Collections & Models

ATOMMIC is organized in collections, each of which implements a specific task. The following collections are currently available, implementing various models as listed.

MultiTask Learning (MTL)

End-to-End Recurrent Attention Network

End-to-End Recurrent Attention Network (SERANet), as presented in [Huang2019].

Huang2019

Huang, Q., Chen, X., Metaxas, D., Nadar, M.S. (2019). Brain Segmentation from k-Space with End-to-End Recurrent Attention Network. In: , et al. Medical Image Computing and Computer Assisted Intervention – MICCAI 2019. Lecture Notes in Computer Science(), vol 11766. Springer, Cham. https://doi.org/10.1007/978-3-030-32248-9_31

Example configuration:

model:
    model_name: SERANET
    use_reconstruction_module: true
    input_channels: 2
    reconstruction_module: unet
    reconstruction_module_output_channels: 2
    reconstruction_module_channels: 32
    reconstruction_module_pooling_layers: 4
    reconstruction_module_dropout: 0.0
    # or
    #  reconstruction_module: cascadenet
    #  reconstruction_module_hidden_channels: 32
    #  reconstruction_module_n_convs: 2
    #  reconstruction_module_batchnorm: true
    #  reconstruction_module_num_cascades: 5
    reconstruction_module_num_blocks: 3
    segmentation_module_input_channels: 32
    segmentation_module_output_channels: 2
    segmentation_module_channels: 32
    segmentation_module_pooling_layers: 4
    segmentation_module_dropout: 0.0
    recurrent_module_iterations: 2
    recurrent_module_attention_channels: 32
    recurrent_module_attention_pooling_layers: 4
    recurrent_module_attention_dropout: 0.0
    # task & dataset related parameters
    coil_combination_method: SENSE
    coil_dim: 1
    complex_valued_type: stacked  # stacked, complex_abs, complex_sqrt_abs
    complex_data: true
    consecutive_slices: 1
    dimensionality: 2
    estimate_coil_sensitivity_maps_with_nn: false
    fft_centered: false
    fft_normalization: backward
    spatial_dims:
        - -2
        - -1
    magnitude_input: true
    normalization_type: minmax
    normalize_segmentation_output: true
    unnormalize_loss_inputs: false
    unnormalize_log_outputs: false

Image domain Deep Structured Low-Rank Network

Image domain Deep Structured Low-Rank Network (IDSLR), as presented in [Pramanik2021].

Pramanik2021

Pramanik A, Wu X, Jacob M. Joint calibrationless reconstruction and segmentation of parallel MRI. arXiv preprint arXiv:2105.09220. 2021 May 19.

Example configuration:

model:
    model_name: IDSLR
    use_reconstruction_module: true
    input_channels: 24  # coils * 2
    reconstruction_module_output_channels: 24  # coils * 2
    segmentation_module_output_channels: 2
    channels: 64
    num_pools: 2
    padding_size: 11
    drop_prob: 0.0
    normalize: true
    padding: true
    norm_groups: 2
    num_iters: 5
    # task & dataset related parameters
    coil_combination_method: SENSE
    coil_dim: 1
    complex_valued_type: stacked  # stacked, complex_abs, complex_sqrt_abs
    complex_data: true
    consecutive_slices: 1
    dimensionality: 2
    estimate_coil_sensitivity_maps_with_nn: false
    fft_centered: false
    fft_normalization: backward
    spatial_dims:
        - -2
        - -1
    magnitude_input: true
    normalization_type: minmax
    normalize_segmentation_output: true
    unnormalize_loss_inputs: false
    unnormalize_log_outputs: false

Image domain Deep Structured Low-Rank UNet

Image domain Deep Structured Low-Rank network using a UNet (and not only the decoder part) as segmentation model (IDSLRUNet), as presented in [Pramanik2021].

Pramanik2021

Pramanik A, Wu X, Jacob M. Joint calibrationless reconstruction and segmentation of parallel MRI. arXiv preprint arXiv:2105.09220. 2021 May 19.

Example configuration:

model:
    model_name: IDSLRUNET
    use_reconstruction_module: true
    input_channels: 24  # coils * 2
    reconstruction_module_output_channels: 24  # coils * 2
    segmentation_module_output_channels: 2
    channels: 64
    num_pools: 2
    padding_size: 11
    drop_prob: 0.0
    normalize: true
    padding: true
    norm_groups: 2
    num_iters: 5
    # task & dataset related parameters
    coil_combination_method: SENSE
    coil_dim: 1
    complex_valued_type: stacked  # stacked, complex_abs, complex_sqrt_abs
    complex_data: true
    consecutive_slices: 1
    dimensionality: 2
    estimate_coil_sensitivity_maps_with_nn: false
    fft_centered: false
    fft_normalization: backward
    spatial_dims:
        - -2
        - -1
    magnitude_input: true
    normalization_type: minmax
    normalize_segmentation_output: true
    unnormalize_loss_inputs: false
    unnormalize_log_outputs: false

Multi-Task Learning for MRI Reconstruction and Segmentation

Multi-Task Learning for MRI Reconstruction and Segmentation (MTLRS), as presented in [Karkalousos2023].

Karkalousos2023

Karkalousos, D., Išgum, I., Marquering, H., Caan, M. W. A., (2023). MultiTask Learning for accelerated-MRI Reconstruction and Segmentation of Brain Lesions in Multiple Sclerosis. In Proceedings of Machine Learning Research (Vol. 078).

Example configuration:

model:
    model_name: MTLRS
    joint_reconstruction_segmentation_module_cascades: 5
    task_adaption_type: multi_task_learning
    use_reconstruction_module: true
    reconstruction_module_recurrent_layer: IndRNN
    reconstruction_module_conv_filters:
        - 64
        - 64
        - 2
    reconstruction_module_conv_kernels:
        - 5
        - 3
        - 3
    reconstruction_module_conv_dilations:
        - 1
        - 2
        - 1
    reconstruction_module_conv_bias:
        - true
        - true
        - false
    reconstruction_module_recurrent_filters:
        - 64
        - 64
        - 0
    reconstruction_module_recurrent_kernels:
        - 1
        - 1
        - 0
    reconstruction_module_recurrent_dilations:
        - 1
        - 1
        - 0
    reconstruction_module_recurrent_bias:
        - true
        - true
        - false
    reconstruction_module_depth: 2
    reconstruction_module_time_steps: 8
    reconstruction_module_conv_dim: 2
    reconstruction_module_num_cascades: 1
    reconstruction_module_dimensionality: 2
    reconstruction_module_no_dc: true
    reconstruction_module_keep_prediction: true
    reconstruction_module_accumulate_predictions: true
    segmentation_module: AttentionUNet
    segmentation_module_input_channels: 1
    segmentation_module_output_channels: 2
    segmentation_module_channels: 64
    segmentation_module_pooling_layers: 2
    segmentation_module_dropout: 0.0
    # task & dataset related parameters
    coil_combination_method: SENSE
    coil_dim: 1
    complex_valued_type: stacked  # stacked, complex_abs, complex_sqrt_abs
    complex_data: true
    consecutive_slices: 1
    dimensionality: 2
    estimate_coil_sensitivity_maps_with_nn: false
    fft_centered: false
    fft_normalization: backward
    spatial_dims:
        - -2
        - -1
    magnitude_input: true
    normalization_type: minmax
    normalize_segmentation_output: true
    unnormalize_loss_inputs: false
    unnormalize_log_outputs: false

Reconstruction Segmentation method using UNet

Reconstruction Segmentation method using UNets for both the reconstruction and segmentation (RecSegUNet), as presented in [Sui2021].

