Deep learning-based Motion-corrected Image Reconstruction for High-resolution Spiral First-pass Myocardial Perfusion Imaging

Cardiovascular magnetic resonance (CMR) is susceptible to motion-induced artifacts from cardiac and respiratory motion, leading to poor image quality. The inter-frame motion artifacts make quantitative analysis for cardiac function evaluation difficult. Hence motion correction is an important pre-processing step before robust quantification of myocardial perfusion. We developed a deep learning-based framework for rapid and accurate motion correction of CMR perfusion imaging using a 2D U-Net that estimates the deformation field from a moving frame to a fixed frame.