Hey, thanks again! I thought the same and found several rather stupid mistakes in my fsl designs. Smoothing was different for one.. and there where several other things..Sorry for the bother, must really be something related to my carelessness. Am currently checking for further mistakes in the fsl batch.
Best, Nils
________________________________________
Von: FSL - FMRIB's Software Library [[log in to unmask]] im Auftrag von Jesper Andersson [[log in to unmask]]
Gesendet: Dienstag, 11. Oktober 2011 13:38
An: [log in to unmask]
Betreff: Re: [FSL] AW: [FSL] Multiple runs in one firstlevel feat
Dear Nils,
you are right that the results should be similar. I find it unlikely that the differences in noise modelling should cause so big differences that the t-maps are unrecognisable between the packages (unless you have a very severe outlier problem). My first guess would be that you have made a mistake in setting up the model with one of the packages. I'd recommend to start your "debugging" by using matlab to have a look at the design matrices you obtained with the two packages and look primarily for large differences in onset and or duration of your blocks.
Also you should keep the smoothing constant across the two packages if you want to compare them.
Good luck Jesper
On 11 Oct 2011, at 13:25, Kohn, Nils wrote:
Thanks for the answer.
In SPM you don’t have to concatenate the scans, you can enter more than one session, which is what I did. As far as I understand SPM calculates the sum and mean of session means (with only one error term), if you have a contrast which includes a blocked event per session.
As far as I understand, fsl does some bayesian statistics for the higher level feat and includes one error term per session. So here’s a basic difference, right?
I calculated a contrast between two sessions in fsl and spm, and as I said the results per subject look very different and also on the group level.
But shouldn’t the results from the two approaches be basically the same? SPM should be more influenced by outliers but overall, should not be that different, or am I totally wrong??
I don’t understand why there is such a big difference and would be thankful for any help!!
Best, Nils
________________________________
Von: FSL - FMRIB's Software Library [mailto:[log in to unmask]] Im Auftrag von David Soto
Gesendet: Montag, 10. Oktober 2011 22:09
An: [log in to unmask]<mailto:[log in to unmask]>
Betreff: Re: [FSL] Multiple runs in one firstlevel feat
Hi,
so guess you must have concatenated the onset times in SPM...
in FSL you should do (1) a lower level analyses for each of the runs separately, all EVs and contrasts specified separately for each block run- which guess would specify regions that 'activate' overall on the run, thus entering +1
then you do (2) a higher-level analyses across runs within subjects (FIXED EFFECTS) to get an average response for each subject for each of your contrasts - since your EVS are blocked you may specify your contrasts (eg, A>B) at this level
and finally you do (3) a higher-level analyses across subjects (MIXED EFFECTS FLAME 1 or 1+2 for increased sensitivity) on each of your lower level contrasts of parameter estimates - copes
remember to check the BET output of your structural scans and also the registration steps were fine!
regards, David
On Mon, Oct 10, 2011 at 8:03 PM, Kohn, Nils <[log in to unmask]<mailto:[log in to unmask]>> wrote:
Hi everyone,
I'm currently trying to analyse a dataset from several subjects with fsl (first time!).
I acquired the data in three runs, each has a block design (with three tasks), but each run had to be somewhat self-paced due to the nature of the task, so the onsets are not equal (durations are equal) over all three runs and additionally each run has a different number of volumes.
I previously analysed these subjects in SPM and made one firstlevel design for each subject. Since we would like to addtionally do an ICA analysis we would prefer to do all analyses in FSL (GLM and ICA).
I set up an FSL design, three first levels per subject, next step was entering the feat output from the runs into a second analysis. I used the same onsets and durations as well as design in FSL and SPM.
Here comes my problem: The results from the FSL and SPM analysis differ strongly, the analyses show completely different activated areas. Sadly SPM makes way more sense. Any ideas why this could be the case?
Is there a possibility to analyse all three runs in one design?
Thanks for any help!!
Best, Nils
|