This repository contains the code submission for the RRSG 2019 challenge to reproduce the results in
Pruessmann, K. P., Weiger, M. , Börnert, P. and Boesiger, P. (2001), Advances in sensitivity encoding with arbitrary k‐space trajectories. Magn. Reson. Med., 46: 638-651.
- Primal-Dual-Toolbox built with gpuNUFFT.
- BART toolbox
- medutils
- python3 (Anaconda)
To install the Primal-Dual-Toolbox with gpuNUFFT and the BART toolbox, please follow the instructions on the corresponding github pages. The medutils package can be installed as follows.
pip install git+https://github.com/khammernik/medutils.git
The code was tested on Ubuntu 16.04, Cuda 9.2 and a python3 anaconda environment.
The challenge dataset contains radial brain and heart data, which will be downloaded automatically when the script is executed. The non-uniform Fourier transform is computed using the gpuNUFFT. The Primal-Dual-Toolbox provides an interface to connect the gpuNUFFT with python. Coil sensitivity maps are estimated using the BART toolbox toolbox. The CG SENSE optimizer is implemented in the medutils package.
To perform the required tasks run
./run_recon.sh
which will generate subfolders containing the reconstructions as h5 file and additional plots. The required figures for the challenge are saved to the root folder.
Figure 1: Brain reconstructions for the acceleration factors R 1, 2, 3 and 4.
Figure 2: Convergence of the true image error for the brain data
Figure 3: Convergence of the tolerance for the brain data
Figure 4: Heart reconstructions for a different number of projections (55, 33, 22, 11).
Kerstin Hammernik
Institute of Computer Graphics and Vision,
Graz University of Technology
[email protected]