Hi,
From step 2, I think the best thing to do is–
1) run a single Fixed Effects model for each individual. This will create the (i believe) average copes for each person (one for each contrast - i.e. you will end up with a folder with 10 copes per person). Apart from being, apparently, the better way to run the analysis (there's a lot of debate about this somewhere on the FSL site), this also allows you to check the results, registration for each person individually.
2) run a second level mixed effects analysis with the FE results. You can select each person's feat directory to run a second level analysis for all contrasts at once.
Otherwise you can run a separate second level analysis for each cope separately. I take this latter approach as for some reason the first one has not worked for me. The way I do this is–
a) within your overall second level analysis folder, create a folder for each contrast (e.g., cope1-10). Name the cope folders exactly the same as the actual cope contrasts (cope1, not cope01 for e.g.).
b) Starting with cope2, setup the first model in the FEAT gui. Save it into the cope2 folder, then copy the .fsf file only into the folders for each other cope contrast.
c) use the sed command to change the cope numbers for each subsequent analysis for each of the other copes. E.g., once you have copied the cope2 .fsf file into the folder for cope1, on the command line type
sed -i 's/cope2/cope1/g' *.fsf
Repeat for other copes. You wind up with one cope directory per lower level contrast, each with one cope corresponding to each of your second level analyses inside. You can of course write a shell script that speeds this up a lot, but they're the basics (at least for me)
Cheers
Chris
On May 12, 2011, at 8:21 PM, Erie Boorman wrote:
Hello,
I am attempting to run an analysis for 10 subjects with four runs per subject. Each single subject has 30 COPEs. I have tried to follow the instructions listed on the FSL webpage: http://www.fmrib.ox.ac.uk/fsl/feat5/detail.html#MultiSessionMultiSubject listed under the section entitled Multi-Session & Multi-Subject (Repeated Measures - Three Level Analysis).
Following the instructions to the best of my ability, I did the following:
1) Ran a lower-level FEAT for each run for each subject.
2) Constructed a higher-level FEAT where the inputs were lower-level FEATs for each run for each subject (i.e. 40 FEATs corresponding to 40 runs). In the stats tab, I selected the Fixed Effects option and set up the design with 40 EVs, where each EV picks out the 4 sessions that correspond to a particular subject. I also included 10 contrasts to represent the 10 subject means.
Rather than output something like subject_N.gfeat/cope1.feat, this second level analysis output 30 cope.feat directories, each with 10 cope.nii.gz files in its stats directory. The file structure was as follows: Group_Ana/cope[n].feat/stats/cope[j], where n = 1:30 lower level contrasts and j = 1:10 subjects. This is different from what the instructions implied would be output.
From an earlier post by Lara on 16 Mar 2011, it was advised to:
1. Analyze each session separately for each subject (including registration).
2. Use a higher-level analysis to combine a subject’s sessions using the fixed-effects stats option, selecting GROUP MEAN model (single EV, with all 1’s in that EV and a single contrast with a single “1” in it), making sure that inputs are lower level FEAT’s.
3. Once I have #1 and #2 completed for each subject, I then run a separate higher-level analysis to look at effects at the group level. For this, I select the option “Inputs are 3D cope images” in the “Data” tab. For the input, I select the [subject].gfeat/cope[n].feat/stats/cope1 image. The rest of the higher-level settings I leave as normal...
If I were to adopt this second approach, I think it would mean selecting 300 inputs (30 3D cope images x 10 subjects), which would be very time-consuming. Is there a simpler way to do this?
Any help regarding which approach to adopt and how to implement it would be greatly appreciated.
Kind regards,
-Erie Boorman
|