Meeting Banner
Abstract #3193

On the usage of deep neural networks as a tensor-to-tensor translation between MR measurements and electrical properties

Ilias Giannakopoulos1,2, José Serallés2, Georgy Guryev1,2, Luca Daniel2, Elfar Adalsteinsson2,3, Lawrence Wald4,5,6, Daniel Sodickson7,8,9, Jacob White2, and Riccardo Lattanzi7,8,9
1Skoltech Center for Computational and Data-Intensive Science and Engineering, Skolkovo Institute of Science and Technology, Moscow, Russian Federation, 2Department of Electrical Engineering & Computer Science, Massachusetts Institute of Technology, Cambridge, MA, United States, 3Institute for Medical Engineering and Science, Cambridge, MA, United States, 4Department of Radiology, Massachusetts General Hospital, Charlestown, MA, United States, 5Department of Radiology, Harvard Medical School, Boston, MA, United States, 6Harvard-MIT Health Sciences and Technology, Cambridge, MA, United States, 7Center for Advanced Imaging Innovation and Research (CAI2R), Department of Radiology, New York University School of Medicine, New York, NY, United States, 8The Bernard and Irene Schwartz Center for Biomedical Imaging (CBI), Department of Radiology, New York University School of Medicine, New York, NY, United States, 9Sackler Institute of Graduate Biomedical Sciences, New York University School of Medicine, New York, NY, United States

Electrical properties (EP) can be retrieved from magnetic resonance measurements. We employed numerical simulations to investigate the use of convolutional neural networks (CNN) as a tensor-to-tensor translation between transmit magnetic field pattern ($$$b_1^+$$$) and EP distribution for simple tissue-mimicking phantoms. Given the volumetric nature of the problem, we chose a 3D UNET and trained the network on $$$10000$$$ data. We investigated on the usage of regularization to account for overfitting and observed that multiple dropouts through the layers of the network yield optimal EP reconstructions for $$$1000$$$ testing data.


This abstract and the presentation materials are available to 2020 meeting attendees and eLibrary customers only; a login is required.

Join Here