FetMRQC [paper1,paper2] is a tool for quality assessment (QA) and quality control (QC) of T2-weighted (T2w) fetal brain MR images. It consists of two parts.
- A rating interface (visual report) to standardize and facilitate quality annotations of T2w fetal brain MRI images, by creating interactive HTML-based visual reports from fetal brain scans. It uses a pair of low-resolution (LR) T2w images with corresponding brain masks to provide snapshots of the brain in the three orientations of the acquisition in the subject-space.
- A QA/QC model that can predict the quality of given T2w scans. It works on BIDS-formatted datasets.
While raw data cannot be shared, the extracted image quality metrics are available on Zenodo.
Disclaimer. Fetal brain QC is not intended for clinical use.
Contents
FetMRQC is available as a docker image on docker hub.
You can follow this link to install Docker and NVIDIA Container Toolkit. After installing docker, you can download FetMRQC with
docker pull thsanchez/fetmrqc:0.1.3
This image was built using Ubuntu 22.04 and CUDA 12.1. Note that the image is heavy, around 35GB.
You have two options to run the FetMRQC docker. A wrapper script run_docker.py is provided in the FetMRQC repository and can run both main pipelines: the reports pipeline that generates visual reports from which quality can be assessed and the inference pipeline that computes the image quality metrics and performs inference using a pre-trained FetMRQC model. Both start from a BIDS formatted directory of raw T2-weighted stacks of 2D slices. More details are available in the usage section.
The run_docker.py script automatically handles the mounting of folders onto the docker container, and can help people unfamiliar with docker containers. For those who are familiar with docker, we provide an example of how the inference pipeline can be run by directly calling the docker
docker run --rm -it \
--gpus all --gpus all --ipc=host \
--ulimit memlock=-1 --ulimit stack=67108864 \
-v <local BIDS data folder>:/data/data \
-v <local masks folder>:/data/masks \
-v <local masks folder>:/data/seg \
-v <local output folder>:/data/out \
thsanchez/fetmrqc:0.1.3 qc_inference_pipeline \
--bids_dir /data/data \
--masks_dir /data/masks \
--seg_dir /data/out/seg \
--bids_csv /data/out/bids_csv.csv \
--iqms_csv /data/out/iqms_csv.csv \
--out_csv /data/out/out_csv.csv \
--fetmrqc20_iqms \
--device <cpu or cuda>
FetMRQC (fetal_brain_qc) was developed in Ubuntu 22.04 and tested for python 3.9.15
To install this repository, first clone it via
git clone [email protected]:Medical-Image-Analysis-Laboratory/fetmrqc.git
and enter into the directory. Create a conda environment using conda env create -n fetal_brain_qc python=3.9.15 Activate it using conda activate fetal_brain_qc and then install the repository using python -m pip install -e .
FetMRQC relies on pretrained models from various other works. In order to run it from source, you will need to download them and place them into the fetal_brain_qc/models folder.
- MONAIfbs [2]. Download the pretrained model from Zenodo, and to add it to
fetal_brain_qc/models/MONAIfbs_dynunet_ckpt.pt. - fetal-IQA [3,4,5]. Download the checkpoint
pytorch.ckptof fetal-IQA from Zenodo. Rename it tofetal_IQA_pytorch.ckptand put it intofetal_brain_qc/models. - pl-fetal-brain-assessment [6]. Download the checkpoint of pl-fetal-brain-assessment from Zenodo. Rename it to
FNNDSC_qcnet_ckpt.hdf5and put it intofetal_brain_qc/models - Pretrained nnUNetv2 [7] model. Download the checkpoint folder from Zenodo, unzip it and put it into
fetal_brain_qc/models.
Create a new environment for nnUNet (exit the current one by doing conda deactivate), using conda env create -n nnunet and activate it. Then, following the nnUNet installation instructions, install pytorch:
Install PyTorch as described on their website (conda/pip). Please install the latest version with support for your hardware (cuda, mps, cpu). DO NOT JUST pip install nnunetv2 WITHOUT PROPERLY INSTALLING PYTORCH FIRST. For maximum speed, consider compiling pytorch yourself (experienced users only!).
After that, clone the repository from nnUNet V2 (v2.0) using git clone https://github.com/MIC-DKFZ/nnUNet.git. Go into the repository and install it using python -m pip install -e . If the installation is successful, you should be able to call from the command line nnUNetv2_predict.
