[Development of the DSA Method for Coronary Angiography Using Deep Learning Techniques]

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Diagnostics

Background:

• Digital Subtraction Angiography (DSA) is crucial for diagnosing coronary artery disease but is hindered by motion artifacts from patient breathing and heartbeats.
• Current DSA techniques are not typically applied to coronary arteries due to these artifacts, limiting diagnostic accuracy and pathogenesis research.
• Advancements in deep learning offer potential solutions for overcoming these imaging challenges.

Purpose of the Study:

• To develop and evaluate a novel DSA method specifically for coronary arteries utilizing deep learning techniques.
• To mitigate motion artifacts inherent in coronary DSA imaging.
• To enhance the visibility of coronary vessels for improved diagnostic capabilities.

Main Methods:

• A convolutional neural network (CNN)-based model was developed and trained on 21,025 image patches from 29 coronary angiogram cases.
• Mask images were generated by inputting live images into the trained CNN model.
• DSA images were obtained by subtracting the derived mask images from the contrast-enhanced live images.

Main Results:

• Both subjective (human observation) and objective (pixel value standard deviation) evaluations confirmed the effectiveness of the proposed DSA method.
• The developed technique successfully reduced motion artifacts caused by cardiac and respiratory movements.
• Vessel visibility in coronary DSA images was significantly improved.

Conclusions:

• New deep learning-based DSA techniques for coronary arteries have been successfully developed.
• The proposed method effectively reduces motion artifacts and enhances vessel visibility, making coronary DSA feasible in clinical settings.
• This technique holds potential for application to other organs, enabling DSA imaging without requiring patients to hold their breath.