Images from Inter4K [1]: a high spatio-temporal resolution publicly available natural video dataset are used to learn Dynamic MR image reconstruction.

Methods can be found in [link].

Three networks and trajectories were investigated:

• Cartesian real-time with unrolled VarNet [2]
• Radial real-time with multicoil UNet
• Low latency spiral imaging with FastDVDNet (Hyperslice [3])

• Top Line: RGB Video, Undersampled Cartesian, Undersampled Radial, Undersampled Spiral.

• Bottom Line: Target, VarNet Reconstruction, multicoil 3DUNet Reconstruction, FastDVDNet Reconstruction.

Provided code includes trajectories, model training and pre-trained models as implemented for the paper.

The ethics does not allow sharing medical image data therefore only Inter4K data and models are made available.

For installation please:

1. Download github repository.

2. From Project folder, create Docker image and launch interactive docker container:

docker compose up --build -d

3. Download and unzip Inter4K Dataset in DatasetFolder (see DatasetFolder/README.md if does not work):

docker compose exec tensorflow python download_Inter4k_Dataset.py

4. Test training by using one of the following commands :

nohup docker compose exec tensorflow python train\_network.py -m VarNet > training_VarNet.log & # for VarNet Cartesian training (longest)
nohup docker compose exec tensorflow python train\_network.py -m 3DUNet > training_3DUNet.log & # for full model 3DUNet radial training
nohup docker compose exec tensorflow python train\_network.py -m FastDVDNet > training_FastDVDNet.log & # for FastDVDNet Spiral training

5. Shut down docker container:

docker compose down

0. Alternatively can be used with VScode (.devcontainer folder) for development within the docker container.

Note that only Linux is supported.

Results are saved in ./Training_folder (as in the already trained example models ./Training_folder/Default_FastDVDNet)

Logs can also be visualised by using tensorboard:

tensorboard --logdir ./Training\_folder/Default\_FastDVDNet

[1] Stergiou, A., & Poppe, R. (2023). AdaPool: Exponential Adaptive Pooling for Information-Retaining Downsampling. IEEE Transactions on Image Processing, 32, 251–266. https://doi.org/10.1109/TIP.2022.3227503

[2] Hammernik, K., Klatzer, T., Kobler, E., Recht, M. P., Sodickson, D. K., Pock, T., & Knoll, F. (2018). Learning a variational network for reconstruction of accelerated MRI data. Magnetic Resonance in Medicine, 79(6), 3055–3071. https://doi.org/10.1002/mrm.26977

[3] Jaubert, O., Montalt-Tordera, J., Knight, D., Arridge, S., Steeden, J., & Muthurangu, V. (2023). HyperSLICE: HyperBand optimized spiral for low-latency interactive cardiac examination. Magnetic Resonance in Medicine. https://doi.org/10.1002/MRM.29855