Deep learning-based coronary artery motion estimation and compensation for short-scan cardiac CT

Area of Science:

• Medical Imaging
• Cardiovascular Imaging
• Artificial Intelligence in Medicine

Background:

• Cardiac CT scans are susceptible to motion artifacts due to heart movement.
• These artifacts can impair the assessment of small structures like coronary arteries.
• Existing compensation methods may lack efficiency and accuracy.

Purpose of the Study:

• To introduce Deep PAMoCo, a novel deep learning-based method for motion compensation in cardiac CT.
• To estimate and correct coronary artery motion artifacts.
• To enhance the diagnostic value of cardiac CT imaging.

Main Methods:

• Deep PAMoCo utilizes partial angle reconstructions (PARs) by segmenting short scan data.
• PARs are deformed using a motion vector field (MVF) estimated by a deep neural network.
• Compensated PARs are summed to produce a motion-free reconstruction.

Main Results:

• Deep PAMoCo effectively removes motion artifacts, reducing CT value errors from ~300 HU to ~25 HU.
• Performance is independent of contrast, radius, and motion amplitude.
• Clinical scans show superior accuracy and processing time compared to state-of-the-art methods.

Conclusions:

• Deep PAMoCo offers an efficient solution for motion artifact reduction in cardiac CT.
• The method significantly improves the diagnostic quality of cardiac CT scans.
• It is particularly beneficial for scans with substantial motion corruption.