Dear Christina,
Two previous discussions suggest it is not possible to run FSL group analysis ('FLAME') on results from a different software package:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind1508&L=FSL&D=0&1=FSL&9=A&I=-3&J=on&d=No+Match%3BMatch%3BMatches&z=4&P=251772
https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind1611&L=FSL&D=0&1=FSL&9=A&I=-3&J=on&d=No+Match%3BMatch%3BMatches&z=4&P=81702
FSL normalizes the signal of the whole 4D dataset as a whole, while AFNI normalizes the signal per voxel. This affects the scaling of the coefficients. Interestingly, the following study compared mixed effects approaches from AFNI (3dMEMA) and FSL (FLAME):
https://www.sciencedirect.com/science/article/pii/S1053811911014625
I have only had a look at it, so I am not sure how the different normalizations were dealt with. I think the study compared 3dMEMA and FLAME using the same coefficient maps and the same standard error maps (alternatively: the t-statistic maps, t-stat = coef/se).
Btw if you want to run group analyses in FSL because FLAME is a mixed effects approach, keep in mind that AFNI's 3dMEMA is doing almost the same as FLAME. In AFNI there are different ways of doing group analyses: 3dMVM is the most popular one, but there are also 3dLME and 3dMEMA.
If you find a way of dealing with different normalizations, you could generate dummy FSL's first level outputs, with necessary folders and maps called the same way as in FSL.
Best,
Wiktor Olszowy
|