Skip to content

pipeline.md

Latest commit

 

History

History
103 lines (67 loc) · 5.12 KB

pipeline.md

File metadata and controls

103 lines (67 loc) · 5.12 KB
Raw

Enigma-PD-WML pipeline details

All processing steps are called via a bash script analysis_script.sh

This processes each subject's T1-weighted and FLAIR MRI images with the following main steps (each is a separate function in the bash script):

%%{init: {"flowchart": {"htmlLabels": false}} }%%
flowchart TD
    fsl_anat("`Run fsl_anat on T1 image`")
    flair_prep("`Pre-process FLAIR image and align it to T1 image`")
    vent_dist("`Create periventricular vs deep white matter masks for T1 and FLAIR images`")
    unet_prep("`Crop T1/FLAIR images, ready for UNet-pgs`")
    unet("`Run UNet-pgs to segment WMLs`")
    process_output("`Split WML segmentations into periventricular vs deep white matter + align to standard MNI template`")
    fsl_anat --> flair_prep
    flair_prep --> vent_dist
    vent_dist --> unet_prep
    unet_prep --> unet
    unet --> process_output
Loading

After processing, the pipeline also generates an interactive QC output. See the QC docs for more information on those steps.

fsl_anat

First, FSL's fsl_anat tool is run on the input T1 image and its output saved to derivatives/enigma-pd-wml/<subject>/<session>/input/t1-mni.anat folder of each session.

See FSL's documentation for details of the various processing steps this includes.

FLAIR preparation

Next, various FSL tools are used to pre-process the input FLAIR image and align it to the T1 image. Results are written to the derivatives/enigma-pd-wml/<subject>/<session>/input/flair-bet directory of each session. This includes steps to:

Ventricular distance mapping

Next, various FSL tools are used to create periventricular vs deep white matter masks for both the T1 and FLAIR images. These will be used at a later stage to divide WML segmentations into periventicular vs deep white matter. Results are written to the derivatives/enigma-pd-wml/<subject>/<session>/input/vent_dist_mapping directory of each session.

This includes steps to:

  • create binary masks of the ventricles and white matter with make_bianca_mask

  • Align these masks with the FLAIR brain with flirt

  • Use FSL's distancemap tool and fslmaths to produce the final periventricular vs deep white matter masks

Prepare images for UNet-pgs

Next, various FSL tools are used to prepare the images for UNet-pgs. The main purpose of this step is to crop T1 / FLAIR images with fslroi (if they're larger than 500 pixels in the x or y dimension).

UNet-pgs

Run the UNet-pgs segmentation workflow. This uses the WMHs_segmentation_PGS.sh script from the pgs docker image. This will produce a WML binary segmentation mask under the derivatives/enigma-pd-wml/<subject>/<session>/output/results.nii.gz of each session.

Process outputs of UNet-pgs

Next, the output from UNet-pgs is processed with various fsl tools. This splits the WML segmentation from UNet-pgs into periventricular vs deep white matter, as well as linearly/non-linearly aligning it to the standard MNI T1 1 mm template.

This includes steps to:

  • Align WML segmentations from UNet-pgs with the cropped T1 image, the full size T1 image and the full size FLAIR image via flirt

  • Use masks from the ventricular distance mapping step and fslmaths to divide WML segmentations into periventricular and deep white matter

  • Combine the periventricular and deep white matter masks using fslmaths to restrict the segmentations to the white matter, excluding grey matter, CSF and areas outside of the brain

  • Produce WML segmentations using the John Hopkins ICBM-DTI-81 white-matter labels and Oxford-GSK-Imanova striatal probabilistic connectivity atlases

  • Linearly transform/align all segmentations with MNI T1 via flirt

  • Non-linearly transform/align all segmentations with MNI T1 via applywarp

Final pipeline outputs

The pipeline will generate multiple .zip files - one per session, stored within the corresponding session sub-folder, e.g. derivatives/enigma-pd-wml/sub-1/ses-1/sub-1_ses-1_results.zip.

A csv file data/derivatives/enigma-pd-wml/t1_volumes.csv which contains the estimated brain volume (in mm³) in native T1-weighted space and MNI space, as well as the factor used to scale the brain from T1-weighted to MNI space for each subject and session will also be generated.

Please send these zip files to the ENIGMA-PD Vasc team.