@article{Yu,
author = {Yu, S and Dong, H and Yang, G and Slabaugh, G and Dragotti, PL and Ye, X and Liu, F and Arridge, S and Keegan, J and Firmin, D and Guo, Y},
title = {Deep De-Aliasing for Fast Compressive Sensing MRI},
url = {http://arxiv.org/abs/1705.07137v1},
}

Download

TY  - JOUR
AB  - Fast Magnetic Resonance Imaging (MRI) is highly in demand for many clinicalapplications in order to reduce the scanning cost and improve the patientexperience. This can also potentially increase the image quality by reducingthe motion artefacts and contrast washout. However, once an image field of viewand the desired resolution are chosen, the minimum scanning time is normallydetermined by the requirement of acquiring sufficient raw data to meet theNyquist-Shannon sampling criteria. Compressive Sensing (CS) theory has beenperfectly matched to the MRI scanning sequence design with much less requiredraw data for the image reconstruction. Inspired by recent advances in deeplearning for solving various inverse problems, we propose a conditionalGenerative Adversarial Networks-based deep learning framework for de-aliasingand reconstructing MRI images from highly undersampled data with great promiseto accelerate the data acquisition process. By coupling an innovative contentloss with the adversarial loss our de-aliasing results are more realistic.Furthermore, we propose a refinement learning procedure for training thegenerator network, which can stabilise the training with fast convergence andless parameter tuning. We demonstrate that the proposed framework outperformsstate-of-the-art CS-MRI methods, in terms of reconstruction error andperceptual image quality. In addition, our method can reconstruct each image in0.22ms--0.37ms, which is promising for real-time applications.
AU  - Yu,S
AU  - Dong,H
AU  - Yang,G
AU  - Slabaugh,G
AU  - Dragotti,PL
AU  - Ye,X
AU  - Liu,F
AU  - Arridge,S
AU  - Keegan,J
AU  - Firmin,D
AU  - Guo,Y
TI  - Deep De-Aliasing for Fast Compressive Sensing MRI
UR  - http://arxiv.org/abs/1705.07137v1
ER  - 

Download