Deep learning segmentation of major vessels in X-ray coronary angiography

Area of Science:

• Cardiovascular Imaging
• Medical Image Analysis
• Artificial Intelligence in Medicine

Background:

• X-ray coronary angiography is crucial for diagnosing coronary artery disease.
• Quantitative coronary angiography (QCA) offers objective measures but requires extensive training for vessel segmentation.
• Current computer-aided methods often need manual correction for accurate coronary vessel segmentation.

Purpose of the Study:

• To develop a robust deep learning method for automated major coronary vessel segmentation in X-ray coronary angiography.
• To improve the accuracy and efficiency of quantitative coronary angiography (QCA) analysis.

Main Methods:

• Utilized fully convolutional deep learning networks for major vessel segmentation.
• Trained and tested the model on a large dataset of 3302 diseased major vessels from 2042 patients.
• Validated the model's performance on an external dataset with varying image characteristics.

Main Results:

• Achieved an average F1 score of 0.917 for accurate vessel segmentation.
• 93.7% of images showed high accuracy (F1 score > 0.8).
• Demonstrated real-time prediction with minimal preprocessing and high connectivity in capturing stenosis.

Conclusions:

• The proposed deep learning method provides robust and accurate major vessel segmentation in X-ray coronary angiography.
• Automated segmentation facilitates QCA analysis, potentially improving diagnostic efficiency for coronary artery disease.
• This approach shows promise for real-time clinical applications in cardiovascular imaging.