My ultimate goal is to create an appropriate interaction term to implement what is often referred to as an analysis of psychophysiological interactions (PPI), e.g., multiple regression predicting voxel-wise brain activity from an interaction between activity in a specific brain region (seed) and a task parameter. Gitelman et al. (see full ref. below) suggest that this sort of analysis requires an interaction term that's based on the neuronal response of the seed region and all other voxels, *not* the hemodynamic response, as is recorded in an fMRI time series. The need to do this is predicated on the fact that interactions at the hemodynamic level do not necessarily reflect interactions at the neuronal level (per the data they provide in the article), and it's the neuronal interactions that one typically cares about in fMRI. My goal is to use AFNI to implement the type of analysis that they outline in the article (and from what I understand is now standard in SPM's PPI toolbox). My understanding is that creating the interaction term they specify requires:
1) Extracting time series data from a seed region
2) Deconvolving the time series from 1) using an a priori hemodynamic response function
3) Multiplying the deconvolved signal from 2) by a function representing the experimental task
4) Convolving the result of 3) with the a priori hemodynamic response function
... then using the output of 4) as an explanatory variable for input to multiple regression (e.g., 3dDeconvolve).
My understanding of this procedure is that 2) above involves the creation of a sort of "neuronal response" time series via the deconvolution of the original time series with an HRF. Stating this in the terms of signal processing, I'd think that the result of this would be a sort of post-hoc impulse response function - hence my original question about how to extract such a thing via AFNI tools. It could be that the neuronal response time series I have in mind is analogous to "mind-reading," though I I'm not trying to make any inferences about the mind using this parameter alone... just trying to predict connectivity between brain regions as a function of my task.
Is it possible to use AFNI tools to implement this sort of analysis, and if so, what's the command syntax?
Thanks,
John Herrington
The reference:
D.R. Gitelman, W.D. Penny, J. Ashburner, and K.J. Friston. Modeling regional
and psychophsyiologic interactions in fMRI: the importance of hemodynamic
deconvolution. NeuroImage, 19:200-207, 2003