Generation of Realistic MR Elastography Brain Stiffness Map Mimics using a 3D Conditional Generative Adversarial Network

Motivation: Existing MRE datasets are limited in size, limiting their use in computational models. Augmented datasets could allow for increased application of MRE images. Goal(s): Our goal was to use a generative adversarial network to produce artificial brain stiffness maps which could be controlled to appear like true images from younger and older adults. Approach: A generative adversarial network was trained on a dataset of neurologically healthy subjects to create a model capable of generating artificial brain stiffness maps. Results: The generated images demonstrated similar characteristics to true images from the corresponding age group. Impact: This work demonstrates that a generative adversarial network can produce realistic brain stiffness images. Improvements to this technique will allow for these images to be used alongside true MRE images to support computational modeling efforts which utilize brain stiffness information.