New models addition along with U-Net

[Bala93]

Thanks for this repository.

The repo has mainly used U-Net as the base network for MRI reconstruction. I was thinking if we could add base networks which has shown promising results like the following:

• L. Sun, Z. Fan, Y. Huang, X. Ding, and J. Paisley, “Compressed sensing MRI using a recursive dilated network,” 32nd AAAI Conf. Artif. Intell. AAAI 2018, pp. 2444–2451, 2018.
• H. Wu, Y. Wu, L. Sun, C. Cai, Y. Huang, and X. Ding, “A deep ensemble network for compressed sensing MRI,” in Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 2018, vol. 11301 LNCS, pp. 162–171, doi: 10.1007/978-3-030-04167-0_15.
• R. Souza and R. Frayne, “A Hybrid Frequency-domain/Image-domain Deep Network for Magnetic Resonance Image Reconstruction,” no. Dc, pp. 1–8, 2018, [Online].

I have the proper PyTorch implementations for these methods. I could raise a PR. From my experience, this repository serves as the base for all the works related to MRI reconstruction. Should we think of adding new features ? Please let me know your views.

Thank you.

[mmuckley]

Hello @Bala93, we'd love to have more models. That being said, the repository is focused around the fastMRI dataset, so we should probably focus on models that either

1. Push top metrics on the fastMRI dataset. They don't have to be SotA but they should be in the range of the baseline models.
2. Introduce new and interesting perspectives using the dataset.

Do you know of downstream papers that used these models on fastMRI? I looked at the Sun and Souza papers and it looked like neither one was focused on fastMRI. I wasn't able to read the Wu paper, but it doesn't seem extremely well-cited.

A couple of good papers fitting the above criteria might be

• Eo, Taejoon, et al. "KIKI‐net: cross‐domain convolutional neural networks for reconstructing undersampled magnetic resonance images." Magnetic resonance in medicine 80.5 (2018): 2188-2201.
  • Basis for several top models submitted to fastMRI competitions.
• Yu, Songhyun, Bumjun Park, and Jechang Jeong. "Deep iterative down-up cnn for image denoising." Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops. 2019.
  • Not actually used on fastMRI but it's been a building block for competition models.

I'm sure there are others out there.