Sui2021

Sui, B, Lv, J, Tong, X, Li, Y, Wang, C. Simultaneous image reconstruction and lesion segmentation in accelerated MRI using multitasking learning. Med Phys. 2021; 48: 7189– 7198. https://doi.org/10.1002/mp.15213

Example configuration:

model:
    model_name: RECSEGNET
    use_reconstruction_module: true
    input_channels: 1
    reconstruction_module_output_channels: 1
    reconstruction_module_channels: 64
    reconstruction_module_pooling_layers: 2
    reconstruction_module_dropout: 0.0
    segmentation_module_output_channels: 2
    segmentation_module_channels: 64
    segmentation_module_pooling_layers: 2
    segmentation_module_dropout: 0.0
    # task & dataset related parameters
    coil_combination_method: SENSE
    coil_dim: 1
    complex_valued_type: stacked  # stacked, complex_abs, complex_sqrt_abs
    complex_data: true
    consecutive_slices: 1
    dimensionality: 2
    estimate_coil_sensitivity_maps_with_nn: false
    fft_centered: false
    fft_normalization: backward
    spatial_dims:
        - -2
        - -1
    magnitude_input: true
    normalization_type: minmax
    normalize_segmentation_output: true
    unnormalize_loss_inputs: false
    unnormalize_log_outputs: false

Segmentation Network MRI

Segmentation Network MRI (SegNet), as presented in [Sun2019].

Sun2019

Sun, L., Fan, Z., Ding, X., Huang, Y., Paisley, J. (2019). Joint CS-MRI Reconstruction and Segmentation with a Unified Deep Network. In: Chung, A., Gee, J., Yushkevich, P., Bao, S. (eds) Information Processing in Medical Imaging. IPMI 2019. Lecture Notes in Computer Science(), vol 11492. Springer, Cham. https://doi.org/10.1007/978-3-030-20351-1_38

Example configuration:

model:
    model_name: SEGNET
    use_reconstruction_module: true
    input_channels: 24  # coils * 2
    reconstruction_module_output_channels: 24  # coils * 2
    segmentation_module_output_channels: 2
    channels: 64
    num_pools: 2
    padding_size: 11
    drop_prob: 0.0
    normalize: true
    padding: true
    norm_groups: 2
    num_cascades: 5
    segmentation_final_layer_conv_dim: 2
    segmentation_final_layer_kernel_size: 3
    segmentation_final_layer_dilation: 1
    segmentation_final_layer_bias: False
    segmentation_final_layer_nonlinear: relu
    # task & dataset related parameters
    coil_combination_method: SENSE
    coil_dim: 1
    complex_valued_type: stacked  # stacked, complex_abs, complex_sqrt_abs
    complex_data: true
    consecutive_slices: 1
    dimensionality: 2
    estimate_coil_sensitivity_maps_with_nn: false
    fft_centered: false
    fft_normalization: backward
    spatial_dims:
        - -2
        - -1
    magnitude_input: true
    normalization_type: minmax
    normalize_segmentation_output: true
    unnormalize_loss_inputs: false
    unnormalize_log_outputs: false

Quantitative MR Imaging (qMRI)

quantitative Cascades of Independently Recurrent Inference Machines

quantitative Cascades of Independently Recurrent Inference Machines (qCIRIM), as presented in [Zhang2022].

Zhang2022

Zhang C, Karkalousos D, Bazin PL, Coolen BF, Vrenken H, Sonke JJ, Forstmann BU, Poot DH, Caan MW. A unified model for reconstruction and R2* mapping of accelerated 7T data using the quantitative recurrent inference machine. NeuroImage. 2022 Dec 1;264:119680.

Example configuration:

model:
    model_name: qCIRIM
    use_reconstruction_module: true
    reconstruction_module_recurrent_layer: IndRNN
    reconstruction_module_conv_filters:
        - 64
        - 64
        - 2
    reconstruction_module_conv_kernels:
        - 5
        - 3
        - 3
    reconstruction_module_conv_dilations:
        - 1
        - 2
        - 1
    reconstruction_module_conv_bias:
        - true
        - true
        - false
    reconstruction_module_recurrent_filters:
        - 64
        - 64
        - 0
    reconstruction_module_recurrent_kernels:
        - 1
        - 1
        - 0
    reconstruction_module_recurrent_dilations:
        - 1
        - 1
        - 0
    reconstruction_module_recurrent_bias:
        - true
        - true
        - false
    reconstruction_module_depth: 2
    reconstruction_module_time_steps: 8
    reconstruction_module_conv_dim: 2
    reconstruction_module_num_cascades: 1
    reconstruction_module_dimensionality: 2
    reconstruction_module_no_dc: true
    reconstruction_module_keep_prediction: true
    reconstruction_module_accumulate_predictions: true
    quantitative_module_recurrent_layer: IndRNN
    quantitative_module_conv_filters:
        - 64
        - 64
        - 4
    quantitative_module_conv_kernels:
        - 5
        - 3
        - 3
    quantitative_module_conv_dilations:
        - 1
        - 2
        - 1
    quantitative_module_conv_bias:
        - true
        - true
        - false
    quantitative_module_recurrent_filters:
        - 64
        - 64
        - 0
    quantitative_module_recurrent_kernels:
        - 1
        - 1
        - 0
    quantitative_module_recurrent_dilations:
        - 1
        - 1
        - 0
    quantitative_module_recurrent_bias:
        - true
        - true
        - false
    quantitative_module_depth: 2
    quantitative_module_time_steps: 8
    quantitative_module_conv_dim: 2
    quantitative_module_num_cascades: 1
    quantitative_module_no_dc: true
    quantitative_module_keep_prediction: true
    quantitative_module_accumulate_predictions: true
    quantitative_module_signal_forward_model_sequence: MEGRE
    quantitative_module_dimensionality: 2
    quantitative_module_gamma_regularization_factors:
        - 150.0
        - 150.0
        - 1000.0
        - 150.0
    # task & dataset related parameters
    coil_combination_method: SENSE
    coil_dim: 1
    complex_valued_type: stacked  # stacked, complex_abs, complex_sqrt_abs
    consecutive_slices: 1
    dimensionality: 2
    estimate_coil_sensitivity_maps_with_nn: false
    fft_centered: false
    fft_normalization: backward
    spatial_dims:
        - -2
        - -1
    shift_B0_input: false
    normalization_type: minmax
    unnormalize_loss_inputs: false
    unnormalize_log_outputs: false

quantitative Recurrent Inference Machines

quantitative Recurrent Inference Machines (qRIMBlock), as presented in [Zhang2022].