Note. We used the commit af8c9fb1fe5c695020aa7c30c174a5e61efdd95d of nnUNet.
Note. You do not need to set the paths as told in the nnUNet installation
You can now exit the new environment and re-activate fetal_brain_qc.
Note. This installation is done in order to be able to run qc_compute_segmentation (more on it below), in which you need to provide nnunet_env_path, the path to the nnUNet python environment that you can obtain using conda env list
FetMRQC starts from a BIDS dataset (containing NIfTI formatted images).
For reports generation, the recommended workflow is to use qc_reports_pipeline.
usage: qc_reports_pipeline [-h] --bids_dir BIDS_DIR --masks_dir MASKS_DIR [--reports_dir REPORTS_DIR] [--ckpt_path CKPT_PATH] [--mask_pattern MASK_PATTERN] [--bids_csv BIDS_CSV] [--seed SEED]
Given a `bids_dir`, lists the LR series in the directory, computes the brain masks using MONAIfbs and uses the masks to compute visual reports that can be used for manual rating.
optional arguments:
-h, --help show this help message and exit
--bids_dir BIDS_DIR BIDS directory containing the LR series. (default: None)
--masks_dir MASKS_DIR
Root of the BIDS directory where brain masks will be stored. (default: None)
--reports_dir REPORTS_DIR
Directory where the reports will be stored. (Default is <bids_dir>/derivatives/reports) (default: None)
--ckpt_path CKPT_PATH
Path to the checkpoint of the MONAIfbs model. (default: /home/tsanchez/Documents/mial/repositories/qc_fetal_brain/fetal_brain_qc/models/MONAIfbs_dynunet_ckpt.pt)
--mask_pattern MASK_PATTERN
Pattern according to which the masks will be stored. By default, masks will be stored in "<masks_dir>/sub-{subject}[/ses-{session}][/{datatype}]/sub-{subject}[_ses-{session}][_acq-{acquisition}][_run-{run}]_{suffix}.nii.gz", and the different fields
will be substituted based on the structure of bids_dir. (default: sub-{subject}[/ses-{session}][/{datatype}]/sub-{subject}[_ses-{session}][_acq-{acquisition}][_run-{run}]_{suffix}.nii.gz)
--bids_csv BIDS_CSV CSV file where the list of available LR series and masks will be stored. (default: bids_csv.csv)
--seed SEED Seed for the random number generator. (default: 42)
Remark. This script runs the whole reports generating pipeline of FetMRQC, i.e. brain extraction -> listing of BIDS directory and masks into a CSV file -> Report generation -> Index file generation.
For inference, the recommended workflow is qc_inference_pipeline.
usage: qc_inference_pipeline [-h] --bids_dir BIDS_DIR --masks_dir MASKS_DIR --seg_dir SEG_DIR [--bids_csv BIDS_CSV] [--iqms_csv IQMS_CSV] [--out_csv OUT_CSV] [--fetmrqc20_iqms | --no-fetmrqc20_iqms] [--use_all_iqms] [--ckpt_path CKPT_PATH] [--mask_pattern MASK_PATTERN]
[--seed SEED] [--device {cpu,cuda}]
Given a `bids_dir`, lists the LR series in the directory, computes brain masks and segmentations, uses them to extract IQMs and perform inference using one of the pretrained FetMRQC models. The output is a CSV file containing the predictions and IQMs.
