To followup on Gang. In your case, the answer is C.
Since each participant is contributing a single value for each voxel, 3dttest++ has no way of knowing that it was a "dead voxel" for a particular participant. I imagine it would be hard to identify a "dead" voxel by the stats files alone going into a group analysis program as 1) zero is a valid number and 2) I suspect that the value of those dead voxels after preprocessing ends up being something other than zero given the certain level of interpolation related to preprocessing (slice timing, motion correction, alignment, smoothing).
I suspect that you could use the entire time series and some interesting masking procedures to try and identify "dead voxels" and interpolate a value into them from neighboring voxels.
Edited 2 time(s). Last edit at 09/23/2014 03:51PM by Peter Molfese.