Hi,
I'm using FSL to apply a parcellation scheme to extract time series from resting state scans.
As a first step, for each subject I'm registering the parcels ("Parcels_combo.nii") to their BET'd structural scan using the command below.
flirt -in Parcels_combo.nii -ref struc.nii -out results/parc2anat -omat results/parc2anat.mat
My problem is that this step (and the subsequent steps) have worked in the majority cases, however in about a fifth of cases the resulting parc2anat.nii.gz image is stretched and distorted when I inspect in fslview. In these cases the subsequent time series extraction is unsuccessful.
Across both the successful and unsuccessful structural scans, I have compared some obvious things like voxel dimensions and orientation labels, they are all equivalent so I can see no clear reason why the FLIRT fails in a subset. Indeed, they are all scans collected from the same scanner, under the same protocol, and in the same patient population.
I'm wondering if there might be some other options I could apply during the FLIRT transformation that might help improve the registration in this subset, given that there seems to be no obvious underlying problem with the structural scans or the parcels folder.
Many thanks,
Claire
PS - if relevant, the subsequent steps I am applying in FSL are:
flirt -in struc.nii -ref rest_mcf.nii.gz -out results/anat2rest -omat results/anat2rest.mat
convert_xfm results/parc2anat.mat -concat results/anat2rest.mat -omat results/parc2rest.mat
flirt -in Parcels_combo.nii -ref rest_mcf.nii.gz -applyxfm -init results/parc2rest.mat -out results/Parcels_combo_reor -interp nearestneighbour