optional arguments:
-h, --help show this help message and exit
--bids_dir BIDS_DIR BIDS directory containing the LR series. (default: None)
--masks_dir MASKS_DIR
Root of the BIDS directory where brain masks will be/are stored. If masks already exist, they will be used. (default: None)
--seg_dir SEG_DIR Root of the directory where brain segmentations will be stored. If segmentations already exist, they will be used. (default: None)
--bids_csv BIDS_CSV CSV file where the list of available LR series and masks will be stored. (default: bids_csv.csv)
--iqms_csv IQMS_CSV CSV file where the computed IQMs will be stored. (default: iqms_csv.csv)
--out_csv OUT_CSV CSV file where the predictions from FetMRQC will be stored. (default: out_csv.csv)
--fetmrqc20_iqms, --no-fetmrqc20_iqms
Whether the IQMs from FetMRQC-20 should be computed (default: True)
--use_all_iqms Whether all IQMs should be computed (default: False)
--ckpt_path CKPT_PATH
Path to the checkpoint of the MONAIfbs model. (default: /home/tsanchez/Documents/mial/repositories/qc_fetal_brain/fetal_brain_qc/models/MONAIfbs_dynunet_ckpt.pt)
--mask_pattern MASK_PATTERN
Pattern according to which the masks will be stored. By default, masks will be stored in "<masks_dir>/sub-{subject}[/ses-{session}][/{datatype}]/sub-{subject}[_ses-{session}][_acq-{acquisition}][_run-{run}]_{suffix}.nii.gz", and the different fields
will be substituted based on the structure of bids_dir. (default: sub-{subject}[/ses-{session}][/{datatype}]/sub-{subject}[_ses-{session}][_acq-{acquisition}][_run-{run}]_{suffix}.nii.gz)
--seed SEED Seed to control the randomization (to be used with randomize=True). (default: 42)
--device {cpu,cuda} Device to use for inference. (default: cuda)
Remark. This script runs the whole inference pipeline of FetMRQC, i.e. brain extraction -> listing of BIDS directory and masks into a CSV file -> Segmentation of the brain -> IQMs extraction -> Inference using a pretrained FetMRQC model
Each part of the pipeline can also be called individually:
qc_brain_extraction:
usage: qc_brain_extraction [-h] --bids_dir BIDS_DIR --masks_dir MASKS_DIR [--ckpt_path CKPT_PATH] [--mask_pattern MASK_PATTERN]
Given a `bids_dir`, lists the LR series in the directory and computes the brain masks using MONAIfbs (https://github.com/gift-surg/MONAIfbs/tree/main). Save the masks into the `masks_dir` folder, following the same hierarchy as the `bids_dir`
optional arguments:
-h, --help show this help message and exit
--bids_dir BIDS_DIR BIDS directory containing the LR series. (default: None)
--masks_dir MASKS_DIR
Root of the BIDS directory where brain masks will be stored. (default: None)
--ckpt_path CKPT_PATH
Path to the checkpoint of the MONAIfbs model. (default: /home/tsanchez/Documents/mial/repositories/qc_fetal_brain/fetal_brain_qc/models/MONAIfbs_dynunet_ckpt.pt)
--mask_pattern MASK_PATTERN
Pattern according to which the masks will be stored. By default, masks will be stored in "<masks_dir>/sub-{subject}[/ses-{session}][/{datatype}]/sub-{subject}[_ses-{session}][_acq-{acquisition}][_run-{run}]_{suffix}.nii.gz", and the different fields
will be substituted based on the structure of bids_dir. (default: sub-{subject}[/ses-{session}][/{datatype}]/sub-{subject}[_ses-{session}][_acq-{acquisition}][_run-{run}]_{suffix}.nii.gz)
qc_list_bids_csv
usage: qc_list_bids_csv [-h] --bids_dir BIDS_DIR [--mask_patterns MASK_PATTERNS [MASK_PATTERNS ...]] [--mask_patterns_base MASK_PATTERNS_BASE [MASK_PATTERNS_BASE ...]] [--out_csv OUT_CSV] [--anonymize_name | --no-anonymize_name] [--suffix SUFFIX] [--seed SEED]
[--skip_masks | --no-skip_masks]
Given a `bids_dir`, lists the LR series in the directory and tries to find corresponding masks given by `mask_patterns`. Then, saves all the found pairs of (LR series, masks) in a CSV file at `out_csv`
optional arguments:
-h, --help show this help message and exit
--bids_dir BIDS_DIR BIDS directory containing the LR series. (default: None)
--mask_patterns MASK_PATTERNS [MASK_PATTERNS ...]
Pattern(s) to find the LR masks corresponding to the LR series. Patterns will be of the form "sub-{subject}[/ses-{session}][/{datatype}]/sub-{subject}[_ses-{session}][_acq-{acquisition}][_run-{run}]_{suffix}.nii.gz", and the different fields will be
substituted based on the structure of bids_dir. The base directory from which the search will be run can be changed with `--mask-pattern-base`. (default: None)
--mask_patterns_base MASK_PATTERNS_BASE [MASK_PATTERNS_BASE ...]
