Amplitude Modulation Analysis

Hi there,

We are conducting an amplitude modulated analysis with a behavioral measure that has a 4-point rating scale. Is there a way to construct a graph in AFNI that illustrates the relationship between BOLD activation and our behavioral measure? If not, is there an efficient way to extract that information?



Hi Mary,

To be sure, are you running 3dDeconvolve/3dREMLfit using -stim_times_AM2, where each event has an attached modulator in {0,1,2,3} (or {-1.5, -.5, .5, 1.5})?

If so, that assumes a linear relationship between the modulator and the amplitude of the BOLD response, where the output (at each voxel) is a beta for the mean response, and a beta for the incremental effect of the modulator.

Since the output is a mean and slope for the modulated conditions, there isn’t really so much to plot.

What you might want to do is model all 4 levels separately, and then run a simple linear fit to that. That could be compared with the outputs from 3dDeconvolve (they should be similar), and it could be viewed as a 4-point time series, or even as scatter plots over ROIs/clusters, say.

Does this seem reasonable?

  • rick

Hi Rick,

Thank you so much for the response! Yes, I think that makes sense. When you suggest running all four levels separately, does that mean creating an unique regressor file for each level of our behavioural variable so this would appear as separate subbriks or is this something that can be done within the larger analysis?

Thanks again!


Hi Mary,

Yes, that is right. With 4 separate conditions/regressors, you can then evaluate how linear they are, separate from the assumption of it in the AM regression. That would yield 4 betas in separate volumes. With only 4 though, it will be difficult to say anything strongly. But a scatter plot across entire ROIs might make it more clear.

  • rick

Thanks Rick!