Dear Jay,
I'm not sure why you are seeing such large-scale changes here. The outline of your method seems fine.
The only things I can recommend are to be careful about the following:
- check that all your registrations are accurate
- look at the values in the Jacobians and make sure that the effect is not driven by one (or a few) outliers
- make sure that any smoothing you do is consistent and appropriate
- check if your randomise call is doing appropriate multiple-comparison correction and whether it is particularly sensitive to any thresholds you might have used
Apart from that there isn't anything I can think of.
The main one is to check that you registrations are all good and that it isn't an outlier-driven result.
All the best,
Mark
On 1 Feb 2013, at 13:58, SUBSCRIBE FSL Jay <[log in to unmask]> wrote:
> Hello,
>
> I have posted this question regarding TBM approach last week but did not get any coments. I have 30 controls who were scanned (structural T1) at time-point1 and time-point2( 6 weeks after time-point1). The following is the pipeline that i used for TBM analyses
>
> 1) Images from time-point1 and time-point2 were flirted to standard space.
> 2) Images from time-point1 in standard space were fnirted to Images from time-point2 in standard space to obtain the Jacobian of the field(providing the affine transform as input). Then i have log transformed the field.
> 3) I uses randomise on the log transformed jacobian to check if there is any significant expansion or contraction within the group. The contrast file has 2 contrasts. 1 (to check for expansion) and -1 (to check for contraction).
>
> What i see in the corrected p-vaules after thresholding at 0.99 is that there are huge significant changes showing that the controls significantly changed both ways i.e expansion and contraction (see attached screen shot). I wouldnt expect that the controls would change so much. Can anyone comment on these TBM processing steps.
>
> Regards
> <tbm.png>
|