Skip to content

quality-check - Evaluate registration quality

Three different quality check procedures are available in ClinicaDL: one for the t1-linear preprocessing pipeline, one for the t1-volume pipeline and one for the pet-linear pipeline.

quality-check t1-linear - Evaluate t1-linear registration

The quality check procedure relies on a pretrained network that learned to classify images that are adequately registered to a template from others for which the registration failed. It reproduces the quality check procedure performed in [Wen et al., 2020]. It is an adaptation of [Fonov et al., 2022], using their pretrained models. Their original code can be found on GitHub.

Warning

This quality check procedure is specific to the t1-linear pipeline and should not be applied to other preprocessing procedures as the results may not be reliable. Moreover, you should be aware that this procedure may not be well adapted to de-identified data (for example images from OASIS-1) where parts of the images were removed (e.g. the face) or modified to guarantee anonymization.

Prerequisites

You need to execute the clinica run t1-linear prior to running this task.

Running the task

The task can be run with the following command line: 

clinicadl quality-check t1-linear [OPTIONS] CAPS_DIRECTORY OUTPUT_TSV
where:

  • CAPS_DIRECTORY (Path) is the folder containing the results of the t1-linear pipeline and the output of the present command, both in a CAPS hierarchy.
  • OUTPUT_TSV (str) is the path to the output TSV file (filename included).

Options:

  • --participants_tsv (Path) is the path to a TSV file containing the subjects/sessions list to check (filename included). Default will process all sessions available in caps_directory.
  • --threshold (float) is the threshold applied to the output probability when deciding if the image passed or failed. Default value: 0.5.
  • --batch_size (int) is the size of the batch used in the DataLoader. Default value: 1.
  • --n_proc (int) is the number of workers used by the DataLoader. Default value: 2.
  • --gpu/--no_gpu (bool) Use GPU for computing optimization. Default behaviour is to try to use a GPU and to raise an error if it is not found.
  • --use_tensor (bool) is a flag allowing the pipeline to run on the extracted tensors and not on the nifti images.
  • --network (str) is the architecture chosen for the network (to chose between darq, sq101 and deep_qc)

Outputs

The output of the quality check is a TSV file in which all the sessions (identified with their participant_id and session_id) are associated with a pass_probability value and a True/False pass value depending on the chosen threshold. An example of TSV file is:

participant_id session_id pass_probability
sub-CLNC01 ses-M000 0.9936990737915039
sub-CLNC02 ses-M000 0.9772214889526367
sub-CLNC03 ses-M000 0.7292165160179138
sub-CLNC04 ses-M000 0.1549495905637741
... ... ...

quality-check t1-volume - Evaluate t1-volume registration and gray matter segmentation

The quality check procedure is based on thresholds on different statistics that were empirically linked to images of bad quality. Three steps are performed to remove images with the following characteristics:

  1. a maximum value below 0.95,
  2. a percentage of non-zero values below 15% or higher than 50%,
  3. a similarity with the DARTEL template around the frontal lobe below 0.40. The similarity corresponds to the normalized mutual information. This allows checking that the eyes are not included in the brain volume.

Warning

This quality check procedure is specific to the t1-volume pipeline and should not be applied to other preprocessing procedures as the results may not be reliable.

Prerequisites

You need to execute the clinica run t1-volume pipeline prior to running this task.

Running the task

The task can be run with the following command line: 

clinicadl quality-check t1-volume [OPTIONS] CAPS_DIRECTORY OUTPUT_DIRECTORY GROUP_LABEL
where:

  • CAPS_DIRECTORY (Path) is the folder containing the results of the t1-volume pipeline and the output of the present command, both in a CAPS hierarchy.
  • OUTPUT_DIRECTORY (Path) is the path to an output directory in which TSV files will be created.
  • GROUP_LABEL (str) is the identifier for the group of subjects used to create the DARTEL template. You can check which groups are available in the groups/ folder of your caps_directory.

Outputs

This pipeline outputs 4 files:

  • QC_metrics.tsv containing the three QC metrics for all the images,
  • pass_step-1.tsv including only the images which passed the first step,
  • pass_step-2.tsv including only the images which passed the two first steps,
  • pass_step-3.tsv including only the images which passed all the three steps.

Manual quality check

This quality check is really conservative and may keep some images that are not of good quality. You may want to check the last images kept at each step to assess if their quality is good enough for your application.

quality-check pet-linear - Evaluate pet-linear registration

This quality check procedure uses a metric that identifies images that have been misregistered by comparing the output of the clinica run pet-linear pipeline with a contour mask used as the reference for registration. The contour mask is created by combining the CBM 2009c Nonlinear Symmetric brain mask and head mask (that you can find (here)[http://www.bic.mni.mcgill.ca/~vfonov/icbm/2009/mni_icbm152_nlin_sym_09c_nifti.zip]), which aligns with the MNI reference.

To determine if an image is well-registered, we compare the contour mask with the PET image. We first normalize the image and set a threshold of 0.35. We then calculate the sum of pixels within the contour mask, and if this value is too high, we can assume the image is not well-registered.

Note

It is important to note that the t1-linear pipeline must be run prior to the pet-linear pipeline, as an image that does not pass the t1-linear quality check will not pass the pet-linear check either. For best results, we recommend running the quality-check pet-linear on the list of subjects that have passed the t1-linear quality check.

You can ask us to have the process for creating the mask and the different experiments that were conducted to choose the best metric.

Warning

Please note that this quality check is conservative and may keep some images that are not of good quality. We advise you to check the quality of the last images kept at each step to assess if their quality is good enough for your application. Finally, this quality check procedure is specific to the pet-linear pipeline and should not be applied to other preprocessing procedures, as the results may not be reliable.

Prerequisites

You need to execute the clinica run pet-linear pipeline prior to running this task.

Running the task

The task can be run with the following command line:

clinicadl quality-check pet-linear [OPTIONS] CAPS_DIRECTORY OUTPUT_TSV TRACER
                       {pons|cerebellumPons|pons2|cerebellumPons2}
where:

  • CAPS_DIRECTORY (Path) is the folder containing the results of the pet-linear pipeline
  • OUTPUT_TSV (Path) is the output TSV file in which you find for each subject, the pass_probability of being well-registrated.
  • TRACER is the label given to the PET acquisition, specifying the tracer used (trc-). It can be for instance 18FFDG for 18F-fluorodeoxyglucose or 18FAV45 for 18F-florbetapir.
  • The reference region is used to perform intensity normalization (i.e. dividing each voxel of the image by the average uptake in this region) resulting in a standardized uptake value ratio (SUVR) map. It can be cerebellumPons orcerebellumPons2 (used for amyloid tracers) and pons or pons2 (used for FDG). See PET introduction for more details about masks versions.

Outputs

The output of the quality check is a TSV file in which all the sessions (identified with their participant_id and session_id) are associated with a pass_probability value and a True/False pass value depending on the chosen threshold. An example of TSV file is:

participant_id session_id pass_probability pass
sub-CLNC01 ses-M000 0.9936990737915039 True
sub-CLNC02 ses-M000 0.9772214889526367 True
sub-CLNC03 ses-M000 0.7292165160179138 True
sub-CLNC04 ses-M000 0.1549495905637741 False
... ... ... ...