eduzhai > Applied Sciences > Engineering >

MRI Image Reconstruction via Learning Optimization Using Neural ODEs

  • Save

... pages left unread,continue reading

Document pages: 11 pages

Abstract: We propose to formulate MRI image reconstruction as an optimization problemand model the optimization trajectory as a dynamic process using ordinarydifferential equations (ODEs). We model the dynamics in ODE with a neuralnetwork and solve the desired ODE with the off-the-shelf (fixed) solver toobtain reconstructed images. We extend this model and incorporate the knowledgeof off-the-shelf ODE solvers into the network design (learned solvers). Weinvestigate several models based on three ODE solvers and compare models withfixed solvers and learned solvers. Our models achieve better reconstructionresults and are more parameter efficient than other popular methods such asUNet and cascaded CNN. We introduce a new way of tackling the MRIreconstruction problem by modeling the continuous optimization dynamics usingneural ODEs.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×