Marcus Lauer asked the following questions through e-mail:

Hello there. I'm a new AFNI user and I have a couple of questions.

What are the units used by the output of 3dDeconvolve when you use the -iresp flag? I'd like to do things such as plotting the resulting IRFs at certain voxels as examples of the differences in IRF between trial types, but I don't know what the units of my y-axis are. More generally, I don't know what the values represent. Are they arbitrary units which are only meaningful relative to one another? Are they percent signal change from some baseline? Are they percent signal change from the average of these voxels over time?

A bit more background: My input data is positive short integers (int16) in the 0-10,000 range. When I run 3dDeconvolve with and -iresp for each trial type, and have output files for each of our four trial types. Although the output files are technically short ints as well, when the scale factor comes into play, they appear to actually represent signed floats between about -50 and 100.

------

Hi Marcus,

Since 3dDeconvolve basically breaks your original signal at each voxel into a few components during deconvolution/regression analysis:

Signal = Baseline (trend component) + Sum of Condition/Task/Regressor components + Noise

like the original signal, all the coefficients and their statisitical values are dimensionless. And the beta values would be quite arbitrary depending on the magnitude of your regressors. Having said that, if you normalize your original signal, those beta values (regressor coefficients) would be percent signal change relative to the baseline signal, and in this case your Y-axis would be %.

The default datatype of the output in 3dDeconvole is set to be float because the impulse response functions are modelled with continuous curves. Even if your input data is in short positive integers, you still could get negative float numbers of impulse reponse or regressor weights because, after the baseline signal is filtered out, the impulse response functions could plunge into the negative side.

Gang