Zhang2022

Zhang C, Karkalousos D, Bazin PL, Coolen BF, Vrenken H, Sonke JJ, Forstmann BU, Poot DH, Caan MW. A unified model for reconstruction and R2* mapping of accelerated 7T data using the quantitative recurrent inference machine. NeuroImage. 2022 Dec 1;264:119680.

Example configuration:

model:
    model_name: qCIRIM
    use_reconstruction_module: true
    reconstruction_module_recurrent_layer: GRU
    reconstruction_module_conv_filters:
        - 64
        - 64
        - 2
    reconstruction_module_conv_kernels:
        - 5
        - 3
        - 3
    reconstruction_module_conv_dilations:
        - 1
        - 2
        - 1
    reconstruction_module_conv_bias:
        - true
        - true
        - false
    reconstruction_module_recurrent_filters:
        - 64
        - 64
        - 0
    reconstruction_module_recurrent_kernels:
        - 1
        - 1
        - 0
    reconstruction_module_recurrent_dilations:
        - 1
        - 1
        - 0
    reconstruction_module_recurrent_bias:
        - true
        - true
        - false
    reconstruction_module_depth: 2
    reconstruction_module_time_steps: 8
    reconstruction_module_conv_dim: 2
    reconstruction_module_num_cascades: 1
    reconstruction_module_dimensionality: 2
    reconstruction_module_no_dc: true
    reconstruction_module_keep_prediction: true
    reconstruction_module_accumulate_predictions: true
    quantitative_module_recurrent_layer: GRU
    quantitative_module_conv_filters:
        - 64
        - 64
        - 4
    quantitative_module_conv_kernels:
        - 5
        - 3
        - 3
    quantitative_module_conv_dilations:
        - 1
        - 2
        - 1
    quantitative_module_conv_bias:
        - true
        - true
        - false
    quantitative_module_recurrent_filters:
        - 64
        - 64
        - 0
    quantitative_module_recurrent_kernels:
        - 1
        - 1
        - 0
    quantitative_module_recurrent_dilations:
        - 1
        - 1
        - 0
    quantitative_module_recurrent_bias:
        - true
        - true
        - false
    quantitative_module_depth: 2
    quantitative_module_time_steps: 8
    quantitative_module_conv_dim: 2
    quantitative_module_num_cascades: 1
    quantitative_module_no_dc: true
    quantitative_module_keep_prediction: true
    quantitative_module_accumulate_predictions: true
    quantitative_module_signal_forward_model_sequence: MEGRE
    quantitative_module_dimensionality: 2
    quantitative_module_gamma_regularization_factors:
        - 150.0
        - 150.0
        - 1000.0
        - 150.0
    # task & dataset related parameters
    coil_combination_method: SENSE
    coil_dim: 1
    complex_valued_type: stacked  # stacked, complex_abs, complex_sqrt_abs
    consecutive_slices: 1
    dimensionality: 2
    estimate_coil_sensitivity_maps_with_nn: false
    fft_centered: false
    fft_normalization: backward
    spatial_dims:
        - -2
        - -1
    shift_B0_input: false
    normalization_type: minmax
    unnormalize_loss_inputs: false
    unnormalize_log_outputs: false

quantitative End-to-End Variational Network

quantitative End-to-End Variational Network (qVarNet), as presented in [Zhang2022].

Zhang2022

Zhang C, Karkalousos D, Bazin PL, Coolen BF, Vrenken H, Sonke JJ, Forstmann BU, Poot DH, Caan MW. A unified model for reconstruction and R2* mapping of accelerated 7T data using the quantitative recurrent inference machine. NeuroImage. 2022 Dec 1;264:119680.

Example configuration:

model:
    model_name: qVN
    use_reconstruction_module: false
    reconstruction_module_num_cascades: 2
    reconstruction_module_channels: 8
    reconstruction_module_pooling_layers: 2
    reconstruction_module_in_channels: 2
    reconstruction_module_out_channels: 2
    reconstruction_module_padding_size: 11
    reconstruction_module_normalize: true
    reconstruction_module_no_dc: false
    reconstruction_module_accumulate_predictions: false
    quantitative_module_num_cascades: 1
    quantitative_module_channels: 4
    quantitative_module_pooling_layers: 2
    quantitative_module_in_channels: 8
    quantitative_module_out_channels: 8
    quantitative_module_padding_size: 11
    quantitative_module_normalize: true
    quantitative_module_no_dc: false
    quantitative_module_dimensionality: 2
    quantitative_module_accumulate_predictions: false
    quantitative_module_signal_forward_model_sequence: MEGRE
    quantitative_module_gamma_regularization_factors:
        - 150.0
        - 150.0
        - 1000.0
        - 150.0
    # task & dataset related parameters
    coil_combination_method: SENSE
    coil_dim: 1
    complex_valued_type: stacked  # stacked, complex_abs, complex_sqrt_abs
    consecutive_slices: 1
    dimensionality: 2
    estimate_coil_sensitivity_maps_with_nn: false
    fft_centered: false
    fft_normalization: backward
    spatial_dims:
        - -2
        - -1
    shift_B0_input: false
    normalization_type: minmax
    unnormalize_loss_inputs: false
    unnormalize_log_outputs: false

MRI Reconstruction (REC)

Cascades of Independently Recurrent Inference Machines

Cascades of Independently Recurrent Inference Machines (CIRIM), as presented in [Karkalousos2022].

Karkalousos2022

Karkalousos D, Noteboom S, Hulst HE, Vos FM, Caan MWA. Assessment of data consistency through cascades of independently recurrent inference machines for fast and robust accelerated MRI reconstruction. Phys Med Biol. 2022 Jun 8;67(12). doi: 10.1088/1361-6560/ac6cc2. PMID: 35508147.

