One AI Model for Multi-scenario Reconstructions: Physics-informed Synthetic Data Boosts Generalizable Deep Learning for Fast MRI
This work presents a Physics-Informed Synthetic data learning framework for Fast MRI, called PISF, which is the first to enable generalizable DL for multi-scenario MRI reconstruction using solely one trained model.
- We demonstrate that training DL models on synthetic data, integrated with enhanced learning techniques, can achieve comparable or even better in vivo MRI reconstruction compared to models trained on a matched realistic dataset—PISF reduces the demand for real-world MRI data by up to 96%.
- Our PISF shows impressive generalizability in multi-vendor multi-center imaging—it can reconstruct high-quality images of 4 anatomies and 5 contrasts across 5 vendors and centers using a single trained network.
- PISF’s superior adaptability to patients has been verified through 10 experienced doctors’ evaluations (4 neuro radiologists and 1 neurosurgeon for brain tumor patients, and 3 cardiac radiologists and 2 cardiologists for myocardial hypertrophy patients)—its overall image quality steps into the excellent level in reader study.
In summary, PISF provides a feasible and cost-effective way to markedly boost the widespread usage of DL in various fast MRI applications, while freeing from the intractable ethical and practical considerations of in vivo human data acquisitions.
The preprint paper can be seen at https://doi.org/10.48550/arXiv.2307.13220.
Email: Xiaobo Qu (quxiaobo@xmu.edu.cn) CC: Zi Wang (wangzi1023@stu.xmu.edu.cn)
Homepage: http://csrc.xmu.edu.cn
The testing codes of PISF are released here.
Install conda environment:
conda env create -f environment
Run the main code for reconstruction:
python PISF_Recon_Enhance.py
Python environment should be: python=3.6.13, pytorch=1.10.1
Implementation tips: If you want to test on your own collected data, they should be stored in the same format as the demo data we provided and be 8-coil or compressed/extended to 8-coil.
Data availability: All used public datasets are available at their websites, including https://fastmri.org, http://www.mridata.org, and https://ocmr.info. Other in-house MRI datasets from our own collection are available from the corresponding author upon reasonable request.
Note: The software is used for testing only, and cannot be used commercially.
If you want to use the code, please cite the following paper:
Zi Wang et al., One for Multiple: Physics-informed Synthetic Data Boosts Generalizable Deep Learning for Fast MRI Reconstruction, arXiv:2307.13220, DOI: 10.48550/arXiv.2307.13220, 2023.