khanlab / diffparc-surf

Snakebids app for diffparc-surf, including dwi processing

diffparc-surf

Snakebids app for diffusion-based parcellation

diffparc-surf's Issues

Minimal eddy-current correction

With FSL Eddy disabled, there is only the rudimentary motion-correction applied. Will need to improve this if working with e.g. high b-value datasets, or just to improve distortions.

dwi2response for csd is the old FA-based algorithm

Should be using tournier instead. Can make this a config.yml setting.

Test dataset

Only a dry-run test dataset is currently in the repo -- can provide an openneuro dataset with T1w and dwi to use as a real test dataset, maybe shared via dropbox..

EPI-based distortion correction is missing if topup is not used..

If topup is disabled, the workflow doesn't have any way to correct EPI-based distortions.
Should have some non-FSL alternative for this.

README

Readme is currently empty, can adapt one from another snakebids app

Dicom import

Currently the app requires nifti or bids data.

For dicoms, one option could be to supply a formatted dicom zip, e.g.:
{subject}_{session}.zip
(the upstream app could create the dicom zip from uploaded files and metadata)

Given this, then this workflow would need to:

• parse subject, session from the filename
• extract files, convert to nifti
• figure out what files are T1w and what are DWI, and ignore anything else
• rest of the workflow identical

Generate tsv summary files

Produce a tsv file per subject/session, with summary metrics (surface displacement, surface area, connectivity) for each parcel and hemisphere.

If "tidy" format, columns can be:
subject, session, hemi, parcel, measure_type, aggregation_type

measure_type=[surfdisp,surfarea,conn,...]
aggregation_type=[mean,max,min,stdev ...]
parcel=[limbic,exec,caudal_motor,...]
hemi=[L,R]

Atlas segmentation is inaccurate using single non-linear template registration

Was a problem in the original diffparc too. One option is to use a CNN to segment the brain, e.g. as is now implemented in #3. Downside to that approach is we are stuck with the trained labelling.

CIFTI spec files

To ease visualization in connectome-workbench, can create a .spec file for each subject that includes the .surf.gii files, and the cifti dscalar and dlabel files.

Error in rule resample_dwi_to_t1w (job id 15)

shell:

antsApplyTransforms -d 3 --input-image-type 3 --input sub-3179/ses-Baseline/dwi/sub-3179_ses-Baseline_desc-preproc_dwi.nii.gz --reference-image sub-3179/ses-Baseline/dwi/sub-3179_ses-Baseline_space-T1w_res-orig_desc-dwiref_proc-crop_avgb0.nii.gz --transform sub-3179/ses-Baseline/dwi/sub-3179_ses-Baseline_from-dwi_to-T1w_type-itk_xfm.txt --interpolation Linear --output sub-3179/ses-Baseline/dwi/sub-3179_ses-Baseline_space-T1w_res-orig_desc-preproc_dwi.nii.gz --verbose

output:

...

  Applying transform(s) to timePoint/channel/dimension5 129 (out of 130).
Output warped image: sub-3179/ses-Baseline/dwi/sub-3179_ses-Baseline_space-T1w_res-orig_desc-preproc_dwi.nii.gz
Killed

Using pre-compiled binaries from https://github.com/ANTsX/ANTs/releases/tag/v2.4.2