direct.nn.recurrentvarnet package

Submodules

direct.nn.recurrentvarnet.config module

class direct.nn.recurrentvarnet.config.RecurrentVarNetConfig(model_name='???', engine_name=None, num_steps=15, recurrent_hidden_channels=64, recurrent_num_layers=4, no_parameter_sharing=True, learned_initializer=True, initializer_initialization=InitType.SENSE, initializer_channels=(32, 32, 64, 64), initializer_dilations=(1, 1, 2, 4), initializer_multiscale=1, normalized=False)

Bases: ModelConfig

num_steps = 15

recurrent_hidden_channels = 64

recurrent_num_layers = 4

no_parameter_sharing = True

learned_initializer = True

initializer_initialization = 'sense'

initializer_channels = (32, 32, 64, 64)

initializer_dilations = (1, 1, 2, 4)

initializer_multiscale = 1

normalized = False

__init__(model_name='???', engine_name=None, num_steps=15, recurrent_hidden_channels=64, recurrent_num_layers=4, no_parameter_sharing=True, learned_initializer=True, initializer_initialization=InitType.SENSE, initializer_channels=(32, 32, 64, 64), initializer_dilations=(1, 1, 2, 4), initializer_multiscale=1, normalized=False)

direct.nn.recurrentvarnet.recurrentvarnet module

class direct.nn.recurrentvarnet.recurrentvarnet.RecurrentInit(in_channels, out_channels, channels, dilations, depth=2, multiscale_depth=1)

Bases: Module

Recurrent State Initializer (RSI) module of Recurrent Variational Network as presented in [1]_.

The RSI module learns to initialize the recurrent hidden state \(h_0\), input of the first RecurrentVarNetBlock of the RecurrentVarNet.

References:

[1]

Yiasemis, George, et al. “Recurrent Variational Network: A Deep Learning Inverse Problem Solver Applied to the Task of Accelerated MRI Reconstruction.” ArXiv:2111.09639 [Physics], Nov. 2021. arXiv.org, http://arxiv.org/abs/2111.09639.

__init__(in_channels, out_channels, channels, dilations, depth=2, multiscale_depth=1)

Inits RecurrentInit.

Parameters:

• in_channels (int) – Input channels.

• out_channels (int) – Number of hidden channels of the recurrent unit of RecurrentVarNet Block.

• channels (Tuple[int, ...]) – Channels \(n_d\) in the convolutional layers of initializer.

• dilations (Tuple[int, ...]) – Dilations \(p\) of the convolutional layers of the initializer.

• depth (int) – RecurrentVarNet Block number of layers \(n_l\). Default: 2.

• multiscale_depth (int) – Number of feature layers to aggregate for the output, if 1, multi-scale context aggregation is disabled. Default: 1.

forward(x)

Computes initialization for recurrent unit given input.

Parameters:

x (Tensor) – Initialization for RecurrentInit.

Return type:

Tensor

Returns:

Initial recurrent hidden state from input.

class direct.nn.recurrentvarnet.recurrentvarnet.RecurrentVarNet(forward_operator, backward_operator, in_channels=COMPLEX_SIZE, num_steps=15, recurrent_hidden_channels=64, recurrent_num_layers=4, no_parameter_sharing=True, learned_initializer=False, initializer_initialization=None, initializer_channels=(32, 32, 64, 64), initializer_dilations=(1, 1, 2, 4), initializer_multiscale=1, normalized=False, **kwargs)

Bases: Module

Recurrent Variational Network implementation as presented in [1]_.

References:

[1]

Yiasemis, George, et al. “Recurrent Variational Network: A Deep Learning Inverse Problem Solver Applied to the Task of Accelerated MRI Reconstruction.” ArXiv:2111.09639 [Physics], Nov. 2021. arXiv.org, http://arxiv.org/abs/2111.09639.

__init__(forward_operator, backward_operator, in_channels=COMPLEX_SIZE, num_steps=15, recurrent_hidden_channels=64, recurrent_num_layers=4, no_parameter_sharing=True, learned_initializer=False, initializer_initialization=None, initializer_channels=(32, 32, 64, 64), initializer_dilations=(1, 1, 2, 4), initializer_multiscale=1, normalized=False, **kwargs)

Inits RecurrentVarNet.

Parameters:

• forward_operator (Callable) – Callable

• Operator. (Backward)

• backward_operator (Callable) – Callable

• Operator.

• num_steps (int) – int

• iterations (Number of) – math:T.

• in_channels (int) – int

• data. (Input channel number. Default is 2 for complex)

• recurrent_hidden_channels (int) – int

• Default (is disabled.) –

• recurrent_num_layers (int) – int

• ( (Number of layers for the recurrent unit of the RecurrentVarNet Block) – math:n_l). Default: 4.

• no_parameter_sharing (bool) – bool

• False (If) – class:RecurrentVarNetBlock is used for all num_steps. Default: True.

