> I'm trying to figure out what tools are available for longitudinal
> analysis of fMRI data, ideally using a linear mixed models approach
> that is used for other longitudinal data.
I am not aware of sowftare implementing this in the voxelwise setting.
If you could specify a ROI and extract data from there, you could of
course use R to specify and test your model, using different random
structures. I suspect you'd be better off with modelling the residuals
with an autocorrelation structure. First, it is appropriate for the
longitudinal setting, and second there is more of it in neuroimaging
> Additionally, how should missing data be dealt with? Is there a way
> to impute missing data using a missing at random framework, or
This strikes me as a hard combination, and I'd be surprised (and
interested) to hear there is anything for neuroimaging in this corner.
You have to generate random data with the correlation struture implied
by the model, which adds to the difficulty of imputing under the
spatial correlation structure. There is a fairly recent book (Daniels
& Hogan) on longitudinal models + missingness, where they propose to
use freely available sofwtare (that you can call from R) for this
Dept. of Psychiatry, University of Ulm