Base folder(s) from which the LR masks must be listed. The method will look for masks at `mask-pattern-base`/`mask-patterns`. In this case, both `mask-patterns` and `mask-pattern-base` should be of the same length. (default: None)
--out_csv OUT_CSV CSV file where the list of available LR series and masks is stored. (default: bids_csv.csv)
--anonymize_name, --no-anonymize_name
Whether an anonymized name must be stored along the paths in `out_csv`. This will determine whether the reports will be anonymous in the end. (default: True)
--suffix SUFFIX Suffix used to query the data (default: T2w)
--seed SEED Seed for the random number generator. (default: None)
--skip_masks, --no-skip_masks
Whether the masks should be skipped. (default: False)
qc_generate_reports
usage: Given a BIDS CSV file, generates visual reports for annotation. [-h] --bids_csv BIDS_CSV --out_dir OUT_DIR [--sr | --no-sr] [--add_js | --no-add_js] [--block_if_exclude | --no-block_if_exclude] [--dataset DATASET]
optional arguments:
-h, --help show this help message and exit
--bids_csv BIDS_CSV Path where the bids config csv file is stored. (default: None)
--out_dir OUT_DIR Path where the reports will be stored. (default: None)
--sr, --no-sr Whether the reports to be generated are for SR data. (default: False)
--add_js, --no-add_js
Whether some javascript should be added to the report for interaction with the index file. (default: True)
--block_if_exclude, --no-block_if_exclude
Whether the ratings should be blocked after the first rating is set below the include threshold. (default: False)
--dataset DATASET Name of the dataset to be used for the reports. (default: None)
qc_generate_index
usage: qc_generate_index [-h] [--reports_dirs REPORTS_DIRS [REPORTS_DIRS ...]] [--add_script_to_reports | --no-add_script_to_reports] [--use_ordering_file | --no-use_ordering_file] [--navigation | --no-navigation] [--seed SEED]
Given a list of reports, generates an index.html file to navigate through them.
optional arguments:
-h, --help show this help message and exit
--reports_dirs REPORTS_DIRS [REPORTS_DIRS ...]
Paths where the reports are located (default: None)
--add_script_to_reports, --no-add_script_to_reports
Whether some javascript should be added to the report for interaction with the index file. (default: False)
--use_ordering_file, --no-use_ordering_file
Whether ordering.csv should be used to construct the ordering of index.html. The file should be located in the report-path folder. (default: False)
--navigation, --no-navigation
Whether the user should be able to freely navigate between reports. This is disabled for rating, to force user to process reports sequentially. (default: False)
--seed SEED Seed to control the randomization (to be used with randomize=True). (default: 42)
qc_segmentation
usage: qc_segmentation [-h] --bids_csv BIDS_CSV [--out_path OUT_PATH] [--ckpt_path_model CKPT_PATH_MODEL] [--nnunet_res_path NNUNET_RES_PATH] [--nnunet_env_path NNUNET_ENV_PATH] [--device {cuda,cpu}]
Compute segmentation on low resolution clinical acquisitions, using a pretrained deep learning model.
optional arguments:
-h, --help show this help message and exit
--bids_csv BIDS_CSV Path where the bids config csv file is located. (default: None)
--out_path OUT_PATH Path where the segmentations will be stored. (if not specified in bids_csv) (default: None)
--ckpt_path_model CKPT_PATH_MODEL
Path to the checkpoint to be used. (default: None)
--nnunet_res_path NNUNET_RES_PATH
Path to the nnunet folder containing the checkpoint. (default: /home/tsanchez/Documents/mial/repositories/qc_fetal_brain/fetal_brain_qc/models/nnUNet)
--nnunet_env_path NNUNET_ENV_PATH
Path to the nnunet folder containing the checkpoint (from `conda env list`). (default: /home/tsanchez/anaconda3//envs/nnunet)
--device {cuda,cpu} Device to use for the nnUNet inference. (default: cuda)
qc_compute_iqms
Note. If you want to run the metrics computation using segmentation-based metrics, you need to run qc_compute_segmentation prior to calling this script.
usage: qc_compute_iqms [-h] --out_csv OUT_CSV [--metrics METRICS [METRICS ...]] [--use_all_metrics | --no-use_all_metrics] [--normalization {None,sub_ses,site,run}] --bids_csv BIDS_CSV [--ckpt_path_slice_iqa CKPT_PATH_SLICE_IQA] [--ckpt_path_stack_iqa CKPT_PATH_STACK_IQA]
[--device DEVICE] [--continue_run | --no-continue_run] [--use_prob_seg | --no-use_prob_seg] [--verbose | --no-verbose]
Computes quality metrics from given images.
optional arguments:
-h, --help show this help message and exit
--out_csv OUT_CSV Path where the IQA results will be stored. (default: None)
--metrics METRICS [METRICS ...]
