Hello FSL writers and users,
I'm working with a cross-sectional longitudinal fMRI data set, which is unbalanced. (oh jeez!) I was hoping to use FSL as much as possible, using additional packages only sparingly at higher levels, if I need to, in order to estimate changes in bold signal within individuals and across time, taking into consideration lower-level (run and subject) variances.
DATA
We have data from about 150 participants ranging in age from about 10 years to about 20 years at the time of first testing (cross-sectional, time 1), about 90 participants came back one year later (time 2), and another 30 or so about a year after that (time 3). Each visit entails 4 fMRI runs. Only participants with 4 complete runs are included.
MY LEVEL OF UNDERSTANDING
I can accurately generate run and visit level PEs, and approximate higher level PEs (in a convoluted way), using lower-level variance estimates. I understand that the only information going into these higher level analyses are the lower level PEs and Var estimates.
MY TRIED APPROACHES
My attempts in using R to replicate FSL results have been fair for PEs for single-visit data, but I don't think that my method is a viable one given the complexity of the data set. Further, my back-calculation of the Var from standard error is off. I cannot find the Behrens et al., 2003 reference online cited by Beckmann et al., 2003 Neuroimage paper. I don't feel comfortable with moving forward strictly within R if I can't do this.
I've tried using BUGS to estimate PEs, but this software doesn't seem to generally allow different variance estimates for each PE ( I could be wrong ), or if does (Gelman, 2006, Bayesian Analysis), it appears to require a 'sensible' scaling parameter, which I'd hate to do on a voxel-by-voxel basis.
MY GOAL
I'd like to be able to use FSL for all lower-level data analyses, passing up the lower-level variances into higher-level estimates for change with respect to age. If this is possible, or not, entirely within FSL, I'd like to know. If not, if someone has suggestions on how I might be able to use what I can from FSL, I'd appreciate that information as well.
Any suggestions for directions or on how to turn back are welcome.
Thank you!
David
|