Deep learning based on u-shaped architectures has been successfully used as a means for the dipole inversion crucial to Quantitative Susceptibility Mapping (QSM). In the present work we propose a novel deep regression network by stacking two u-shaped networks and consequently both, the background field removal and the dipole inversion can be performed in a single feed forward network architecture. Based on learning the theoretical forward model using synthetic data examples, we show a proof-of-concept for solving the background field problem and dipole inversion in a single end-to-end trained network using in vivo Magnetic Resonance Imaging (MRI) data.
This abstract and the presentation materials are available to members only; a login is required.