I want to update my question...
My goal is to compare 3D anatomical volumes using machine learning (details not important). Of course it is problematic since the data comes from different studies and have different resolutions, orientations and different spatial location within the FOV.
Using a non-linear warping to e.g. MNI space will not be a good option since it might remove subtle differences between said groups (first age to test the network and then moving on to diagnosis).
That is why we ended up with using fslvbm!
We don't want any stats output from fslvbm but we want to use the generated volumes.
The question is: What volumes should we use?
The user gudie says:
"each voxel of each registered grey matter image is multiplied by the Jacobian of the warp field (see Good et al., 2001). All the modulated registered GM images are concatenated into a 4D image in the stats directory ("GM_mod_merg") and then smoothed ("GM_mod_merg_s3" for instance) by a range of Gaussian kernels; sigma = 2, 3, 4mm, i.e., approximately from FWHM = 2x2.3 = 4.6mm to FWHM = 9mm. "
Should we use the jacobians or the product of the jacobians and the registered grey matter images? I.e. un-merge the 4D image in the stats directory (GM_mod_merg)?
We basically want a 3D volume per subject that when compared would correspond to anatomical differences that does not suffer from the drawbacks of the standard non-linear warping typically used by the standard neuro imaging packages.
Also, should we re-sample the input to fslvbm_2_tempalte to the datasets lowest resoltion since they come from different sites?
Thanks!!
########################################################################
To unsubscribe from the FSL list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1
