Model Overview
quantitative Cascades of Independently Recurrent Inference Machines (qCIRIM) for 12x accelerated quantitative MRI mapping of R2*, S0, B0, phi maps on the AHEAD 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/QMRI_qCIRIM_AHEAD_gaussian2d_12x/blob/main/QMRI_qCIRIM_AHEAD_gaussian2d_12x.atommic
mode: test
Usage
You need to download the AHEAD dataset to effectively use this model. Check the AHEAD page for more information.
Model Architecture
model:
model_name: qCIRIM
use_reconstruction_module: false
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: 5
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_maps_scaling_factor: 1e-3
quantitative_maps_regularization_factors:
- 150.0
- 150.0
- 1000.0
- 150.0
quantitative_loss:
ssim: 1.0
kspace_quantitative_loss: false
total_quantitative_loss_weight: 1.0 # balance between reconstruction and quantitative loss
quantitative_parameters_regularization_factors:
- R2star: 1.0
- S0: 1.0
- B0: 1.0
- phi: 1.0
Training
optim:
name: adam
lr: 1e-4
betas:
- 0.9
- 0.98
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 qmri task, with --evaluation_type per_slice.
Results
Evaluation against R2*, S0, B0, phi targets
12x: MSE = 0.004702 +/- 0.02991 NMSE = 0.1239 +/- 0.3383 PSNR = 28.28 +/- 11.31 SSIM = 0.8814 +/- 0.1774
Limitations
This model was trained on very few subjects on the AHEAD dataset. It is not guaranteed to generalize to other datasets.
References
[1] ATOMMIC
[2] Alkemade A, Mulder MJ, Groot JM, et al. The Amsterdam Ultra-high field adult lifespan database (AHEAD): A freely available multimodal 7 Tesla submillimeter magnetic resonance imaging database. NeuroImage 2020;221.