Keywords: MR Spectroscopy, Geometric Deep Learning, U-Net
Abstract: Geometric deep learning has recently gained influence, as it allows the extension of convolutional neural networks to non euclidean domains. In this paper graph neural networks (GNNs) are used for reconstruction and coil combination of undersampled concentric-ring k-space MRSI data. We show that graph U-nets perform better on undersampled data than GNNs. Specifically, results suggest that the omission of self-connecting edges results in a more stable behavior and better training for graph U-nets.
Paper Type: methodological development
Primary Subject Area: Image Acquisition and Reconstruction
Secondary Subject Area: Application: Radiology
Paper Status: original work, not submitted yet
Source Code Url: https://github.com/weiserjpaul/Graph-U-Net
Data Set Url: We are very sorry, we had unforeseen technical difficulties. The data will be uploaded later on.
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.