Thanks for the quick reply Gang!
I should have specified that it is the shape of the hemodynamic response I'm trying to get, so 3dttest++ doesn't seem appropriate.
I don't mind phase being correlated with my regressors. The idea is to implement Menon's phase regression method (Ravi Menon, Mag Res Med 2002) in a slightly different way. Since phase correlates with magnitude (and therefore with the regressors of interest) in and around large veins, but not in the brain tissue, regressing phase time series from the magnitude time series should remove most of vein-borned signal. And since it is the shape of the (large vein signal-free) magnitude response I want to get, I don't mind issues of degrees of freedom.
I could always regress out phase with 3dTproject before extracting the hemodynamic response from magnitude with 3dDeconvolve, but I'm afraid that would over-correct for vein signal even more than if I do it in a single step.
I hope this makes my goal clearer.
Thanks again for the help
Sébastien Proulx