Logically, you can't do a t-test (etc.) when you just have the means of the 2 different cases. The reason is that there is no way, from this data alone, to judge the variance of the data that went into the means.

For example: is 3 significantly different from 4? Hard to answer unless you define "significance", which in statistics is done by determining the variability of the data. So if the standard deviation of the data whose mean is 3 is 0.1, then it is a long ways from 4; but if the standard deviation of the data whose mean is 3 is 27, then 3 is pretty close to 4.

This is why 3dttest (etc.) requires multiple input bricks -- not just to compute the means of the cases, but their variances.

If all you have left is the mean bricks, then you are kind of in trouble. If you have the original data from which the means was derived, then you can either use that data as input to 3dttest, or use that data to compute the standard error of the means and then compute the t-statistic explicitly using 3dcalc.

bob cox