Motion estimation and correction in cardiac CT angiography images using convolutional neural networks

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare
• Cardiovascular Imaging

Background:

• Cardiac motion artifacts significantly degrade coronary computed tomography angiography (CCTA) image quality.
• This degradation can lead to misinterpretations and hinder the diagnosis of coronary artery disease (CAD).

Purpose of the Study:

• To present a novel motion compensation (MC) approach, CoMPACT (Coronary Motion estimation by Patch Analysis in CT data).
• To improve the interpretability of CCTA images affected by cardiac motion.

Main Methods:

• Simulated cardiac motion using CoMoFACT (Coronary Motion Forward Artifact model for CT data) on clinical CCTA data.
• Training convolutional neural networks (CNNs) to estimate 2D motion vectors from 2.5D image patches.
• Integrating CNNs into an iterative MC pipeline with distance-weighted motion vector extrapolation.

Main Results:

• CNNs achieved accurate motion direction and magnitude prediction in phantom and clinical studies.
• The CoMPACT pipeline significantly reduced artifact levels in clinical cases with real cardiac motion.
• Observer studies showed a reduction in mean artifact levels from 3.08 to 2.28 on a 5-point Likert scale with CoMPACT MC.

Conclusions:

• CoMPACT effectively reduces cardiac motion artifacts in CCTA.
• The proposed method enhances image interpretability, aiding in the diagnosis of CAD.
• AI-driven motion compensation offers a promising solution for improving cardiac CT image quality.