Ctrl+K

Deep Cardiac MRI Reconstruction with ADMM (CMRxRecon Challenge 2023)

Deep Cardiac MRI Reconstruction with ADMM (CMRxRecon Challenge 2023)#

fig

Figure 1: Pipeline of our proposed methods.#

tabs

Figure 2: Average results and inference times.#

This project contains necessary configuration files to reproduce experiments of 2nd top-ranking approach to both tasks of CMRxRecon Challenge 2023 as presented in Deep Cardiac MRI Reconstruction with ADMM. This will also help you set up the training and inference data directories.

Setting up data directory#

This project aims to help you set up directories for training and validation data using the specified directory structures necessary to run with DIRECT. You will need to run this process twice: once for “Cine” data and once for “Mapping” data.

Prerequisites#

Before you begin, make sure you have downloaded the CMRxRecon Challenge data. Check the challenge website for more information.

Assumed Base Path Structure#

The script assumes that CMRxRecon data is organized according to the following directory structure:

base_path
├── MultiCoil
│   ├── Cine_or_Mapping
│   │   ├── TrainingSet
│   │   │   ├── FullSample
│   │   │   ├── AccFactor04
│   │   │   ├── AccFactor08
│   │   │   └── AccFactor10
│   │   ├── ValidationSet
│   │   │   ├── FullSample
│   │   │   ├── AccFactor04
│   │   │   ├── AccFactor08
│   │   │   └── AccFactor10
│   │   ├── TestSet
│   │   │   ├── FullSample
│   │   │   ├── AccFactor04
│   │   │   ├── AccFactor08
│   │   │   └── AccFactor10

Create Symbolic Directories#

The following creates files of fully sampled data with the respective masks, and creates symbolic paths of data in single directories to be used with DIRECT.

python3 create_data_dir.py --base_path path_to_base_data --target_path path_to_target_directory --data_type Cine
python3 create_data_dir.py --base_path path_to_base_data --target_path path_to_target_directory --data_type Mapping

You can append –create_training_data_with_masks on the commands above for the training data with sampling masks creation option.

Experiments#

Configuration Files#

We provide configuration files for DIRECT for experiments presented in Deep Cardiac MRI Reconstruction with ADMM in the CMRxRecon configs folder.

Training#

In direct/ run the following command to begin training on the training data.

direct train <output_folder> \
            --training-root <target_path>/MultiCoil/training/ \
            --validation-root <target_path>/MultiCoil/training/ \
            --cfg projects/CMRxRecon/configs/base_<name_of_experiment>.yaml \
            --num-gpus <number_of_gpus> \
            --num-workers <number_of_workers> --resume

Note that for validation a subset of the training data is used since full validation data have not been released.

Inference#

Note that inference is performed for a single dataset, therefore a single acceleration factor. For example, the following entry for inference will perform predictions for acceleration factor of 4x on validation data. Change kspace_key: kspace_sub04 to kspace_key: kspace_sub08 for 8x and kspace_key: kspace_sub10 for 10x.

inference:
    batch_size: 8
    dataset:
        name: CMRxRecon
        kspace_key: kspace_sub04
        compute_mask: true
        transforms:
            cropping:
                crop: null
            sensitivity_map_estimation:
                estimate_sensitivity_maps: true
            normalization:
                scaling_key: masked_kspace
                scale_percentile: 0.99
            masking: null
        text_description: inference-4x
    crop: null

In direct/ run the following command to perform inference, for instance on 4x:

direct predict <output_directory>
            --checkpoint <path_or_url_to_checkpoint> \
            --cfg projects/CMRxRecon/configs/base_<name_of_experiment>.yaml \
            --data-root <target_path>/MultiCoil/<Cine_or_Mapping>/validation/AccFactor<04_or_08_or_10> \
            --num-gpus <number_of_gpus> \
            --num-workers <number_of_workers> \
            [--other-flags]

Note#

Fully sampled validation dataset and Test data have been released after the challenge.

Citing this work#

Please use the following BiBTeX entries if you use our proposed methods in your work:

@article{DIRECTTOOLKIT,
    title = {DIRECT: Deep Image REConstruction Toolkit},
    author = {
        George Yiasemis and Nikita Moriakov and Dimitrios Karkalousos and Matthan
        Caan and Jonas Teuwen
    },
    year = 2022,
    journal = {Journal of Open Source Software},
    publisher = {The Open Journal},
    volume = 7,
    number = 73,
    pages = 4278,
    doi = {10.21105/joss.04278},
    url = {https://doi.org/10.21105/joss.04278}
}

@article{lyu2024stateoftheart,
    title = {
        The state-of-the-art in Cardiac MRI Reconstruction: Results of the CMRxRecon
        Challenge in MICCAI 2023
    },
    author = {
        Jun Lyu and Chen Qin and Shuo Wang and Fanwen Wang and Yan Li and Zi Wang and
        Kunyuan Guo and Cheng Ouyang and Michael Tänzer and Meng Liu and Longyu Sun
        and Mengting Sun and Qin Li and Zhang Shi and Sha Hua and Hao Li and Zhensen
        Chen and Zhenlin Zhang and Bingyu Xin and Dimitris N. Metaxas and George
        Yiasemis and Jonas Teuwen and others
    },
    year = 2024,
    eprint = {2404.01082},
    archiveprefix = {arXiv},
    primaryclass = {eess.IV}
}

@article{yiasemis2023vsharp,
    title = {
        vSHARP: variable Splitting Half-quadratic ADMM algorithm for Reconstruction
        of inverse-Problems
    },
    author = {George Yiasemis and Nikita Moriakov and Jan-Jakob Sonke and Jonas Teuwen},
    year = 2023,
    month = {Sep},
    journal = {arXiv.org},
    doi = {10.48550/arXiv.2309.09954},
    url = {https://doi.org/10.48550/arXiv.2309.09954},
    note = {arXiv:2309.09954 [eess.IV]},
    eprint = {2309.09954},
    archiveprefix = {arXiv},
    primaryclass = {eess.IV}
}

@inproceedings{yiasemis2024deep,
    title = {Deep Cardiac MRI Reconstruction with ADMM},
    author = {Yiasemis, George and Moriakov, Nikita and Sonke, Jan-Jakob and Teuwen, Jonas},
    year = 2024,
    booktitle = {
        Statistical Atlases and Computational Models of the Heart. Regular and
        CMRxRecon Challenge Papers
    },
    publisher = {Springer Nature Switzerland},
    address = {Cham},
    pages = {479--490},
    doi = {10.1007/978-3-031-52448-6\_45},
    isbn = {978-3-031-52448-6},
    url = {https://doi.org/10.1007/978-3-031-52448-6\_45}
}
Contents