Metrics to be evaluated. (default: ['rank_error', 'dilate_erode_mask_full', 'mask_volume', 'filter_sobel_mask_full', 'nrmse_window', 'filter_laplace_mask', 'filter_laplace_mask_full', 'dilate_erode_mask', 'rank_error_center', 'centroid',
'rank_error_center_relative', 'seg_sstats', 'im_size', 'ncc_intersection', 'ncc_window', 'psnr_window', 'seg_snr', 'seg_volume'])
--use_all_metrics, --no-use_all_metrics
Whether all metrics should be evaluated (default: False)
--normalization {None,sub_ses,site,run}
Whether input data should be normalized (default: None)
--bids_csv BIDS_CSV Path where the bids config csv file is located. (default: None)
--ckpt_path_slice_iqa CKPT_PATH_SLICE_IQA
Path to the checkpoint of the fetal IQA pytorch model (by Junshen Xu at MIT). (default: /home/tsanchez/Documents/mial/repositories/qc_fetal_brain/fetal_brain_qc/models/fetal_IQA_pytorch.ckpt)
--ckpt_path_stack_iqa CKPT_PATH_STACK_IQA
Path to the checkpoint of the fetal IQA tensorflow model (by Ivan Legorreta FNNDSC). (default: /home/tsanchez/Documents/mial/repositories/qc_fetal_brain/fetal_brain_qc/models/FNNDSC_qcnet_ckpt.hdf5)
--device DEVICE Device to be used for the deep learning model. (default: cuda:0)
--continue_run, --no-continue_run
Whether QC run should re-use existing results if a metrics.csv file at `out_path`/metrics.csv. (default: True)
--use_prob_seg, --no-use_prob_seg
Whether to use the probability segmentation or the binary segmentation. (default: True)
--verbose, --no-verbose
Enable verbose. (default: False)
qc_inference
usage: qc_inference [-h] --iqms_csv IQMS_CSV --out_csv OUT_CSV [--ckpt_path CKPT_PATH] [--classification | --no-classification] [--regression | --no-regression] [--fetmrqc20 | --no-fetmrqc20] [--custom_model CUSTOM_MODEL]
Performs FetMRQC inference, given a pretrained model.
optional arguments:
-h, --help show this help message and exit
--iqms_csv IQMS_CSV Path where the IQMs csv file is located. (default: None)
--out_csv OUT_CSV CSV file where the predicted results will be stored. (default: None)
--ckpt_path CKPT_PATH
Path to the checkpoint of the fetal IQA pytorch model. (default: None)
--classification, --no-classification
Whether to perform classification. (default: False)
--regression, --no-regression
Whether to perform regression. (default: False)
--fetmrqc20, --no-fetmrqc20
Whether to use FetMRQC20 IQMs. (default: False)
--custom_model CUSTOM_MODEL
Path to a custom model, trained using run_train_fetmrqc.py. (default: None)
qc_training:
usage: Train a FetMRQC model. [-h] [--dataset DATASET] [--first_iqm FIRST_IQM] [--classification] [--regression] [--fetmrqc20 | --no-fetmrqc20] [--iqms_list IQMS_LIST [IQMS_LIST ...]] [--save_path SAVE_PATH]
optional arguments:
-h, --help show this help message and exit
--dataset DATASET Path to the csv file dataset.
--first_iqm FIRST_IQM
First IQM in the csv of the dataset.
--classification Whether to perform classification or regression.
--regression Whether to perform classification or regression.
--fetmrqc20, --no-fetmrqc20
Whether to use FetMRQC20 IQMs. (default: False)
--iqms_list IQMS_LIST [IQMS_LIST ...]
Custom list of IQMs to use. By default, all IQMs are used.
