XPDNet for 5x & 10x accelerated MRI Reconstruction on the CC359 dataset.
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.
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.
1pretrained: true
2checkpoint: https://huggingface.co/wdika/REC_XPDNet_CC359_12_channel_poisson2d_5x_10x_NNEstimationCSM/blob/main/REC_XPDNet_CC359_12_channel_poisson2d_5x_10x_NNEstimationCSM.atommic
3mode: test
You need to download the CC359 dataset to effectively use this model. Check the
CC359 page for more information.
1model:
2 model_name: XPDNet
3 num_primal: 5
4 num_dual: 1
5 num_iter: 10
6 use_primal_only: true
7 kspace_model_architecture: CONV
8 kspace_in_channels: 2
9 kspace_out_channels: 2
10 dual_conv_hidden_channels: 16
11 dual_conv_num_dubs: 2
12 dual_conv_batchnorm: false
13 image_model_architecture: MWCNN
14 imspace_in_channels: 2
15 imspace_out_channels: 2
16 mwcnn_hidden_channels: 16
17 mwcnn_num_scales: 0
18 mwcnn_bias: true
19 mwcnn_batchnorm: false
20 normalize_image: true
21 dimensionality: 2
22 reconstruction_loss:
23 l1: 0.1
24 ssim: 0.9
25 estimate_coil_sensitivity_maps_with_nn: true
1 optim:
2 name: adamw
3 lr: 1e-4
4 betas:
5 - 0.9
6 - 0.999
7 weight_decay: 0.0
8 sched:
9 name: CosineAnnealing
10 min_lr: 0.0
11 last_epoch: -1
12 warmup_ratio: 0.1
13
14trainer:
15 strategy: ddp_find_unused_parameters_false
16 accelerator: gpu
17 devices: 1
18 num_nodes: 1
19 max_epochs: 20
20 precision: 16-mixed
21 enable_checkpointing: false
22 logger: false
23 log_every_n_steps: 50
24 check_val_every_n_epoch: -1
25 max_steps: -1
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.
5x: MSE = 0.004192 +/- 0.004255 NMSE = 0.06401 +/- 0.06475 PSNR = 24.27 +/- 4.135 SSIM = 0.7609 +/- 0.09962
10x: MSE = 0.00581 +/- 0.00445 NMSE = 0.08987 +/- 0.07376 PSNR = 22.65 +/- 3.225 SSIM = 0.6997 +/- 0.1119
This model was trained on the CC359 using a UNet coil sensitivity maps estimation and might differ from the results reported on the challenge leaderboard.
[2] Beauferris, Y., Teuwen, J., Karkalousos, D., Moriakov, N., Caan, M., Yiasemis, G., Rodrigues, L., Lopes, A., Pedrini, H., Rittner, L., Dannecker, M., Studenyak, V., Gröger, F., Vyas, D., Faghih-Roohi, S., Kumar Jethi, A., Chandra Raju, J., Sivaprakasam, M., Lasby, M., … Souza, R. (2022). Multi-Coil MRI Reconstruction Challenge—Assessing Brain MRI Reconstruction Models and Their Generalizability to Varying Coil Configurations. Frontiers in Neuroscience, 16.
https://doi.org/10.3389/fnins.2022.919186
