MR image reconstruction using deep density priors

Kerem C. Tezcan, Christian F. Baumgartner, Roger Luechinger, Klaas P. Pruessmann, Ender Konukoglu

Keywords: MRI, image reconstruction, density estimation, VAE

TL;DR: We use a VAE to learn the distribution of fully sampled MR images and then use this VAE as the prior for undersampled MR image reconstruction.

Abstract: We present a recently published work on undersampled MR image reconstruction (Tezcan et al., 2018) relying on deep learning (DL). The method uses a variational autoencoder trained on fully sampled images as the prior in a maximum a posteriori formulation of the reconstruction problem. Doing this allows decoupling the prior from the encoding, i.e. undersampling scheme and coil setting, allowing using the same network with any encoding without retraining, an aspect not guaranteed for any other reconstruction method using DL. Results indicate highly competitive performance

Code Of Conduct: I have read and accept the code of conduct.

Community Implementations: [ 1 code implementation](https://www.catalyzex.com/paper/mr-image-reconstruction-using-deep-density/code)