Example configuration:

model:
    model_name: CIRIM
    recurrent_layer: IndRNN
    conv_filters:
        - 64
        - 64
        - 2
    conv_kernels:
        - 5
        - 3
        - 3
    conv_dilations:
        - 1
        - 2
        - 1
    conv_bias:
        - true
        - true
        - false
    recurrent_filters:
        - 64
        - 64
        - 0
    recurrent_kernels:
        - 1
        - 1
        - 0
    recurrent_dilations:
        - 1
        - 1
        - 0
    recurrent_bias:
        - true
        - true
        - false
    depth: 2
    time_steps: 8
    conv_dim: 2
    num_cascades: 8
    no_dc: true
    keep_prediction: true
    accumulate_predictions: true
    # task & dataset related parameters
    coil_combination_method: SENSE
    coil_dim: 1
    complex_valued_type: stacked  # stacked, complex_abs, complex_sqrt_abs
    consecutive_slices: 1
    dimensionality: 2
    estimate_coil_sensitivity_maps_with_nn: false
    fft_centered: false
    fft_normalization: backward
    spatial_dims:
        - -2
        - -1
    normalization_type: minmax
    unnormalize_loss_inputs: false
    unnormalize_log_outputs: false

Convolutional Recurrent Neural Networks

Convolutional Recurrent Neural Networks (CRNNet), as presented in [Qin2019].

Qin2019

C. Qin, J. Schlemper, J. Caballero, A. N. Price, J. V. Hajnal and D. Rueckert, “Convolutional Recurrent Neural Networks for Dynamic MR Image Reconstruction,” in IEEE Transactions on Medical Imaging, vol. 38, no. 1, pp. 280-290, Jan. 2019, doi: 10.1109/TMI.2018.2863670.

Example configuration:

model:
    model_name: CRNNet
    num_iterations: 10
    hidden_channels: 64
    n_convs: 3
    batchnorm: false
    no_dc: false
    accumulate_predictions: true
    # task & dataset related parameters
    coil_combination_method: SENSE
    coil_dim: 1
    complex_valued_type: stacked  # stacked, complex_abs, complex_sqrt_abs
    consecutive_slices: 1
    dimensionality: 2
    estimate_coil_sensitivity_maps_with_nn: false
    fft_centered: false
    fft_normalization: backward
    spatial_dims:
        - -2
        - -1
    normalization_type: minmax
    unnormalize_loss_inputs: false
    unnormalize_log_outputs: false

Deep Cascade of Convolutional Neural Networks

Deep Cascade of Convolutional Neural Networks (CascadeNet), as presented in [Schlemper2017].

Schlemper2017

Schlemper, J., Caballero, J., Hajnal, J. V., Price, A., & Rueckert, D., A Deep Cascade of Convolutional Neural Networks for MR Image Reconstruction. Information Processing in Medical Imaging (IPMI), 2017.

Example configuration:

model:
    model_name: CascadeNet
    num_cascades: 10
    hidden_channels: 64
    n_convs: 5
    batchnorm: false
    no_dc: false
    accumulate_predictions: false
    # task & dataset related parameters
    coil_combination_method: SENSE
    coil_dim: 1
    complex_valued_type: stacked  # stacked, complex_abs, complex_sqrt_abs
    consecutive_slices: 1
    dimensionality: 2
    estimate_coil_sensitivity_maps_with_nn: false
    fft_centered: false
    fft_normalization: backward
    spatial_dims:
        - -2
        - -1
    normalization_type: minmax
    unnormalize_loss_inputs: false
    unnormalize_log_outputs: false

Down-Up Net

Down-Up NET (DUNet), inspired by [Hammernik2021].

Hammernik2021

Hammernik, K, Schlemper, J, Qin, C, et al. Systematic valuation of iterative deep neural networks for fast parallel MRI reconstruction with sensitivity-weighted coil combination. Magn Reson Med. 2021; 86: 1859– 1872. https://doi.org/10.1002/mrm.28827

Example configuration:

model:
    model_name: DUNet
    num_iter: 10
    reg_model_architecture: DIDN
    didn_hidden_channels: 64
    didn_num_dubs: 2
    didn_num_convs_recon: 1
    data_consistency_term: VS
    data_consistency_lambda_init: 0.1
    data_consistency_iterations: 10
    shared_params: false
    # task & dataset related parameters
    coil_combination_method: SENSE
    coil_dim: 1
    complex_valued_type: stacked  # stacked, complex_abs, complex_sqrt_abs
    consecutive_slices: 1
    dimensionality: 2
    estimate_coil_sensitivity_maps_with_nn: false
    fft_centered: false
    fft_normalization: backward
    spatial_dims:
        - -2
        - -1
    normalization_type: minmax
    unnormalize_loss_inputs: false
    unnormalize_log_outputs: false

End-to-End Variational Network

End-to-End Variational Network (VarNet), as presented in [Sriram2020].

Sriram2020

Sriram A, Zbontar J, Murrell T, Defazio A, Zitnick CL, Yakubova N, Knoll F, Johnson P. End-to-end variational networks for accelerated MRI reconstruction. InInternational Conference on Medical Image Computing and Computer-Assisted Intervention 2020 Oct 4 (pp. 64-73). Springer, Cham.

Example configuration:

model:
    model_name: VN
    num_cascades: 8
    channels: 18
    pooling_layers: 4
    padding_size: 11
    normalize: true
    no_dc: false
    # task & dataset related parameters
    coil_combination_method: SENSE
    coil_dim: 1
    complex_valued_type: stacked  # stacked, complex_abs, complex_sqrt_abs
    consecutive_slices: 1
    dimensionality: 2
    estimate_coil_sensitivity_maps_with_nn: false
    fft_centered: false
    fft_normalization: backward
    spatial_dims:
        - -2
        - -1
    normalization_type: minmax
    unnormalize_loss_inputs: false
    unnormalize_log_outputs: false

Independently Recurrent Inference Machines

Independently Recurrent Inference Machines (RIMBlock), as presented in [Karkalousos2022].

Karkalousos2022

Karkalousos D, Noteboom S, Hulst HE, Vos FM, Caan MWA. Assessment of data consistency through cascades of independently recurrent inference machines for fast and robust accelerated MRI reconstruction. Phys Med Biol. 2022 Jun 8;67(12). doi: 10.1088/1361-6560/ac6cc2. PMID: 35508147.