• same (the) – class:RecurrentVarNetBlock is used for all num_steps. Default: True.

• learned_initializer (bool) – bool

• Default – False.

• initializer_initialization (Optional[InitType]) – str, Optional

• 'sense' (Type of initialization for the RSI module. Can be either)

• 'input-image'. ('zero-filled' or)

• Default – None.

• initializer_channels (Optional[Tuple[int, ...]]) – tuple

• Channels – math:n_d in the convolutional layers of the RSI module. Default: (32, 32, 64, 64).

• initializer_dilations (Optional[Tuple[int, ...]]) – tuple

• Dilations – math:p of the convolutional layers of the RSI module. Default: (1, 1, 2, 4).

• initializer_multiscale (int) – int

• output (RSI module number of feature layers to aggregate for the)

• 1 (if)

• aggregation (multi-scale context)

• Default

• normalized (bool) – bool

• True (If) – class:NormConv2dGRU will be used as a regularizer in the RecurrentVarNetBlocks.

:param : class:NormConv2dGRU will be used as a regularizer in the RecurrentVarNetBlocks. :param Default: False.

compute_sense_init(kspace, sensitivity_map)

Computes sense initialization \(x_{\text{SENSE}}\):

\[x_{\text{SENSE}} = \sum_{k=1}^{n_c} {S^{k}}^* \times y^k\]

where \(y^k\) denotes the data from coil \(k\).

Parameters:

• kspace (Tensor) – torch.Tensor

• shape (Sensitivity map of)

• sensitivity_map (Tensor) – torch.Tensor

• shape

Returns:

torch.Tensor Sense initialization \(x_{\text{SENSE}}\).

Return type:

input_image

forward(masked_kspace, sampling_mask, sensitivity_map, **kwargs)

Computes forward pass of RecurrentVarNet.

Parameters:

• masked_kspace (Tensor) – torch.Tensor

• shape (Coil sensitivities of)

• sampling_mask (Tensor) – torch.Tensor

• shape

• sensitivity_map (Tensor) – torch.Tensor

• shape

Returns:

torch.Tensor k-space prediction.

Return type:

kspace_prediction

class direct.nn.recurrentvarnet.recurrentvarnet.RecurrentVarNetBlock(forward_operator, backward_operator, in_channels=2, hidden_channels=64, num_layers=4, normalized=False)

Bases: Module

Recurrent Variational Network Block \(\mathcal{H}_{\theta_{t}}\) as presented in [1]_.

References:

[1]

Yiasemis, George, et al. “Recurrent Variational Network: A Deep Learning Inverse Problem Solver Applied to the Task of Accelerated MRI Reconstruction.” ArXiv:2111.09639 [Physics], Nov. 2021. arXiv.org, http://arxiv.org/abs/2111.09639.

__init__(forward_operator, backward_operator, in_channels=2, hidden_channels=64, num_layers=4, normalized=False)

Inits RecurrentVarNetBlock.

Parameters:

• forward_operator (Callable) – Callable

• Transform. (Backward Fourier)

• backward_operator (Callable) – Callable

• Transform.

• in_channels (int) – int,

• data. (Input channel number. Default is 2 for complex)

• hidden_channels (int) – int,

• Default (Hidden channels.) –

• num_layers (int) – int,

• of (Number of layers) – math:n_l recurrent unit. Default: 4.

• normalized (bool) – bool

• True (If) – class:NormConv2dGRU will be used as a regularizer. Default: False.

:param : class:NormConv2dGRU will be used as a regularizer. Default: False.

forward(current_kspace, masked_kspace, sampling_mask, sensitivity_map, hidden_state, coil_dim=1, spatial_dims=(2, 3))

Computes forward pass of RecurrentVarNetBlock.

Parameters:

• current_kspace (Tensor) – torch.Tensor

• shape (Recurrent unit hidden state of)

• masked_kspace (Tensor) – torch.Tensor

• shape

• sampling_mask (Tensor) – torch.Tensor

• shape

• sensitivity_map (Tensor) – torch.Tensor

• shape

• hidden_state (Optional[Tensor]) – torch.Tensor or None

• shape

• coil_dim (int) – int

• Default (Spatial dimensions.)

• spatial_dims (Tuple[int, int]) – tuple of ints

• Default – (2, 3).

Returns:

torch.Tensor New k-space prediction of shape (N, coil, height, width, complex=2). hidden_state: torch.Tensor Next hidden state of shape (N, hidden_channels, height, width, num_layers).

Return type:

new_kspace

direct.nn.recurrentvarnet.recurrentvarnet_engine module

class direct.nn.recurrentvarnet.recurrentvarnet_engine.RecurrentVarNetEngine(cfg, model, device, forward_operator=None, backward_operator=None, mixed_precision=False, **models)

Bases: MRIModelEngine

Recurrent Variational Network Engine.

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

Inits :class:`RecurrentVarNetEngine.

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, Tensor]

Module contents