Motion artifact recognition and quantification in coronary CT angiography using convolutional neural networks.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Imaging

Background:

• Diagnostic non-invasive coronary CT angiography (CCTA) requires excellent image quality.
• Cardiac motion artifacts can compromise CCTA diagnostic accuracy and treatment planning.

Purpose of the Study:

• To develop deep learning-based methods for recognizing and quantifying coronary motion artifacts in CCTA.
• To assess the diagnostic reliability and image quality of CCTA using these novel measures.
• To enable the application, steering, and evaluation of motion compensation algorithms.

Main Methods:

• A Coronary Motion Forward Artifact model for CT data (CoMoFACT) was developed to simulate motion artifacts.
• CoMoFACT generated training data from 17 prospectively ECG-triggered clinical cases with controlled motion levels.
• Convolutional neural networks (CNNs) were trained for artifact classification and motion level regression.

Main Results:

• CNNs achieved 93.3% ± 1.8% accuracy in classifying motion-free versus motion-perturbed coronary image patches.
• A regression network predicted the target motion level with a mean absolute error of 1.12 ± 0.07.
• The methods demonstrated transferability and generalization on CCTA cases with real motion artifacts.

Conclusions:

• Deep learning effectively recognizes and quantifies coronary motion artifacts in CCTA.
• These quantitative measures improve diagnostic reliability and image quality assessment.
• The developed approach supports the optimization of motion compensation strategies in CCTA.