VarNet#

class deepinv.models.VarNet(denoiser=None, sensitivity_model=None, num_cascades=12, mode='varnet')[source]#

Bases: ArtifactRemoval, MRIMixin

VarNet or E2E-VarNet model.

These models are from the papers Sriram et al., End-to-End Variational Networks for Accelerated MRI Reconstruction and Hammernik et al., Learning a variational network for reconstruction of accelerated MRI data.

This performs unrolled iterations on the image estimate x (as per the original VarNet paper) or the kspace y (as per E2E-VarNet).

Note

For singlecoil MRI, either mode is valid. For multicoil MRI, the VarNet mode will simply sum over the coils (not preferred). Using E2E-VarNet is therefore preferred. For sensitivity-map estimation for multicoil MRI, pass in sensitivity_model.

Code loosely adapted from E2E-VarNet implementation from facebookresearch/fastMRI.

Parameters:

  • denoiser (Denoiser, torch.nn.Module) – backbone network that parametrises the grad of the regulariser. If None, a small DnCNN is used.

  • sensitivity_model (torch.nn.Module) – network to jointly estimate coil sensitivity maps for multi-coil MRI. If None, do not perform any map estimation. For single-coil MRI, unused.

  • num_cascades (int) – number of unrolled iterations (‘cascades’).

  • mode (str) – if ‘varnet’, perform iterates on the images x as in original VarNet. If ‘e2e-varnet’, perform iterates on the kspace y as in the E2E-VarNet.

backbone_inference(tensor_in, physics, y)[source]#

Perform inference on input tensor.

Uses physics and y for data consistency. If necessary, perform fully-sampled MRI IFFT on model output.

Parameters:

  • tensor_in (torch.Tensor) – input tensor as dictated by VarNet mode (either k-space or image)

  • physics (Physics) – forward physics for data consistency

  • y (torch.Tensor) – input measurements y for data consistency

Returns:

(torch.Tensor) reconstructed image

Return type:

Tensor

Examples using VarNet:#

Tour of MRI functionality in DeepInverse

Tour of MRI functionality in DeepInverse