---
language: en

license: bsd-3-clause

tags:
- MRI image priors
- Generative models
- Diffusion models
- TensorFlow
- PixelCNN
---

## Generative pretrained models on MRI images.

It was introduced in [this paper](https://) and first released at [this page](https://github.com/mrirecon/image-priors). The prior distribution of MRI images learned with generative models has proven to be effective in MRI image reconstruction. Here, we include four PixelCNN models and two diffusion models, one is SMLD and the another one is DDPM. These models are trained with [spreco](https://github.com/mrirecon/spreco). For more details on how these models were trained, please find them in the paper.

| Prior                         | Model     | Phase     | Size                 | Contrast                                                                           | Subscript             |
|-------------------------------|-----------|-----------|----------------------|------------------------------------------------------------------------------------|-----------------------|
| \\(\texttt{P}_\mathrm{SC}\\)  | PixelCNN  | preserved | 1000                 | T1, T2, T2-FLAIR, \\(\texttt{T}^*_\mathrm{2}\\) | SC - Small, complex   |
| \\(\texttt{P}_\mathrm{SM}\\)  | PixelCNN  | unknown   | 1000                 | T1, T2, T2-FLAIR, \\(\texttt{T}^*_\mathrm{2}\\) | SM - Small, magnitude |
| \\(\texttt{P}_\mathrm{LM}\\)  | PixelCNN  | unknown   | ~20000 | MPRAGE                                                                             | LM - Large, magnitude |
| \\(\texttt{P}_\mathrm{LC}\\)  | PixelCNN  | generated | ~20000 | MPRAGE                                                                             | LC - Large, complex   |
| \\(\texttt{D}_\mathrm{SC}\\)  | Diffusion | generated | ~80000 | MPRAGE                                                                             | SC - SMLD, complex    |
| \\(\texttt{D}_\mathrm{PC}\\)  | Diffusion | generated | ~80000 | MPRAGE                                                                             | PC - DDPM, complex    |


## How to use
The Berkeley Advanced Reconstruction Toolbox ([BART](https://mrirecon.github.io/bart/)) toolbox provides many functionalities for MRI image reconstruction. It introduced the application of Tensorflow graph as regularization in [Deep learning with BART](https://doi.org/10.1002/mrm.29485) and there is a [colab notebook](https://colab.research.google.com/github/mrirecon/bart-workshop/blob/master/ismrm2021/bart_tensorflow/bart_tf.ipynb) where you can give quickstart with it. For the codes to evaluate above priors, please find them in this [repository](https://github.com/mrirecon/image-priors).

## BibTex entry and citation info

```bibtex
@article{Luo2023,
  title={Generative Pretrained Image Priors for MRI Reconstruction},
  author={xxx},
  year={2023}
}

@Article{Luo_Magn.Reson.Med._2023,
  author    = {Guanxiong Luo and Moritz Blumenthal and Martin Heide and Martin Uecker},
  title     = {Bayesian {MRI} reconstruction with joint uncertainty
estimation using diffusion models},
  doi       = {10.1002/mrm.29624},
  number    = {4},
  pages     = {1--17},
  volume    = {84},
  journal   = {Magn. Reson. Med.},
  month     = {apr},
  publisher = {Wiley},
  year      = {2023},
}

@article{Blumenthal_Magn.Reson.Med._2023,
author       = {Blumenthal, Moritz and Luo, Guanxiong and Schilling, Martin and Holme, H. Christian M. and Uecker, Martin},
title        = {Deep, deep learning with BART},
journal      = {Magnetic Resonance in Medicine},
volume       = {89},
number       = {2},
pages        = {678-693},
keywords     = {automatic differentiation, deep learning, image reconstruction, inverse problems, MRI, parallel imaging},
doi          = {https://doi.org/10.1002/mrm.29485},
url          = {https://onlinelibrary.wiley.com/doi/abs/10.1002/mrm.29485},
eprint       = {https://onlinelibrary.wiley.com/doi/pdf/10.1002/mrm.29485},
abstract     = {Purpose To develop a deep-learning-based image reconstruction framework for reproducible research in MRI. Methods The BART toolbox offers a rich set of implementations of calibration and reconstruction algorithms for parallel imaging and compressed sensing. In this work, BART was extended by a nonlinear operator framework that provides automatic differentiation to allow computation of gradients. Existing MRI-specific operators of BART, such as the nonuniform fast Fourier transform, are directly integrated into this framework and are complemented by common building blocks used in neural networks. To evaluate the use of the framework for advanced deep-learning-based reconstruction, two state-of-the-art unrolled reconstruction networks, namely the Variational Network and MoDL, were implemented. Results State-of-the-art deep image-reconstruction networks can be constructed and trained using BART's gradient-based optimization algorithms. The BART implementation achieves a similar performance in terms of training time and reconstruction quality compared to the original implementations based on TensorFlow. Conclusion By integrating nonlinear operators and neural networks into BART, we provide a general framework for deep-learning-based reconstruction in MRI.},
year         = {2023}
}

@Article{Luo_Magn.Reson.Med._2020,
  author    = {Guanxiong Luo and Na Zhao and Wenhao Jiang and Edward S. Hui and Peng Cao},
  title     = {{MRI} reconstruction using deep Bayesian estimation},
  doi       = {https://doi.org/10.1002/mrm.28274},
  number    = {4},
  pages     = {2246--2261},
  volume    = {84},
  journal   = {Magn. Reson. Med.},
  month     = {apr},
  publisher = {Wiley},
  year      = {2020},
}

```