jerke123/mridc

Data Consistency for Magnetic Resonance Imaging

Data Consistency for Magnetic Resonance Imaging

---

❗This is a fork from the original MRIDC repo, expanding on 3 things:

• Made the RIM able to accept 3D data (for a batch size of 1). To use this, use the parameter consec_slices > 1.
• Included 2.5D convolutions for the (CI)RIM. This turns convolutional filter from a single 2D filter to 2 sequential 2D filters, one in-plane and one with the same dimensions in slice dimension and height. To use this feature, set dimensionality to 3.
• Included varialbe density poisson masking as an undersampling technique. To use this, use poisson2d as mask type.

Additionally, this repo contains Thesis material used and described in the paper for reference of implementation, aswell as the Thesis PDF document itself.

A comparison of the Poisson masking compared with conventional Gaussian datamasking:

(Row A: Poisson undersampled test. Row B: Gaussian undersampled test)

Data Consistency (DC) is crucial for generalization in multi-modal MRI data and robustness in detecting pathology.

This repo implements the following reconstruction methods:

• Cascades of Independently Recurrent Inference Machines (CIRIM) [1],
• Independently Recurrent Inference Machines (IRIM) [2, 3],
• End-to-End Variational Network (E2EVN), [4, 5]
• the UNet [5, 6],
• Compressed Sensing (CS) [7], and
• zero-filled reconstruction (ZF).

The CIRIM, the RIM, and the E2EVN target unrolled optimization by gradient descent. Thus, DC is implicitly enforced. Through cascades DC can be explicitly enforced by a designed term [1, 4].

Installation

You can install mridc with pip:

Pip

pip install mridc

From source

git clone https://github.com/wdika/mridc
cd mridc
./reinstall.sh

Usage

Recommended public datasets to use with this repo:

• fastMRI [5].

Documentation

Read the docs here

License

Citation

Check CITATION.cff file or cite using the widget. Alternatively cite as

@misc{mridc,
    author = {Karkalousos, Dimitrios and Caan, Matthan},
    title = {MRIDC: Data Consistency for Magnetic Resonance Imaging},
    year = {2021},
    url = {https://github.com/wdika/mridc},
}

Bibliography

[1] Karkalousos, D. et al. (2021) ‘Assessment of Data Consistency through Cascades of Independently Recurrent Inference Machines for fast and robust accelerated MRI reconstruction’. Available at: https://arxiv.org/abs/2111.15498v1 ( Accessed: 1 December 2021).

[2] Lønning, K. et al. (2019) ‘Recurrent inference machines for reconstructing heterogeneous MRI data’, Medical Image Analysis, 53, pp. 64–78. doi: 10.1016/j.media.2019.01.005.

[3] Karkalousos, D. et al. (2020) ‘Reconstructing unseen modalities and pathology with an efficient Recurrent Inference Machine’, pp. 1–31. Available at: http://arxiv.org/abs/2012.07819.

[4] Sriram, A. et al. (2020) ‘End-to-End Variational Networks for Accelerated MRI Reconstruction’, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 12262 LNCS, pp. 64–73. doi: 10.1007/978-3-030-59713-9_7.

[5] Zbontar, J. et al. (2018) ‘fastMRI: An Open Dataset and Benchmarks for Accelerated MRI’, arXiv, pp. 1–35. Available at: http://arxiv.org/abs/1811.08839.

[6] Ronneberger, O., Fischer, P. and Brox, T. (2015) ‘U-Net: Convolutional Networks for Biomedical Image Segmentation’, in Medical image computing and computer-assisted intervention : MICCAI ... International Conference on Medical Image Computing and Computer-Assisted Intervention, pp. 234–241. doi: 10.1007/978-3-319-24574-4_28.

[7] Lustig, M. et al. (2008) ‘Compressed Sensing MRI’, IEEE Signal Processing Magazine, 25(2), pp. 72–82. doi: 10.1109/MSP.2007.914728.