eduzhai > Applied Sciences > Engineering >

Learning Bloch Simulations for MR Fingerprinting by Invertible Neural Networks

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 10 pages

Abstract: Magnetic resonance fingerprinting (MRF) enables fast and multiparametric MRimaging. Despite fast acquisition, the state-of-the-art reconstruction of MRFbased on dictionary matching is slow and lacks scalability. To overcome theselimitations, neural network (NN) approaches estimating MR parameters fromfingerprints have been proposed recently. Here, we revisit NN-based MRFreconstruction to jointly learn the forward process from MR parameters tofingerprints and the backward process from fingerprints to MR parameters byleveraging invertible neural networks (INNs). As a proof-of-concept, we performvarious experiments showing the benefit of learning the forward process, i.e.,the Bloch simulations, for improved MR parameter estimation. The benefitespecially accentuates when MR parameter estimation is difficult due to MRphysical restrictions. Therefore, INNs might be a feasible alternative to thecurrent solely backward-based NNs for MRF reconstruction.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×