Example configuration:

model:
    model_name: CIRIM
    recurrent_layer: IndRNN
    conv_filters:
        - 64
        - 64
        - 2
    conv_kernels:
        - 5
        - 3
        - 3
    conv_dilations:
        - 1
        - 2
        - 1
    conv_bias:
        - true
        - true
        - false
    recurrent_filters:
        - 64
        - 64
        - 0
    recurrent_kernels:
        - 1
        - 1
        - 0
    recurrent_dilations:
        - 1
        - 1
        - 0
    recurrent_bias:
        - true
        - true
        - false
    depth: 2
    time_steps: 8
    conv_dim: 2
    num_cascades: 1
    no_dc: true
    keep_prediction: true
    accumulate_predictions: true
    # task & dataset related parameters
    coil_combination_method: SENSE
    coil_dim: 1
    complex_valued_type: stacked  # stacked, complex_abs, complex_sqrt_abs
    consecutive_slices: 1
    dimensionality: 2
    estimate_coil_sensitivity_maps_with_nn: false
    fft_centered: false
    fft_normalization: backward
    spatial_dims:
        - -2
        - -1
    normalization_type: minmax
    unnormalize_loss_inputs: false
    unnormalize_log_outputs: false

Joint Deep Model-Based MR Image and Coil Sensitivity Reconstruction Network

Joint Deep Model-Based MR Image and Coil Sensitivity Reconstruction Network (JointICNet), as presented in [Jun2021].

Jun2021

Jun, Yohan, et al. “Joint Deep Model-Based MR Image and Coil Sensitivity Reconstruction Network (Joint-ICNet) for Fast MRI.” 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), IEEE, 2021, pp. 5266–75. DOI.org (Crossref), https://doi.org/10.1109/CVPR46437.2021.00523.

Example configuration:

model:
    model_name: JointICNet
    num_iter: 2
    kspace_unet_num_filters: 16
    kspace_unet_num_pool_layers: 2
    kspace_unet_dropout_probability: 0.0
    kspace_unet_padding_size: 11
    kspace_unet_normalize: true
    imspace_unet_num_filters: 16
    imspace_unet_num_pool_layers: 2
    imspace_unet_dropout_probability: 0.0
    imspace_unet_padding_size: 11
    imspace_unet_normalize: true
    sens_unet_num_filters: 16
    sens_unet_num_pool_layers: 2
    sens_unet_dropout_probability: 0.0
    sens_unet_padding_size: 11
    sens_unet_normalize: true
    # task & dataset related parameters
    coil_combination_method: SENSE
    coil_dim: 1
    complex_valued_type: stacked  # stacked, complex_abs, complex_sqrt_abs
    consecutive_slices: 1
    dimensionality: 2
    estimate_coil_sensitivity_maps_with_nn: false
    fft_centered: false
    fft_normalization: backward
    spatial_dims:
        - -2
        - -1
    normalization_type: minmax
    unnormalize_loss_inputs: false
    unnormalize_log_outputs: false

KIKINet

KIKINet (KIKINet), modified to work with multi-coil k-space data, as presented in [Taejoon2018].

Taejoon2018

Eo, Taejoon, et al. “KIKI-Net: Cross-Domain Convolutional Neural Networks for Reconstructing Undersampled Magnetic Resonance Images.” Magnetic Resonance in Medicine, vol. 80, no. 5, Nov. 2018, pp. 2188–201. PubMed, https://doi.org/10.1002/mrm.27201.

Example configuration:

model:
    model_name: KIKINet
    num_iter: 2
    kspace_model_architecture: UNET
    kspace_in_channels: 2
    kspace_out_channels: 2
    kspace_unet_num_filters: 16
    kspace_unet_num_pool_layers: 2
    kspace_unet_dropout_probability: 0.0
    kspace_unet_padding_size: 11
    kspace_unet_normalize: true
    imspace_model_architecture: UNET
    imspace_in_channels: 2
    imspace_unet_num_filters: 16
    imspace_unet_num_pool_layers: 2
    imspace_unet_dropout_probability: 0.0
    imspace_unet_padding_size: 11
    imspace_unet_normalize: true
    # task & dataset related parameters
    coil_combination_method: SENSE
    coil_dim: 1
    complex_valued_type: stacked  # stacked, complex_abs, complex_sqrt_abs
    consecutive_slices: 1
    dimensionality: 2
    estimate_coil_sensitivity_maps_with_nn: false
    fft_centered: false
    fft_normalization: backward
    spatial_dims:
        - -2
        - -1
    normalization_type: minmax
    unnormalize_loss_inputs: false
    unnormalize_log_outputs: false

Learned Primal-Dual Net

Learned Primal-Dual Net (LPDNet), as presented in [Adler2018].

Adler2018

Adler, Jonas, and Ozan Öktem. “Learned Primal-Dual Reconstruction.” IEEE Transactions on Medical Imaging, vol. 37, no. 6, June 2018, pp. 1322–32. arXiv.org, https://doi.org/10.1109/TMI.2018.2799231.

Example configuration:

model:
    model_name: LPDNet
    num_primal: 5
    num_dual: 5
    num_iter: 5
    primal_model_architecture: UNET
    primal_in_channels: 2
    primal_out_channels: 2
    primal_unet_num_filters: 16
    primal_unet_num_pool_layers: 2
    primal_unet_dropout_probability: 0.0
    primal_unet_padding_size: 11
    primal_unet_normalize: true
    dual_model_architecture: UNET
    dual_in_channels: 2
    dual_out_channels: 2
    dual_unet_num_filters: 16
    dual_unet_num_pool_layers: 2
    dual_unet_dropout_probability: 0.0
    dual_unet_padding_size: 11
    dual_unet_normalize: true
    # task & dataset related parameters
    coil_combination_method: SENSE
    coil_dim: 1
    complex_valued_type: stacked  # stacked, complex_abs, complex_sqrt_abs
    consecutive_slices: 1
    dimensionality: 2
    estimate_coil_sensitivity_maps_with_nn: false
    fft_centered: false
    fft_normalization: backward
    spatial_dims:
        - -2
        - -1
    normalization_type: minmax
    unnormalize_loss_inputs: false
    unnormalize_log_outputs: false

MoDL: Model Based Deep Learning Architecture for Inverse Problems

MoDL: Model Based Deep Learning Architecture for Inverse Problems (MoDL).

Adjusted to optionally perform a data consistency step (Conjugate Gradient), as presented in [Aggarwal2018], [Yaman2020]. If dc is set to False, the network will perform a simple residual learning step.

Aggarwal2018

MoDL: Model Based Deep Learning Architecture for Inverse Problems by H.K. Aggarwal, M.P Mani, and Mathews Jacob in IEEE Transactions on Medical Imaging, 2018

Yaman2020

Yaman, B, Hosseini, SAH, Moeller, S, Ellermann, J, Uğurbil, K, Akçakaya, M. Self-supervised learning of physics-guided reconstruction neural networks without fully sampled reference data. Magn Reson Med. 2020; 84: 3172– 3191. https://doi.org/10.1002/mrm.28378

Example configuration:

model:
    model_name: MoDL
    unrolled_iterations: 5
    residual_blocks: 5
    channels: 64
    regularization_factor: 0.1
    penalization_weight: 1.0
    conjugate_gradient_dc: false
    conjugate_gradient_iterations: 1
    # task & dataset related parameters
    coil_combination_method: SENSE
    coil_dim: 1
    complex_valued_type: stacked  # stacked, complex_abs, complex_sqrt_abs
    consecutive_slices: 1
    dimensionality: 2
    estimate_coil_sensitivity_maps_with_nn: false
    fft_centered: false
    fft_normalization: backward
    spatial_dims:
        - -2
        - -1
    normalization_type: minmax
    unnormalize_loss_inputs: false
    unnormalize_log_outputs: false

MultiDomainNet

Feature-level multi-domain module. Inspired by AIRS Medical submission to the FastMRI 2020 challenge.

