direct.nn.iterdualnet package

Submodules

direct.nn.iterdualnet.config module

class direct.nn.iterdualnet.config.IterDualNetConfig(model_name='???', engine_name=None, num_iter=10, image_normunet=False, kspace_normunet=False, image_unet_num_filters=8, image_unet_num_pool_layers=4, image_unet_dropout=0.0, kspace_unet_num_filters=8, kspace_unet_num_pool_layers=4, kspace_unet_dropout=0.0, image_no_parameter_sharing=True, kspace_no_parameter_sharing=False, compute_per_coil=True)

Bases: ModelConfig

num_iter = 10

image_normunet = False

kspace_normunet = False

image_unet_num_filters = 8

image_unet_num_pool_layers = 4

image_unet_dropout = 0.0

kspace_unet_num_filters = 8

kspace_unet_num_pool_layers = 4

kspace_unet_dropout = 0.0

image_no_parameter_sharing = True

kspace_no_parameter_sharing = False

compute_per_coil = True

__init__(model_name='???', engine_name=None, num_iter=10, image_normunet=False, kspace_normunet=False, image_unet_num_filters=8, image_unet_num_pool_layers=4, image_unet_dropout=0.0, kspace_unet_num_filters=8, kspace_unet_num_pool_layers=4, kspace_unet_dropout=0.0, image_no_parameter_sharing=True, kspace_no_parameter_sharing=False, compute_per_coil=True)

direct.nn.iterdualnet.iterdualnet module

class direct.nn.iterdualnet.iterdualnet.IterDualNet(forward_operator, backward_operator, num_iter=10, image_normunet=False, kspace_normunet=False, image_no_parameter_sharing=True, kspace_no_parameter_sharing=True, compute_per_coil=True, **kwargs)

Bases: Module

Iterative Dual Network solves iteratively the following problem

\[\begin{split}\min_{x} ||A(x) - y||_2^2 + \lambda_I ||x - D_I(x)||_2^2 + \lambda_F ||x - \mathcal{Q}(D_F(f))||_2^2, \quad \left\{ \begin{array} Q = \mathcal{F}^{-1}, f = \mathcal{F}(x) & \text{if compute_per_coil is False} \\ Q = \mathcal{F}^{-1} \circ \mathcal{E}, f = \mathcal{R} \circ \mathcal{F}(x) & \text{otherwise} \end{array}\end{split}\]

by unrolling a gradient descent scheme where \(\mathcal{E}\) and \(\mathcal{R}\) are the expand and reduce operators which use the sensitivity maps. \(D_I\) and \(D_F\) are trainable U-Nets operating in the image and k-space domain.

__init__(forward_operator, backward_operator, num_iter=10, image_normunet=False, kspace_normunet=False, image_no_parameter_sharing=True, kspace_no_parameter_sharing=True, compute_per_coil=True, **kwargs)

Inits IterDualNet.

Parameters:

• forward_operator (Callable) – Forward Operator.

• backward_operator (Callable) – Backward Operator.

• num_iter (int) – Number of iterations. Default: 10.

• image_normunet (bool) – If True will use NormUNet for the image model. Default: False.

• kspace_normunet (bool) – If True will use NormUNet for the kspace model. Default: False.

• image_no_parameter_sharing (bool) – If False, a single image model will be shared across all iterations. Default: True.

• kspace_no_parameter_sharing (bool) – If False, a single kspace model will be shared across all iterations. Default: True.

• compute_per_coil (bool) – If True \(f\) will be transformed into a multi-coil kspace. Default: True.

• **kwargs – Keyword arguments for unet models.

forward(masked_kspace, sampling_mask, sensitivity_map)

Computes forward pass of IterDualNet.

Parameters:

• masked_kspace (Tensor) – torch.Tensor

• shape (Sensitivity map of)

• sampling_mask (Tensor) – torch.Tensor

• shape

• sensitivity_map (Tensor) – torch.Tensor

• shape

Returns:

torch.Tensor Output image of shape (N, height, width, complex=2).

Return type:

out_image

direct.nn.iterdualnet.iterdualnet_engine module

class direct.nn.iterdualnet.iterdualnet_engine.IterDualNetEngine(cfg, model, device, forward_operator=None, backward_operator=None, mixed_precision=False, **models)

Bases: MRIModelEngine

__init__(cfg, model, device, forward_operator=None, backward_operator=None, mixed_precision=False, **models)

Inits IterDualNetEngine.

Parameters:

• cfg (BaseConfig) – Configuration file.

• model (Module) – Model.

• device (str) – Device. Can be “cuda: {idx}” or “cpu”.

• forward_operator (Optional[Callable]) – The forward operator. Default: None.

• backward_operator (Optional[Callable]) – The backward operator. Default: None.

• mixed_precision (bool) – Use mixed precision. Default: False.

• **models (Module) – Additional models.

forward_function(data)

This method performs the model’s forward method given data which contains all tensor inputs.

Must be implemented by child classes.

Return type:

Tuple[Tensor, None]

Module contents