Cross-anatomy Inference of Deep Learning-based MRI Acceleration Methods

Aleksandr Belov, Joël Valentin Stadelmann, Dmitry V. Dylov

Keywords: Fast MRI, Superresolution, Image-to-image translation

TL;DR: We report efficient inference models between anatomies (brain and knee) when k-space data is either undersampled or underresolved.

Abstract: MRI scanners capture images of excellent soft-tissue contrast but involve long acquisition times, which can be mitigated by acquiring undersampled or low-resolution k-space data. Although image-to-image translation and superresolution techniques can correct the artifacts caused by the incomplete k-space, they are known to be anatomy-specific, requiring validation and train sets to belong to the same body part. This short paper covers the causes of the lowered reconstruction quality in the cross-anatomy inference task and provides suggestions for their compensation.

Paper Type: validation/application paper

Primary Subject Area: Image Acquisition and Reconstruction

Secondary Subject Area: Transfer Learning and Domain Adaptation

Paper Status: based on accepted/submitted journal paper

Source Code Url: At this point, the authors do not have authorization from the owners of the IP rights to publish the source code.

Data Set Url: https://fastmri.org/dataset/

Registration: I acknowledge that publication of this at MIDL and in the proceedings requires at least one of the authors to register and present the work during the conference.

Authorship: I confirm that I am the author of this work and that it has not been submitted to another publication before.