Dear Par,

If I understand correctly your question, you are entering normalized functional images and would like to know whether it would be best to enter anatomical images that have been separately normalized (e.g. to a T1 template) or to enter anatomical images that have been first coregistered to the realigned functional volumes and then transformed using the same 'functional normalization' transformation (e.g. the *_sn.mat file obtained during the functional volumes normalization step). I do not have a definite answer to this question, the current implementation in the default conn preprocessing steps using the former approach (normalizes the anatomical volumes to a T1 template -in a single normalization/segmentation step- and then normalizes separately the functional volumes to an EPI template) because we feel that this approach might be more robust in general, but I do not know of any studies comparing these two approaches in detail. Another interesting possibility would be to perform the segmentation directly on the functional volumes, avoiding in this way the problem of inaccuracies in the structural/functional coregistration, at the cost of a more imprecise segmentation due to the lower resolution of the functional volumes. We are hoping to be able to explore the relative advantages of these different options for functional connectivity studies in more detail in the near future to be able to give a more precise recommendation, and I would love to hear your or other users thoughts about this as well!

And regarding your question about the CSF/WM mask threshold, currently we use a 0.5 value threshold on the posterior probability maps (and further perform a 1-voxel erosion step on the resulting masks to avoid partial volume effects) in order to obtain the final masks that are used by the aCompCor method. 

Best
Alfonso 

Originally posted by Pär Flodin: