality score of extracted CATs from 87±6 to 93±4 across 122 datasets.
• The improvement in extraction quality was negatively correlated with the initial extraction quality (R = -0.694, P < 0.001).
• The method achieved processing times under 2 minutes on a standard workstation.

Conclusions:

• The presented method effectively detects and corrects inaccuracies in automatically extracted CATs without compromising initial quality.
• The model-guided approach significantly enhances CAT extraction accuracy, achieving an average quality score of 93.
• This automated improvement holds promise for streamlining clinical interpretation of CCTA data.