You might find it simpler to use the 3dDeconvolve -stim_times_AM2 option to set up regressors that include the linear and quadratic dose curves (being sure to make these 2 sets of parameters orthogonal).
Since I gather this isn't actually time series data, you'd want to choose a response model function that extends over only 1 TR, and is equal to 1 at the "stimulus time". The simplest choice would be 'POLY(1,0,0.5)', then set the TR of your dataset to 1, and set each "stimulus time" to be the index in the input dataset of the corresponding point.
Then program will then generate t-statistics for each coefficient in the -stim_times_AM2 expansion, which would be the constant term, the linear in dose term, and the quadratic in dose term. This might meet your needs.