Detail-preserving multi-scale deep learning reconstruction for cardiac magnetic resonance imaging

Fast data acquisition and high-quality image reconstruction are vital for dynamic MRI, which can capture both anatomical and temporal information. High-resolution acquisition approaches in k-space and super-resolution approaches after reconstruction have been frequently reported. However, these methods may get details lost at high acceleration factors. To address this issue, we propose a multi-scale detail preserving reconstruction method for dynamic MR images. The residuals of multi-scale intermediate images in the iterative procedure are explored and the temporal and spatial dependencies between frames are considered. Promising results are achieved by the proposed method at the high acceleration factor of 11.