--save_path SAVE_PATH
Where to save the model.
qc_niftymic_qc: Run the QC from NiftyMIC.
usage: qc_niftymic_qc [-h] --out-csv OUT_CSV --bids-csv BIDS_CSV [--continue-run | --no-continue-run]
Exclude outlying stacks for each subject. Based on the paper from .
optional arguments:
-h, --help show this help message and exit
--out-csv OUT_CSV Path where the IQA results will be stored. (default: None)
--bids-csv BIDS_CSV Path where the bids config csv file is located. (default: None)
--continue-run, --no-continue-run
Whether QC run should re-use existing results if a metrics.csv file at `out_path`/metrics.csv. (default: True)
qc_ratings_to_csv: When manual QA/QC annotations have been done.
usage: qc_ratings_to_csv [-h] [--out_csv OUT_CSV] ratings_dir bids_csv
Given a `ratings_dir`, and a `bids_csv`, formats the ratings into a single csv file containing all information.
positional arguments:
ratings_dir Directory containing the ratings.
bids_csv CSV file where the list of available LR series and masks is stored.
optional arguments:
-h, --help show this help message and exit
--out_csv OUT_CSV CSV file where the ratings will be stored (default: `<ratings_dir>/ratings.csv`). (default: None)
--sr If set, the csv file will be formatted for the SR study. (default: False)
If you found this work useful, please cite the following articles.
For the large, multi-centric evaluation and model:
Sanchez, T., Esteban, O., Gomez, Y., Pron, A., Koob, M., Dunet, V., Girard, N., Jakab, A., Eixarch, E., Auzias, G. and Bach Cuadra, M. (2024). "FetMRQC: A robust quality control system for multi-centric fetal brain MRI". Medical Image Analysis, p.103282. https://doi.org/10.1016/j.media.2024.103282
@article{sanchez2024fetmrqc,
title={FetMRQC: A robust quality control system for multi-centric fetal brain MRI},
author={Sanchez, Thomas and Esteban, Oscar and Gomez, Yvan and Pron, Alexandre and Koob, M{\'e}riam and Dunet, Vincent and Girard, Nadine and Jakab, Andras and Eixarch, Elisenda and Auzias, Guillaume and Bach Cuadra, Meritxell},
journal={Medical Image Analysis},
pages={103282},
year={2024},
publisher={Elsevier}
}
For the original model (arXiv link):
Sanchez, T., Esteban, O., Gomez, Y., Eixarch, E., Bach Cuadra, M. (2023). "FetMRQC: Automated Quality Control for Fetal Brain MRI." PIPPI MICCAI Workshop 2023. https://doi.org/10.1007/978-3-031-45544-5_1
@inproceedings{sanchez2023fetmrqc,
title={FetMRQC: Automated Quality Control for Fetal Brain MRI},
author={Sanchez, Thomas and Esteban, Oscar and Gomez, Yvan and Eixarch, Elisenda and Bach Cuadra, Meritxell},
booktitle={International Workshop on Preterm, Perinatal and Paediatric Image Analysis},
pages={3--16},
year={2023},
organization={Springer}
}
As fetal brain imaging contains highly sensitive data, sharing the raw data is not possible. However, as FetMRQC relies on extracted image quality metrics, these can be easily shared for other researchers to re-train or enhance FetMRQC based models. They are available on Zenodo.
Copyright 2023 Medical Image Analysis Laboratory. FetMRQC is licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0. Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.
This project was supported by the ERA-net Neuron MULTIFACT – SNSF grant 31NE30_203977.
[1] Esteban, Oscar, et al. "MRIQC: Advancing the automatic prediction of image quality in MRI from unseen sites." PloS one 12.9 (2017): e0184661.
[2] Ranzini, Marta, et al. "MONAIfbs: MONAI-based fetal brain MRI deep learning segmentation." arXiv preprint arXiv:2103.13314 (2021).
[3] Semi-supervised learning for fetal brain MRI quality assessment with ROI consistency
[4] Gagoski, Borjan, et al. "Automated detection and reacquisition of motion‐degraded images in fetal HASTE imaging at 3 T." Magnetic Resonance in Medicine 87.4 (2022): 1914-1922.
[5] Lala, Sayeri, et al. "A deep learning approach for image quality assessment of fetal brain MRI." Proceedings of the 27th Annual Meeting of ISMRM, Montréal, Québec, Canada. 2019.
[6] Legorreta et al. https://github.com/FNNDSC/pl-fetal-brain-assessment
[7] Isensee, F., Jaeger, P. F., Kohl, S. A., Petersen, J., & Maier-Hein, K. H. (2021). nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation. Nature methods, 18(2), 203-211.