Cagan Alkan1, Morteza Mardani1, Shreyas S. Vasanawala1, and John M. Pauly1
1Stanford University, Stanford, CA, United States
1Stanford University, Stanford, CA, United States
Variational information maximization enables end-to-end optimization of MR data sampling and reconstruction in a data-driven manner. Our framework improves the reconstruction quality as measured by the 1.4-2.2dB increase in pSNR compared to the variable density baseline.
Figure 4: Reconstruction result from a slice in the test set for $$$R=5$$$ and $$$R=10$$$. Zero-filled reconstruction corresponds to applying $$$A_\phi^H$$$ on $$$z$$$ directly.
Figure 1: Network architecture consisting of nuFFT based encoder (a) and unrolled reconstruction network (b, c). Sampling locations $$$\phi$$$ are shared between encoder and decoder.