NCPDNet-3D

NCPDNet-3D

---

tags:

• TensorFlow
• MRI reconstruction
• MRI datasets:
• OASIS

---

This is a non-Cartesian 3D MRI reconstruction model for radial trajectories at acceleration factor 4. The model uses 10 iterations and a small vanilla CNN.

Model description

For more details, see https://hal.inria.fr/hal-03188997. This section is WIP.

Intended uses and limitations

This model can be used to reconstruct 3D brain data obtained retrospectively from magnitude scanners of the OASIS database at acceleration factor 4 in a fully radial acquisition setting.

How to use

This model can be loaded using the following repo: https://github.com/zaccharieramzi/fastmri-reproducible-benchmark. After cloning the repo, git clone https://github.com/zaccharieramzi/fastmri-reproducible-benchmark, you can install the package via pip install fastmri-reproducible-benchmark. The framework is TensorFlow.

You can initialize and load the model weights as follows:

import tensorflow as tf

from fastmri_recon.models.subclassed_models.ncpdnet import NCPDNet


model = NCPDNet(
    three_d=True,
    n_iter=6,
    n_filters=16,
    im_size=(176, 256, 256),
    dcomp=True,
    fastmri=False,
    n_primal=2,
)
kspace_shape = 1
inputs = [
    tf.zeros([1, 1, kspace_shape, 1], dtype=tf.complex64),
    tf.zeros([1, 3, kspace_shape], dtype=tf.float32),
    (tf.constant([(176, 256, 256)]), tf.ones([1, kspace_shape], dtype=tf.float32)),
]
model(inputs)
model.load_weights('model_weights.h5')

Using the model is then as simple as:

model([
    kspace,  # shape: [n_batch, 1, n_kspace_samples, 1]
    traj,  # shape: [n_batch, 1, 3, n_kspace_samples]
    (
        output_shape,  # shape: [n_batch, 3]
        dcomp,  # shape: [n_batch, n_kspace_samples]
    )
])

Limitations and bias

The limitations and bias of this model have not been properly investigated.

Training data

This model was trained using the OASIS dataset.

Training procedure

The training procedure is described in https://hal.inria.fr/hal-03188997. This section is WIP.

Evaluation results

On the OASIS validation dataset:

• PSNR: 33.76

Bibtex entry

@article{ramzi2022nc,
  title={NC-PDNet: A density-compensated unrolled network for 2D and 3D non-Cartesian MRI reconstruction},
  author={Ramzi, Zaccharie and Chaithya, GR and Starck, Jean-Luc and Ciuciu, Philippe},
  journal={IEEE Transactions on Medical Imaging},
  volume={41},
  number={7},
  pages={1625--1638},
  year={2022},
  publisher={IEEE}
}