Example configuration:

model:
    model_name: MultiDomainNet
    standardization: true
    num_filters: 64
    num_pool_layers: 2
    dropout_probability: 0.0
    # task & dataset related parameters
    coil_combination_method: SENSE
    coil_dim: 1
    complex_valued_type: stacked  # stacked, complex_abs, complex_sqrt_abs
    consecutive_slices: 1
    dimensionality: 2
    estimate_coil_sensitivity_maps_with_nn: false
    fft_centered: false
    fft_normalization: backward
    spatial_dims:
        - -2
        - -1
    normalization_type: minmax
    unnormalize_loss_inputs: false
    unnormalize_log_outputs: false

ProximalGradient

Proximal/Conjugate Gradient (ProximalGradient), according to [Aggarwal2018], [Yaman2020].

Aggarwal2018

MoDL: Model Based Deep Learning Architecture for Inverse Problems by H.K. Aggarwal, M.P Mani, and Mathews Jacob in IEEE Transactions on Medical Imaging, 2018

Yaman2020

Yaman, B, Hosseini, SAH, Moeller, S, Ellermann, J, Uğurbil, K, Akçakaya, M. Self-supervised learning of physics-guided reconstruction neural networks without fully sampled reference data. Magn Reson Med. 2020; 84: 3172– 3191. https://doi.org/10.1002/mrm.28378

Example configuration:

model:
    model_name: PROXIMALGRADIENT
    conjugate_gradient_dc: true
    conjugate_gradient_iterations: 10
    # task & dataset related parameters
    coil_combination_method: SENSE
    coil_dim: 1
    complex_valued_type: stacked  # stacked, complex_abs, complex_sqrt_abs
    consecutive_slices: 1
    dimensionality: 2
    estimate_coil_sensitivity_maps_with_nn: false
    fft_centered: false
    fft_normalization: backward
    spatial_dims:
        - -2
        - -1
    normalization_type: minmax
    unnormalize_loss_inputs: false
    unnormalize_log_outputs: false

Recurrent Variational Network

Recurrent Variational Network (RecurrentVarNet), as presented in [Yiasemis2021].

Yiasemis2021

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.

Example configuration:

model:
    model_name: RVN
    in_channels: 2
    recurrent_hidden_channels: 64
    recurrent_num_layers: 4
    num_steps: 8
    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
    accumulate_predictions: false
    # task & dataset related parameters
    coil_combination_method: SENSE
    coil_dim: 1
    complex_valued_type: stacked  # stacked, complex_abs, complex_sqrt_abs
    consecutive_slices: 1
    dimensionality: 2
    estimate_coil_sensitivity_maps_with_nn: false
    fft_centered: false
    fft_normalization: backward
    spatial_dims:
        - -2
        - -1
    normalization_type: minmax
    unnormalize_loss_inputs: false
    unnormalize_log_outputs: false

Recurrent Inference Machines

Recurrent Inference Machines (RIMBlock), as presented in [Lonning19].

Lonning19

Lønning K, Putzky P, Sonke JJ, Reneman L, Caan MW, Welling M. Recurrent inference machines for reconstructing heterogeneous MRI data. Medical image analysis. 2019 Apr 1;53:64-78.

Example configuration:

model:
    model_name: CIRIM
    recurrent_layer: GRU
    conv_filters:
        - 64
        - 64
        - 2
    conv_kernels:
        - 5
        - 3
        - 3
    conv_dilations:
        - 1
        - 2
        - 1
    conv_bias:
        - true
        - true
        - false
    recurrent_filters:
        - 64
        - 64
        - 0
    recurrent_kernels:
        - 1
        - 1
        - 0
    recurrent_dilations:
        - 1
        - 1
        - 0
    recurrent_bias:
        - true
        - true
        - false
    depth: 2
    time_steps: 8
    conv_dim: 2
    num_cascades: 1
    no_dc: true
    keep_prediction: true
    accumulate_predictions: true
    # task & dataset related parameters
    coil_combination_method: SENSE
    coil_dim: 1
    complex_valued_type: stacked  # stacked, complex_abs, complex_sqrt_abs
    consecutive_slices: 1
    dimensionality: 2
    estimate_coil_sensitivity_maps_with_nn: false
    fft_centered: false
    fft_normalization: backward
    spatial_dims:
        - -2
        - -1
    normalization_type: minmax
    unnormalize_loss_inputs: false
    unnormalize_log_outputs: false

UNet

UNet (UNet), as presented in [Ronneberger2015].

Ronneberger2015

O. Ronneberger, P. Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. In International Conference on Medical image computing and computer-assisted intervention, pages 234–241. Springer, 2015.

Example configuration:

model:
    model_name: UNet
    channels: 64
    pooling_layers: 4
    in_channels: 2
    out_channels: 2
    padding_size: 11
    dropout: 0.0
    normalize: true
    norm_groups: 2
    # task & dataset related parameters
    coil_combination_method: SENSE
    coil_dim: 1
    complex_valued_type: stacked  # stacked, complex_abs, complex_sqrt_abs
    consecutive_slices: 1
    dimensionality: 2
    estimate_coil_sensitivity_maps_with_nn: false
    fft_centered: false
    fft_normalization: backward
    spatial_dims:
        - -2
        - -1
    normalization_type: minmax
    unnormalize_loss_inputs: false
    unnormalize_log_outputs: false

Variable Splitting Network

Variable Splitting Network (VSNet), as presented in [Duan2019].

Duan2019

Duan, J. et al. (2019) Vs-net: Variable splitting network for accelerated parallel MRI reconstruction, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 11767 LNCS, pp. 713–722. doi: 10.1007/978-3-030-32251-9_78.

Example configuration:

model:
    model_name: VSNet
    num_cascades: 10
    imspace_model_architecture: CONV
    imspace_in_channels: 2
    imspace_out_channels: 2
    imspace_conv_hidden_channels: 64
    imspace_conv_n_convs: 4
    imspace_conv_batchnorm: false
    # task & dataset related parameters
    coil_combination_method: SENSE
    coil_dim: 1
    complex_valued_type: stacked  # stacked, complex_abs, complex_sqrt_abs
    consecutive_slices: 1
    dimensionality: 2
    estimate_coil_sensitivity_maps_with_nn: false
    fft_centered: false
    fft_normalization: backward
    spatial_dims:
        - -2
        - -1
    normalization_type: minmax
    unnormalize_loss_inputs: false
    unnormalize_log_outputs: false

XPDNet

XPDNet (XPDNet), as presented in [Ramzi2021].

