Model Overview
Learned Primal Dual Network (LPDNet) for 4x & 8x accelerated MRI Reconstruction on the fastMRIBrainsMulticoil dataset.
ATOMMIC: Training
To train, fine-tune, or test the model you will need to install ATOMMIC. We recommend you install it after you've installed latest Pytorch version.
pip install atommic['all']
How to Use this Model
The model is available for use in ATOMMIC, and can be used as a pre-trained checkpoint for inference or for fine-tuning on another dataset.
Corresponding configuration YAML files can be found here.
Automatically instantiate the model
pretrained: true
checkpoint: https://huggingface.co/wdika/REC_LPDNet_fastMRIBrainsMulticoil_equispaced_4x_8x_GDCC_1_coil_NNEstimationCSM/blob/main/REC_LPDNet_fastMRIBrainsMulticoil_equispaced_4x_8x_GDCC_1_coil_NNEstimationCSM.atommic
mode: test
Usage
You need to download the fastMRI Brains dataset to effectively use this model. Check the fastMRIBrainsMulticoil page for more information.
Model Architecture
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
dimensionality: 2
reconstruction_loss:
l1: 0.1
ssim: 0.9
estimate_coil_sensitivity_maps_with_nn: true
Training
optim:
name: adam
lr: 1e-4
betas:
- 0.9
- 0.999
weight_decay: 0.0
sched:
name: InverseSquareRootAnnealing
min_lr: 0.0
last_epoch: -1
warmup_ratio: 0.1
trainer:
strategy: ddp_find_unused_parameters_false
accelerator: gpu
devices: 1
num_nodes: 1
max_epochs: 20
precision: 16-mixed
enable_checkpointing: false
logger: false
log_every_n_steps: 50
check_val_every_n_epoch: -1
max_steps: -1
Performance
To compute the targets using the raw k-space and the chosen coil combination method, accompanied with the chosen coil sensitivity maps estimation method, you can use targets configuration files.
Evaluation can be performed using the evaluation script for the reconstruction task, with --evaluation_type per_slice.
Results
Evaluation against RSS targets
4x: MSE = 0.000939 +/- 0.004162 NMSE = 0.02527 +/- 0.09819 PSNR = 32.6 +/- 6.781 SSIM = 0.8815 +/- 0.2009
8x: MSE = 0.001548 +/- 0.00446 NMSE = 0.04132 +/- 0.1069 PSNR = 29.51 +/- 5.934 SSIM = 0.8401 +/- 0.2084
Limitations
This model was trained on the fastMRIBrainsMulticoil batch0 dataset using a UNet coil sensitivity maps estimation and Geometric Decomposition Coil-Compressions to 1-coil, and might differ from the results reported on the challenge leaderboard.
References
[1] ATOMMIC
[2] Muckley MJ, Riemenschneider B, Radmanesh A, Kim S, Jeong G, Ko J, Jun Y, Shin H, Hwang D, Mostapha M, Arberet S, Nickel D, Ramzi Z, Ciuciu P, Starck JL, Teuwen J, Karkalousos D, Zhang C, Sriram A, Huang Z, Yakubova N, Lui YW, Knoll F. Results of the 2020 fastMRI Challenge for Machine Learning MR Image Reconstruction. IEEE Trans Med Imaging. 2021 Sep;40(9):2306-2317. doi: 10.1109/TMI.2021.3075856. Epub 2021 Aug 31. PMID: 33929957; PMCID: PMC8428775.