Dave Van Veen1,2, Arjun Desai1, Reinhard Heckel3,4, and Akshay S. Chaudhari1
1Stanford University, Stanford, CA, United States, 2University of Texas at Austin, Austin, TX, United States, 3Rice University, Houston, TX, United States, 4Technical University of Munich, Munich, Germany
1Stanford University, Stanford, CA, United States, 2University of Texas at Austin, Austin, TX, United States, 3Rice University, Houston, TX, United States, 4Technical University of Munich, Munich, Germany
We find that untrained convolutional neural networks are comparable to supervised methods for accelerating MRI scans in both 2D and 3D. Further, we demonstrate a method for regularizing the network feature maps using undersampled k-space measurements.
Figure 4. Comparison of untrained method against fully supervised baseline on a 3D qDESS axial slice. The k-space was downsampled using a 2D Poisson disc at acceleration factors of 4x and 8x for the top and bottom rows, respectively. The image quality of the reconstruction with the same undersampling factors is higher for 3D qDESS than 2D fastMRI data, likely because the undersampling aliasing is spread across two dimensions ($$$k_y$$$ and $$$k_z$$$).
Figure 1. Network architecture for our untrained method.