(1) You shouldn't use "-rlt" or its relatives if you are going to use 3dDeconvolve. The reason for this is explained at length in the 3dDeconvolve manual.
(2) If you really don't want to run 3dDeconvolve again, then you'll have to get the baseline for the denominator from somewhere else. You could use 3dTstat, for example, to compute the mean of your original (non-detrended) dataset.
bob cox