If you just want to check whether there are some interactions between a continuous covariate (e.g., age) and categorical variables (A and B in your case), you can easily do that with 3dLME.R and get F values for those interactions. The difficult part is when you want to get some specific contrasts, which I have not got chance testing out yet with real data.

> For now, perhaps just assuming a linear relationship between age and
> activation, though being able to plug in some other hypothesized age function
> would be desirable.

Nonlinear relationship should be relatively easy to incorporate into the model if you know the type of non-linearity you're looking for.

Gang