Gaviraghi, M;
Kanber, B;
Ricciardi, A;
Palesi, F;
Grussu, F;
Tur, C;
Calvi, A;
... Wheeler-Kingshott, CAMG; + view all
(2023)
A deep learning network from downsampled diffusion-weighted MRI k-space to image-space.
In:
Eighth National Congress of Bioengineering – Proceedings 2023.
Associazione Gruppo Nazionale Bioingegneria: Padova, Italy.
Preview |
Text
GNB_2023_paper_4743.pdf - Accepted Version Download (569kB) | Preview |
Abstract
Advanced Magnetic Resonance Imaging (MRI) techniques, such as Diffusion Weighed Imaging, usually require long acquisition times and an open challenge is to reduce the acquisition time more and more in order to allow their use in the clinical routine. Downsampling k-space is a way to speed up MRI, but this can generate artefacts in the resulting images when reconstructing them with standard Fourier transform methods. Here, we used deep learning to perform the inverse Fourier transform from k-space to the Diffusion Weighted (DW) images. and used it to assess the quality of images obtained from significantly reduced k-space acquisition strategies. The hypothesis is that a deep learning algorithm would preserve data quality, learned from the fully sampled k-space association. We tested our deep learning algorithm by reducing the number of acquired k-space rows by 30%, which would correspond to a total acquisition time reduction. We considered different types of k-space downsampling. All the trained networks were able to map the relationship between k-space and DW images, reducing artefacts. In conclusion, this work paves the way to designing acquisition strategies for fast diffusion imaging.
Archive Staff Only
 |
View Item |
