AM regression on Unassumed

AFNI Message Board
, ,
1

AFNI version info (20.2.16):

Hi AFNI team,

My fMRI experiment is similar to a classical monetary incentive delay (MID) task (Knutson, Brian, et al. 2000). We used the paradigm to assess one's neural response to the anticipation and receipt of rewarding monetary stimuli.
I modelled my reward cue using CSPLINZERO and task using GAM. I have two questions:

  1. Is it possible to do an AM regression using unassumed (or CSPLINZERO)? Or does AM regression require some assumed shape (e.g. GAM)

  2. I understand that in CSPLINzero the first and the last point are taken to be 0. But what are the pros and cons of using CSPLINZERO over CSPLIN? or vice versa.

Thanks,
Sahithyan

2

Sahithyan,

It is fine to adopt a modulation approach using the basis function CSPLINzero. The assumption is that the BOLD response exhibits a linear association with the modulation variable.

I understand that in CSPLINzero the first and the last point are taken to be 0. But what are the pros and cons of using CSPLINZERO over CSPLIN? or vice versa.

Under normal circumstances when the BOLD response aligns with the stimulus onset, CSPLINzero is a more suitable choice. It avoids allocating a regressor that merely captures random noise at the stimulus onset. Conversely, in situations where participants exhibit anticipation or actions before the stimulus onset, CSPLIN might be a better fit.

Gang Chen

3

Thanks Gang Chen!

I have a follow up question on adapting modulation approach using the basis function CSPLINzero.

So, If I want to modulate my FMRI task activation using the response time as amplitude factor and using CSPLINzero as basis function

Something like this:

*-stim_times_AM2 1 $ M_TaskOnset_RT_married 'CSPLINzero(0,15,7)' -stim_label 1 task_Onset_RT *

My TR = 2.5s

But I am having trouble making sense of the output. So, I will be getting 5 RT modulated values (for each of the 5 time points)?

4

Hi Sahithyan,

Since AM2 is specified and there is a single modulator for a 5 parameter model, there should be 10 betas (plus stats) output: the 5 main/mean h(t) betas for CSPLIN, plus 5 corresponding h(t) betas for the modulator. If there were more modulators, each would add another 5 betas, one corresponding to each of the main parameters. And they should be in that order.

Does that seem like what you are seeing?

  • rick
5

Hi Rick!

Yes, I get 10 betas (plus stats). I was initially confused regarding the order of the betas when there are multiple modulators. Thanks for the clarification.

Regards,
Sahithyan