Meeting Banner
Abstract #1832

Dipole inversion by recurrent inference for quantitative susceptibility mapping

Samy Abo Seada1, Emanoel Ribeiro Sabidussi1, Sebastian Weingärtner2, Dirk H. J. Poot1, and Juan Antonio Hernandez-Tamames1
1Department of Radiology and Nuclear Medicine, Erasmus MC, Rotterdam, Netherlands, 2Department of Imaging Physics, TU Delft, Delft, Netherlands

QSM dipole inversion remains a challenge and recent machine learning approaches have not incorporated the known forward model directly. We propose using recurrent inference machines (RIM), a type of unrolled optimization technique, which are proposed for solving iterative inverse problems specifically. RIMs enable incorporating the forward dipole convolution directly in the learning process. Simulated data was used for training. The QSM reconstruction was tested on simulated data and healthy subject data acquired at 3T.

This abstract and the presentation materials are available to members only; a login is required.

Join Here