Deep convolutional neural networks for accelerated dynamic magnetic resonance imaging

Dynamic magnetic resonance imaging (MRI) scans can be accelerated by utilizing compressed sensing (CS) reconstruction methods that allow for diagnostic quality images to be generated from undersampled data. Unfortunately, CS reconstruction is time-consuming, requiring hours between a dynamic MRI scan and image availability for diagnosis. In this work, we train a convolutional neural network (CNN) to perform fast reconstruction of severely undersampled dynamic cardiac MRI data, and we explore the utility of CNNs for further accelerating dynamic MRI scan times. Compared to state-of-the-art CS reconstruction techniques, our CNN achieves reconstruction speeds that are 150x faster without significant loss of image quality. Additionally, preliminary results suggest that CNNs may allow scan times that are 2x faster than those allowed by CS.