FDG-PET nipype preprocessing pipeline

[jrdalenberg]

Hey all,

I received a mail from @melanieganz about the derivatives meeting in which she linked to this repo.

Previously, my colleagues worked with PMOD to (pre)process data. But when I critically evaluated their pipeline, I noticed that it made a lot of errors during co-registration and normalization.

Since I am working on a study with currently 110 FDG PET scans from patients with hyperkinetic movement disorders and healthy controls (still recruiting up to 140), I decided to build a robust preprocessing pipeline with nipype. The data that I am working with are static 3T T1 and FDG-PET images.

The pipeline makes use of:

• fmirprep for the T1 images (--anat-only),
• AFNI autobox for cropping PET images
• HD-BET for skull-stripping PET images (which outperformed other alternatives that I tried). I built a custom nipype function for it. The repo (https://github.com/MIC-DKFZ/HD-BET) seemed to not be maintained anymore and I was thinking of forking it.
• ANTs for co-registration. I do this in two steps (inspired by fmiprep). First I co-register the skull-stripped mask, then the actual PET image.
• Coreg1, Coreg2, and fmriprep normalisation are then merged into a final transformation matrix
• Final matrix is applied to the original PET image
• Normalized PET image is then smoothed.

The pipeline seems to do a good job for our data set and I very much like to share it with the community.

What would be a good way to move forward?

[jrdalenberg]

I continued this work here: https://github.com/jrdalenberg/PETBrainPreprocessing