Multivariate analysis techniques for fMRI data I am doing a project in which I need to predict fMRI activation values for each voxel of the brain. The voxels are approximately 20,000 and I have 300 examples with 25 features in each. Thus there are large number of dependent variables.
What are the best pratices involved in this multivariate analysis field?
I am new to this field and it would be very helpful if you may guide me in right direction.
 A: A very useful text is The Statistical Analysis of Functional MRI Data by Nicole Lazar (free pdf via Springerlink with institutional access). Chapter 7 covers multivariate approaches to the analysis of fMRI data. You don't mention in your post but the analysis of resting state vs. task generally require different approaches.
Resting state typically relies on principal components analysis (PCA) or independent components analysis (ICA) with both considered as a correlation analysis.
For analyzing voxel activation in the presence of task, I recommend chapter 6 of the book I've linked, which covers spatiotemporal models. I have more experience in this area and a simple approach is to fit the time series to a linear model (i.e. ANOVA) and convolve the design matrix with the so-called "canonical hemodynamic response function (HRF)."
Additionally, I've found course materials from the University of New Mexico helpful when I was starting out in this field: Analysis Methods in Functional Magnetic Resonance Imaging.
