The Code is created based on the method described in the following papers:
[1] A comparative study of CNN-based super-resolution methods in MRI reconstruction and its beyond,
W. Zeng, J. Peng, S. Wang, Q. Liu, Signal Processing: Image Communication, vol. 81, 2020.
https://doi.org/10.1016/j.image.2019.115701.

[2] A Comparative Study of CNN-based Super-resolution Methods in MRI Reconstruction,
W. Zeng, J. Peng, S. Wang, Z. Li, Q. Liu, D. Liang,IEEE 16th ISBI, 2019.

The progress of convolution neural network (CNN) based Super-resolution (SR) has shown its potential in image processing community. Meanwhile, Compressed Sensing MRI (CS-MRI) provides the possibility to accelerate the traditional acquisition process of MRI. In this work, on the basis of decomposing the cascade network to be a series of alternating CNN-based sub-network and data-consistency sub-network, we investigate the performance of the cascade networks in CS-MRI by employing various CNN-based super-resolution methods in the CNN-based sub-network. Furthermore, realizing that existing methods only explore dense connection in the CNN-based sub-network which insufficiently explore the feature information, we propose a dense connected cascade network (DCCN) for more accurate MR reconstruction. Specifically, DCCN network densely connects both CNN-based sub-network and data-consistency sub-network, thus takes advantage of the data-consistency of k-space data in a densely connected fashion. Experimental results on various MR data demonstrated that DCCN is superior to current cascade networks in reconstruction quality.

Reconstruction results by various methods at 90% Pseudo radial undersampling. Red boxes illustrate the enlarged view. From left to right: Fully-sampled MRI image, udersampling MRI image, NLR-CS, CN-CNN, CN-DenseNet, CN-EDSR, CN-RDN and DCCN.

Reconstruction results by various methods at 85% 2D random undersampling. Red boxes illustrate the enlarged view. From left to right: Fully-sampled MRI image, udersampling MRI image, NLR-CS, CN-CNN, CN-DenseNet, CN-EDSR, CN-RDN and DCCN.

Reconstruction results by various methods at 85% 1D Cartesian undersampling. Red boxes illustrate the enlarged view. From left to right: Fully-sampled MRI image, udersampling MRI image, NLR-CS, CN-CNN, CN-DenseNet, CN-EDSR, CN-RDN and DCCN.

theano
cuda
cudnn
python3

'python DCCN.py' is the demo of DCCN.

@article{zeng2019isbi,   
title=A Comparative Study of CNN-based Super-resolution Methods in MRI Reconstruction},   
author={Wei Zeng, Jie Peng, Shanshan Wang, Zhicheng Li, Qiegen Liu, Dong Liang},   
conference={IEEE 16th International Symposium on Biomedical Imaging},   
year={2019},   
}

• Variable augmented neural network for decolorization and multi-exposure fusion [Paper] [Code] [Slide]

• IFR-Net: Iterative Feature Refinement Network for Compressed Sensing MRI [Paper] [Code]

• Iterative scheme-inspired network for impulse noise removal [Paper] [Code]

• Progressively distribution-based Rician noise removal for magnetic resonance imaging [Paper1] [Paper2] [Code]

• VST-Net: Variance-stabilizing transformation inspired network for Poisson denoising [Paper] [Code]

• Complex-valued MRI data from SIAT [Data]