Dear Anderson,
Thank you for the fantastic input. I'm essentially trying to follow this recent 2013 article : http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0059382
1) Where I'm struck is whether the smoothness estimate should be obtained from the FA Maps [OR] the dti_sse.nii.gz output from dtifit. Especially because the fitting to obtain the eigen values (and then FA) is not from a GLM but a OLS.
1.1) On a related note, does it make a big difference, if I determine the smoothness estimate at the MNI resolution as opposed to the acquisition resolution. In the past I always did this in acquisition resolution.
2) As you, Hayasaka and guru Tom Nichols rightly point out, the non stationarity assumption is somewhat ignored in many cases.
3) If I used FSL's smoothest function then I'm some what struck :
case1 : >>smoothest -d <number> -r <filename> -m <filename> ( (1) I do not know what is the degrees of freedom in this case and (b) whether I should use dvi_sse.nii.gz)
case2 : >>smoothest -z <filename> -m <filename> (this is doable)
But Note : even smoothest does not seem to account for Non Stationarity and hence downstream 'cluster' command is also likely afflicted. Please, Please correct me if I'm wrong and if there are alternatives within FSL.
4) I looked into SPM function :
>> spm_est_smoothness(V,VM,ndf)
% V - Filenames or mapped standardized residual images
% VM - Filename of mapped mask image
% ndf - A 2-vector, [n df], the original n & dof of the linear model
I find I'm somewhat back to my question 1, as I do not know (a) what 'dof' to input and (b) plugin dti_sse.nii.gz OR the Z-stat volume.
5) Finally (without exploring in great detail) the NS toolbox, also seems to be incorporated focussed on setting up a GLM (for fMRI analyses).
Anderson, I thank you once again for your time and answers and will be much obliged to hear your response.
Thank you,
Kumar.
|