JISCMail - SPM Archives

On Thu, Apr 15, 2010 at 8:26 AM, <[log in to unmask]> wrote:

Thank you for the response Dr. Gitelman. I have had several excellent responses indicating mathematically why we would not expect to see the reciprocal PPI, and the deconvolultion fits in nicely. However, I also have seen a response indicating that we can infer area A is driving activity area B, and not the other way around. Is this a valid inference to make from my example? I thought we couldn't inference directionality or causality from a functional connectivity analysis.

PPI is not a "functional connectivity" analysis for several reasons. First it doesn't depend on differences in timing of activation per se. The entire time series is examined instantaneously so one is not looking if an area is active before or after some other area. Second, rather than depending simply on the correlation of time courses, it tries to provide additional explanatory power by incorporating either an interaction with a psychological context (or experimental condition) in the case of psychophysiologic interactions or the interactions between the signals from 2 regions in the case of physiophysiologic interactions. In fact PPI specifically discounts the simple "correlation" between the source region (A) and a target region (B) by including this as confound in the design matrix (i.e., the main effect of region A).

In his original paper Karl referred to PPI's as models of contribution, rather than as models of either functional connectivity (for the reasons given above) or effective connectivity. Actually it is a model of effective connectivity, albeit a very simple one in which the contributions from a single area to other regions is being examined (see pages 219 and 220 from Karl's 1997 paper in Neuroimage vol 6). If more regions were included it would be a better model of effective connectivity as is possible with DCM.

As far as causality is concerned, my simple understanding is that this is best thought of in terms of what contributes to the variance or information in a region rather than in relation to timing differences, which would be difficult to infer based on fMRI data in any case.

So yes you can say that area A influences are B if there is a significant PPI with area A as the source. If you also want to say that area B does or does not influence area A you would have to run the opposite PPI.

I think I was/(am) suffering from a misunderstanding of exactly what a PPI is. I have always considered it to be that two areas exhibit significantly increased functional connectivity (i.e. increased correlation of activity) in one condition compared to another. Therefore, if we are detecting increased correlation between two areas, it wouldn't matter which one we selected as the seed. I think I need to sit down a examine this a little bit more.

This would be true if you were just examining the main effect and not the interaction and there was no HRF so that the equation was symmetric. Also different main effects are being included in each of the PPI's. In the equations below Xna and Xnb = deconvolved signal from each region, Ya and Yb = BOLD signal from each region and P is the psychological vector. HRF( ) = convolution with HRF.

GLM equation (leaving off other confounds and error) for the PPI with region A as its source:

Yb = beta1*(HRF(Xna * P)) + beta2*Ya + beta3*P does not equal the equation with region B as its source

Ya = beta1'*(HRF(Xnb * P)) + beta2'*Yb + beta3'*P

Regards,
Darren

Again, thank you for your reponse - and your time!

Christopher

----- Original Message -----
From: Darren Gitelman <[log in to unmask]>
Date: Wednesday, April 14, 2010 7:42 pm
Subject: Re: [SPM] Interpreting PPI directionality?
To: Christopher Wilson <[log in to unmask]>
Cc: SPM <[log in to unmask]>

> Dear Christopher
>
> You are correct. The PPI analysis is not symmetric. This is
> because of
> convolution with the HRF.
>
> Darren Gitelman, MD
>
>
> On Tue, Apr 13, 2010 at 8:59 AM, Christopher Wilson
> <[log in to unmask]>wrote:
> > Hi SPMers, I had a question about interpreting PPI
> results. The question
> > is probably best illustrated with an example.
> >
> > I have run my standard whole-brain analysis and identified two
> interesting> peaks (A and B) from the group-level analysis. I
> then wanted to run two PPI
> > analyses using each region as a seed. When I run a PPI using
> area A as my
> > seed region, I find that area B displays a significant
> pyschophysiological> interaction with area A. However,
> when I use area B as the seed, I do NOT
> > find the same effect with A.
> >
> > How does one interpret or explain this result? I found a
> post by Karl
> > Friston (
> > https://www.jiscmail.ac.uk/cgi-
> bin/wa.exe?A2=SPM;kwlgAQ;20100108150238%2B0000),> which makes
> reference to this, but I still do not understand why this
> > happens, or what it means.
> >
> > Does anyone have some insight?
> >
> > Best wishes,
> >
> > Christopher Wilson
> >
>