Hi Federico
>Dear Experts,
>
>After reading the controversy about using froi and localization task
>(Friston et al. vs. Saxe et al. Neuroimage, 2006) I would understand if the
>cluster/roi homogeneity can be evaluated as follow. I've a 2x2 within
>subjects design A (a1, a2) x B (b1, b2) so the design matrix has four
>regressor a1b1, a1b2, a2b1, a2b2; I model with canonical hrf.
>
>
I'll try to answer that as I did try something similar
>At first level I run four t-contrasts as ([1 0 0 0]; [0 1 0 0]; [0 0 1 0];
>[0 0 0 1]) and in corrispondence to each I get the first eigenvariate, I
>type xY.v after I export these weights in SPSS:
>ROWS are Subjects: sub1, ..., Subn
>COLUMNS are: weight_1_a1b1 ... weight_n_a1b1; ... ; weight_1_a2b2 ...
>weight_n_a2b2.
>
>In this way I can evaluate the weight variability for each subject in each
>condition (take a row I've mean and standard deviation for a1b1, a1b2, a2b1,
>a2b2) and the weight variability across subjects (mean and standard
>deviation in coloumn). If the weight variability is low (the weights of the
>voxels forming this cluster are the same) I can consider the first
>eigenvariate as a cluster mean or in other words this functional roi is
>homogenous. Is this correct?
>
>
I think it is. As an additional test I also computed the mean overall
voxels per subjects and compared with the 1st eigenvariate
>Another question: when I extract the first eigenvariate shall I adjust for
>the effects of interest or it better I choose don't adjust?
>
>
I didn't adjust to get something close to my mean values ..
I think it's an interesting question as one of the problem with the ROI
is the inhomogeneity of voxel signals in another task than the
experimental one. But if one tests this looking at the weights of the
voxel / comparing with 1st eigen vs the average then it can justify the
ROI stats :-\ (at least for me as I tend to agree with the paper..)
Any suggestions folk?
Best
Cyril
|