Ramzi2021

Ramzi, Zaccharie, et al. “XPDNet for MRI Reconstruction: An Application to the 2020 FastMRI Challenge. ArXiv:2010.07290 [Physics, Stat], July 2021. arXiv.org, http://arxiv.org/abs/2010.07290.

Example configuration:

model:
    model_name: XPDNet
    num_primal: 5
    num_dual: 1
    num_iter: 20
    use_primal_only: true
    kspace_model_architecture: CONV
    kspace_in_channels: 2
    kspace_out_channels: 2
    dual_conv_hidden_channels: 16
    dual_conv_num_dubs: 2
    dual_conv_batchnorm: false
    image_model_architecture: MWCNN
    imspace_in_channels: 2
    imspace_out_channels: 2
    mwcnn_hidden_channels: 16
    mwcnn_num_scales: 2
    mwcnn_bias: true
    mwcnn_batchnorm: false
    normalize_image: false
    # task & dataset related parameters
    coil_combination_method: SENSE
    coil_dim: 1
    complex_valued_type: stacked  # stacked, complex_abs, complex_sqrt_abs
    consecutive_slices: 1
    dimensionality: 2
    estimate_coil_sensitivity_maps_with_nn: false
    fft_centered: false
    fft_normalization: backward
    spatial_dims:
        - -2
        - -1
    normalization_type: minmax
    unnormalize_loss_inputs: false
    unnormalize_log_outputs: false

Zero-Filled

Zero-Filled reconstruction using either root-sum-of-squares (RSS) or SENSE (SENSitivity Encoding, as presented in [Pruessmann1999]).

Pruessmann1999

Pruessmann KP, Weiger M, Scheidegger MB, Boesiger P. SENSE: Sensitivity encoding for fast MRI. Magn Reson Med 1999; 42:952-962.

Example configuration:

model:
    model_name: ZF
    # task & dataset related parameters
    coil_combination_method: SENSE
    coil_dim: 1
    complex_valued_type: stacked  # stacked, complex_abs, complex_sqrt_abs
    consecutive_slices: 1
    dimensionality: 2
    estimate_coil_sensitivity_maps_with_nn: false
    fft_centered: false
    fft_normalization: backward
    spatial_dims:
        - -2
        - -1
    normalization_type: minmax
    unnormalize_loss_inputs: false
    unnormalize_log_outputs: false

MRI Segmentation (SEG)

Attention UNet

Attention UNet for MRI segmentation (SegmentationAttentionUNet), as presented in [Oktay2018].

Oktay2018

O. Oktay, J. Schlemper, L.L. Folgoc, M. Lee, M. Heinrich, K. Misawa, K. Mori, S. McDonagh, N.Y. Hammerla, B. Kainz, B. Glocker, D. Rueckert. Attention U-Net: Learning Where to Look for the Pancreas. 2018. https://arxiv.org/abs/1804.03999

Example configuration:

model:
    model_name: SEGMENTATIONATTENTIONUNET
    segmentation_module: AttentionUNet
    segmentation_module_input_channels: 1
    segmentation_module_output_channels: 4
    segmentation_module_channels: 32
    segmentation_module_pooling_layers: 5
    segmentation_module_dropout: 0.0
    # task & dataset related parameters
    coil_combination_method: SENSE  # if complex data
    coil_dim: 1  # if complex data
    complex_data: true  # or false if using magnitude data
    complex_valued_type: stacked (only for complex data)  # stacked, complex_abs, complex_sqrt_abs
    consecutive_slices: 1
    dimensionality: 2
    estimate_coil_sensitivity_maps_with_nn: false
    fft_centered: false  # if complex data
    fft_normalization: backward  # if complex data
    spatial_dims:
        - -2  # if complex data
        - -1  # if complex data
    magnitude_input: true
    normalization_type: minmax
    normalize_segmentation_output: true
    unnormalize_loss_inputs: false
    unnormalize_log_outputs: false

Dynamic UNet

Dynamic UNet for MRI segmentation (SegmentationDYNUNet), as presented in [Isensee2018].

Isensee2018

Isensee F, Petersen J, Klein A, Zimmerer D, Jaeger PF, Kohl S, Wasserthal J, Koehler G, Norajitra T, Wirkert S, Maier-Hein KH. nnu-net: Self-adapting framework for u-net-based medical image segmentation. arXiv preprint arXiv:1809.10486. 2018 Sep 27.

Example configuration:

model:
    model_name: SEGMENTATIONDYNUNET
    segmentation_module: DYNUNet
    segmentation_module_input_channels: 1
    segmentation_module_output_channels: 4
    segmentation_module_channels:
        - 64
        - 128
        - 256
        - 512
    segmentation_module_kernel_size:
        - 3
        - 3
        - 3
        - 1
    segmentation_module_strides:
        - 1
        - 1
        - 1
        - 1
    segmentation_module_dropout: 0.0
    segmentation_module_norm: instance
    segmentation_module_activation: leakyrelu
    segmentation_module_deep_supervision: true
    segmentation_module_deep_supervision_levels: 2
    # task & dataset related parameters
    coil_combination_method: SENSE  # if complex data
    coil_dim: 1  # if complex data
    complex_data: true  # or false if using magnitude data
    complex_valued_type: stacked (only for complex data)  # stacked, complex_abs, complex_sqrt_abs
    consecutive_slices: 1
    dimensionality: 2
    estimate_coil_sensitivity_maps_with_nn: false
    fft_centered: false  # if complex data
    fft_normalization: backward  # if complex data
    spatial_dims:
        - -2  # if complex data
        - -1  # if complex data
    magnitude_input: true
    normalization_type: minmax
    normalize_segmentation_output: true
    unnormalize_loss_inputs: false
    unnormalize_log_outputs: false

Lambda UNet

Lambda UNet for MRI segmentation (SegmentationLambdaUNet), as presented in [Yanglan2021].

