Skip to content

Latest commit

 

History

History
110 lines (82 loc) · 5.29 KB

README.md

File metadata and controls

110 lines (82 loc) · 5.29 KB

PHIMO: Physics-Informed Deep Learning for Motion-Corrected Reconstruction of Quantitative Brain MRI

Hannah Eichhorn, Veronika Spieker, Kerstin Hammernik, Elisa Saks, Kilian Weiss, Christine Preibisch, Julia A. Schnabel

Accepted at MICCAI 2024 | Link to paper

Abstract: We propose PHIMO, a physics-informed learning-based motion correction method tailored to quantitative MRI. PHIMO leverages information from the signal evolution to exclude motion-corrupted k-space lines from a data-consistent reconstruction. We demonstrate the potential of PHIMO for the application of T2* quantification from gradient echo MRI, which is particularly sensitive to motion due to its sensitivity to magnetic field inhomogeneities. A state-of-the-art technique for motion correction requires redundant acquisition of the k-space center, prolonging the acquisition.We show that PHIMO can detect and exclude intra-scan motion events and, thus, correct for severe motion artifacts. PHIMO approaches the performance of the state-of-the-art motion correction method, while substantially reducing the acquisition time by over 40%, facilitating clinical applicability.

Keywords: Self-Supervised Learning · Motion Detection · Data-Consistent Reconstruction · T2* Quantification · Gradient Echo MRI

Citation

If you use this code, please cite our arXiv paper:

@InProceedings{eichhorn2024,
author="Eichhorn, Hannah and Spieker, Veronika and Hammernik, Kerstin and Saks, Elisa and Weiss, Kilian and Preibisch, Christine and Schnabel, Julia A.",
editor="Linguraru, Marius George and Dou, Qi and Feragen, Aasa and Giannarou, Stamatia and Glocker, Ben and Lekadir, Karim and Schnabel, Julia A.",
title="Physics-Informed Deep Learning for Motion-Corrected Reconstruction of Quantitative Brain MRI",
booktitle="Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024",
year="2024",
publisher="Springer Nature Switzerland",
address="Cham",
pages="562--571",
isbn="978-3-031-72104-5"
}

Contents of this repository:

  • iml-dl: code belonging to the above work, using an adapted version of the IML-CompAI Framework and the MERLIN Framework. The weights of the trained models are not shared, since the training was performed on private data.

All computations were performed using Python 3.8.12 and PyTorch 2.0.1.

Setup:

  1. Create a virtual environment with the required packages:

    cd ${TARGET_DIR}/PHIMO
conda env create -f conda_environment.yml
source activate phimo *or* conda activate phimo

  2. Install pytorch with cuda:

    conda install pytorch torchvision torchaudio pytorch-cuda=11.7 -c pytorch -c nvidia
pip install torchinfo
conda install -c conda-forge pytorch-lightning

  3. For setting up wandb please refer to the IML-CompAI Framework.

Steps to reproduce the analysis:

  1. Use the file iml-dl/projects/recon_t2star/preprocessing/create_symbolic_links.py to create symbolic links under iml-dl/data/links_to_data/ (folders need to be adapted here).

  2. Adapt iml-dl/projects/recon_t2star/configs/config_train_random_masks.yaml and ismrm-abstract/iml-dl/projects/recon_t2star/configs/config_train_var_dens_masks.yaml to correct settings (i.e. input and output folders) for training the unrolled reconstruction network on random masks for PHIMO and variable density masks for OR-BA

  3. Run the following commands for training the reconstruction networks:

    # go to right directory and activate the conda environment
cd path_to_code/PHIMO/preprint/iml-dl
conda activate phimo

# launch the experiment
echo 'Starting Script'
python -u ./core/Main.py --config_path ./projects/recon_t2star/configs/config_train_{random_masks or var_dens_masks}.yaml

  4. Adapt iml-dl/projects/moco_t2star/configs/config_test_moco.yaml to correct settings (i.e. input and output folders) for optimising the line detection MLP.

  5. Run the following commands for optimising the MLP (motion detection and correction):

    # go to right directory and activate the conda environment
cd path_to_code/PHIMO/preprint/iml-dl
conda activate phimo

# launch the experiment
echo 'Starting Script'
python -u ./core/Main.py --config_path ./projects/moco_t2star/configs/config_test_moco.yaml

    Note: this needs to be run for each subject separately. Please exchange the select_one_scan variable in the config file for that.

  6. For the evaluation of results and creation of plots for the preprint, adapt the config file iml-dl/projects/moco_t2star/configs/config_evaluate_moco.yaml and run iml-dl/projects/moco_t2star/evaluate_predictions.py

Illustration of PHIMO:

PHIMO Overview of PHIMO. (A) Training an unrolled reconstruction network on randomly undersampled motion-free multi-echo images. (B) MoCo of motion-corrupted data by optimising an MLP to predict slice-wise exclusion masks in a self-supervised fashion for each subject individually. (C) Calculation of empirical correlation coefficient as physics-informed loss for optimising the MLP in (B).