A deep learning-based automated algorithm for labeling coronary arteries in computed tomography angiography images

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Cardiovascular Imaging

Background:

• Coronary artery labeling in CCTA is crucial for diagnosis.
• Manual labeling is time-consuming and prone to inter-observer variability.

Purpose of the Study:

• To develop and validate a fully automated algorithm for coronary artery labeling in CCTA images.
• To assess the accuracy and efficiency of the automated labeling system.

Main Methods:

• Utilized two 3D U-Net architectures for myocardium segmentation.
• Employed a distance transformation algorithm for left circumflex artery delineation.
• Developed an automated algorithm for labeling 16 coronary artery segments.

Main Results:

• Achieved 96.2% accuracy for label presence.
• Demonstrated an average overlap of 94.0% with expert labels.
• Reported an average inter-expert agreement rate of 95.0%.

Conclusions:

• The proposed deep learning algorithm demonstrates high accuracy for automated coronary artery labeling.
• This method shows promise in reducing the burden on radiologists.
• The algorithm offers a reliable tool for CCTA image analysis.