Skip to main content

Deep Learning for Radial SMS Myocardial Perfusion Reconstruction

Dynamic contrast enhanced MRI (DCE MRI) allows clinicians to better detect the downstream effects of coronary artery disease on myocardial tissue. DCE MRI of the heart is typically conducted through the acquisition of 2-4 short axis slices. However, this acquisition scheme provides incomplete coverage of the left ventricle and limits the visualization and quantification of the affected tissue. Radial simultaneous multi-slice (SMS) has been shown to improve DCE cardiac perfusion by providing complete coverage of the left ventricle. This comes with a cost, however: data undersampling and advanced reconstruction techniques significantly increase the computational time required to obtain images. To improve image quality and reduce the reconstruction time, we used a standard U-net architecture in order to learn the iterative compressed sensing reconstruction with total variation constraints.

deep learning
Figure 1. Illustration of the deep learning reconstruction of a gated radial SMS perfusion data not used in training. Truth corresponds to the iterative compressed sensing reconstructed images. Input to the network corresponds to undersampled sum of squares Inverse Fourier Transform images. Output images correspond to images learned by the neural network.  The deep learning framework took ~13 seconds to estimate artifacts for one SMS dataset in comparison to ~35 minutes for iterative compressed sensing reconstruction.