Real-time spiral phase contrast MR is practical for free-breathing assessment of flow. However, for sufficient spatial and temporal resolutions it requires high acceleration rates leading to long reconstruction times. Here we propose to train a 3D U-Net with complex convolutions to accurately reconstruct phase data and flow curves from highly undersampled data. Prospectively acquired in-vivo data were reconstructed with similar image quality but ~4.6x faster than compressed sensing reconstructions which could improve workflow.
This abstract and the presentation materials are available to members only; a login is required.