Yanglan2021

Yanglan Ou, Ye Yuan, Xiaolei Huang, Kelvin Wong, John Volpi, James Z. Wang, Stephen T.C. Wong. LambdaUNet: 2.5D Stroke Lesion Segmentation of Diffusion-weighted MR Images. 2021. https://arxiv.org/abs/2104.13917

Example configuration:

model:
    model_name: SEGMENTATIONLAMBDAUNET
    segmentation_module: LambdaUNet
    segmentation_module_input_channels: 1
    segmentation_module_output_channels: 4
    segmentation_module_channels: 64
    segmentation_module_pooling_layers: 2
    segmentation_module_dropout: 0.0
    segmentation_module_query_depth: 16
    segmentation_module_intra_depth: 1
    segmentation_module_receptive_kernel: 1
    segmentation_module_temporal_kernel: 1
    # task & dataset related parameters
    coil_combination_method: SENSE  # if complex data
    coil_dim: 1  # if complex data
    complex_data: true  # or false if using magnitude data
    complex_valued_type: stacked (only for complex data)  # stacked, complex_abs, complex_sqrt_abs
    consecutive_slices: 1
    dimensionality: 2
    estimate_coil_sensitivity_maps_with_nn: false
    fft_centered: false  # if complex data
    fft_normalization: backward  # if complex data
    spatial_dims:
        - -2  # if complex data
        - -1  # if complex data
    magnitude_input: true
    normalization_type: minmax
    normalize_segmentation_output: true
    unnormalize_loss_inputs: false
    unnormalize_log_outputs: false

UNet

2D UNet for MRI segmentation (SegmentationUNet), as presented in [Ronneberger2015].

Ronneberger2015

O. Ronneberger, P. Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. In International Conference on Medical image computing and computer-assisted intervention, pages 234–241. Springer, 2015.

Example configuration:

model:
    model_name: SEGMENTATIONUNET
    segmentation_module: UNet
    segmentation_module_input_channels: 1
    segmentation_module_output_channels: 4
    segmentation_module_channels: 64
    segmentation_module_pooling_layers: 2
    segmentation_module_dropout: 0.0
    # task & dataset related parameters
    coil_combination_method: SENSE  # if complex data
    coil_dim: 1  # if complex data
    complex_data: true  # or false if using magnitude data
    complex_valued_type: stacked (only for complex data)  # stacked, complex_abs, complex_sqrt_abs
    consecutive_slices: 1
    dimensionality: 2
    estimate_coil_sensitivity_maps_with_nn: false
    fft_centered: false  # if complex data
    fft_normalization: backward  # if complex data
    spatial_dims:
        - -2  # if complex data
        - -1  # if complex data
    magnitude_input: true
    normalization_type: minmax
    normalize_segmentation_output: true
    unnormalize_loss_inputs: false
    unnormalize_log_outputs: false

UNet 3D

3D UNet for MRI segmentation (Segmentation3DUNet), as presented in [Ronneberger2015].

Ronneberger2015

O. Ronneberger, P. Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. In International Conference on Medical image computing and computer-assisted intervention, pages 234–241. Springer, 2015.

Example configuration:

model:
    model_name: SEGMENTATION3DUNET
    segmentation_module: UNet
    segmentation_module_input_channels: 1
    segmentation_module_output_channels: 4
    segmentation_module_channels: 64
    segmentation_module_pooling_layers: 2
    segmentation_module_dropout: 0.0
    # task & dataset related parameters
    coil_combination_method: SENSE  # if complex data
    coil_dim: 1  # if complex data
    complex_data: true  # or false if using magnitude data
    complex_valued_type: stacked (only for complex data)  # stacked, complex_abs, complex_sqrt_abs
    consecutive_slices: 1
    dimensionality: 2
    estimate_coil_sensitivity_maps_with_nn: false
    fft_centered: false  # if complex data
    fft_normalization: backward  # if complex data
    spatial_dims:
        - -2  # if complex data
        - -1  # if complex data
    magnitude_input: true
    normalization_type: minmax
    normalize_segmentation_output: true
    unnormalize_loss_inputs: false
    unnormalize_log_outputs: false

UNETR

UNETR for MRI segmentation (SegmentationUNetR), as presented in [Hatamizadeh2022].

Hatamizadeh2022

Hatamizadeh A, Tang Y, Nath V, Yang D, Myronenko A, Landman B, Roth HR, Xu D. Unetr: Transformers for 3d medical image segmentation. InProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision 2022 (pp. 574-584).

Example configuration:

model:
    model_name: SEGMENTATIONUNETR
    segmentation_module: UNETR
    segmentation_module_input_channels: 1
    segmentation_module_output_channels: 3
    segmentation_module_img_size: (256, 256)
    segmentation_module_channels: 64
    segmentation_module_hidden_size: 768
    segmentation_module_mlp_dim: 3072
    segmentation_module_num_heads: 12
    segmentation_module_pos_embed: conv
    segmentation_module_norm_name: instance
    segmentation_module_conv_block: true
    segmentation_module_res_block: true
    segmentation_module_dropout: 0.0
    segmentation_module_qkv_bias: false
    # task & dataset related parameters
    coil_combination_method: SENSE  # if complex data
    coil_dim: 1  # if complex data
    complex_data: true  # or false if using magnitude data
    complex_valued_type: stacked (only for complex data)  # stacked, complex_abs, complex_sqrt_abs
    consecutive_slices: 1
    dimensionality: 2
    estimate_coil_sensitivity_maps_with_nn: false
    fft_centered: false  # if complex data
    fft_normalization: backward  # if complex data
    spatial_dims:
        - -2  # if complex data
        - -1  # if complex data
    magnitude_input: true
    normalization_type: minmax
    normalize_segmentation_output: true
    unnormalize_loss_inputs: false
    unnormalize_log_outputs: false

V-Net

V-Net for MRI segmentation (SegmentationVNet), as presented in [Milletari2016].

Milletari2016

Fausto Milletari, Nassir Navab, Seyed-Ahmad Ahmadi. V-Net: Fully Convolutional Neural Networks for Volumetric Medical Image Segmentation, 2016. https://arxiv.org/abs/1606.04797

Example configuration:

model:
    use_reconstruction_module: false
    segmentation_module: VNet
    segmentation_module_input_channels: 1
    segmentation_module_output_channels: 4
    segmentation_module_activation: elu
    segmentation_module_dropout: 0.0
    segmentation_module_bias: False
    segmentation_module_padding_size: 15
    # task & dataset related parameters
    coil_combination_method: SENSE  # if complex data
    coil_dim: 1  # if complex data
    complex_data: true  # or false if using magnitude data
    complex_valued_type: stacked (only for complex data)  # stacked, complex_abs, complex_sqrt_abs
    consecutive_slices: 1
    dimensionality: 2
    estimate_coil_sensitivity_maps_with_nn: false
    fft_centered: false  # if complex data
    fft_normalization: backward  # if complex data
    spatial_dims:
        - -2  # if complex data
        - -1  # if complex data
    magnitude_input: true
    normalization_type: minmax
    normalize_segmentation_output: true
    unnormalize_loss_inputs: false
    unnormalize_log_outputs: false