Please see inline responses below.
Best Regards, Donald McLaren
=================
D.G. McLaren, Ph.D.
Postdoctoral Research Fellow, GRECC, Bedford VA
Research Fellow, Department of Neurology, Massachusetts General Hospital and
Harvard Medical School
Office: (773) 406-2464
=====================
This e-mail contains CONFIDENTIAL INFORMATION which may contain PROTECTED
HEALTHCARE INFORMATION and may also be LEGALLY PRIVILEGED and which is
intended only for the use of the individual or entity named above. If the
reader of the e-mail is not the intended recipient or the employee or agent
responsible for delivering it to the intended recipient, you are hereby
notified that you are in possession of confidential and privileged
information. Any unauthorized use, disclosure, copying or the taking of any
action in reliance on the contents of this information is strictly
prohibited and may be unlawful. If you have received this e-mail
unintentionally, please immediately notify the sender via telephone at (773)
406-2464 or email.
On Fri, Sep 9, 2011 at 7:42 PM, Ce Mo <[log in to unmask]> wrote:
> Dear All,
>
> This is my first time performing PPI analysis, and I would really appreciate it if you could provide me with some answers.
>
> Suppose I have two psychological factors A and B, each of which has 2 levels, say A1 A2 B1 B2 respectively. Moreover, I also specified a source seed region R. My objective is to identify the conjunction of regions whose activation can be interpreted as the result of interaction between the source region R and 2 psychological factors A and B. I did a little thinking myself, and came up with 2 possible solutions.
Let's say you have 4 conditions and make it easier and because you
should have modelled it as four conditions in your activation
analysis.
>
>
> 1) Generate 2 individual PPI structures, and thus I got 2 psychophysiological interactions of interest R*A and R*B. Now I combine the 2 factors, the 2 Interactions and R in one single GLM? That is,
>
> y = b1R*A + b2R*B + R + A + B
I assume you mean y=R*A+R*B+A+B and b1 and b2 are the beta values of interest.
You haven't described what A or B represents, but for now I will
assume that A means the difference between A1 and A2 and the same for
B.
My personal suggestion is why stop there, you should model all four conditions.
y=R*A1+ R*A2 + R*B1 + R*B2 +R +A1+A2+B1+B4
In fact, this is exactly what the automated gPPI toolbox does
(http://brainmap.wisc.edu/PPI).
>
> and perform a conjunction of the first two regressors, in other words, conjunction of T contrast [1 0 0 0 0] and [0 1 0 0 0] in the first level. And then take the results into 2nd_level using within subject ANOVA and perform conjunction of contrasts [1 0] and [0 1]
You can't take conjunctions from the first level and use them in the
second level. If you are truly interested in the conjunction -- making
the claim that there is overlap -- then you want to demonstrate it in
the single subject maps.
Contrasts for interaction of A would be [1 -1 0 0 0 0 0 0 0] and B
would be [0 0 1 -1 0 0 0 0]. If you only want this direction of the
test (e.g. A1>A2 and B1>B2), then use a t-test. Otherwise you want to
use an F-test and form the conjunction.
Now that you have single subject conjunctions. Threshold (peak_nii
(http://www.martinos.org/~mclaren/ftp/Utilities_DGM) and convert them
to binary values (imcalc:i1>0) for each subject. Then add up all the
subjects and you will be able to state how many subject overlap at
each voxel.
Taking con_ images to the second level for each contrast and then
doing a conjunction at the second level asks a different question.
Whether the locations of significant group amplitudes overlap, which
doesn't say that individuals overlap because of the statistic being
computed. group mean/standard deviation of the group. If the standard
deviation is lower, for the same mean, then the activation area will
likely be bigger just based on that alone.
>
> The problem of this approach is that I don't know whether the 1st level GLM formed in this way is valid. It seems reasonable to me but I have never seen anybody did this before in published works...
A paper is currently under review on the method I mentioned.
>
> 2) Set up 2 separate GLMs in the traditional PPI fashion, that is,
>
> y = b1R*A + R + A with contrast [1 0 0]
> y = b2R*B + B + R with contrast [1 0 0]
This is bad as your only modelling some of the data which could lead
to errors in model fitting.
>
> and then taking the results into a common 2nd_level ANOVA analysis and perform conjunction in the same manner described above.
I think its fine to take contrasts to the second level, you just need
to be careful about how you interpret the conjunctions using group
maps.
>
> The problem concerning this method is that I wonder whether the results of 2 different 1st level GLMs can be put into 1 common Group Analysis.
Not sure, but since I know there are problems with modelling only some
of the data, I'd avoid using that method.
>
> If both methods are invalid, is there a third method to solve my problem?
See the gPPI toolbox.
> Any help and advice will be greatly appreciated Many thanks in advance!
>
> P.S. One more question about the underlying principle of PPI. Let's say I coded 2 factor levels A1 as 1 and A2 as -1. And I have a source region, say R. If the beta corresponding to the PPI regressor is positive(tested using T contrast [1 0 0]), then the result can be interpreted as the activity of R is stronger in the context of A1 than it is in the context of A2. Why is that?
The interpretation is that the activity of R during A1 leads to a
larger change in region B than A2. If R is +, then its a larger
increase, if R is -, then its a larger decrease. An illustration can
be found in Karl's 1997 paper. Also, you can see the effect by drawing
two lines, one with a slope of 3 and the other a slope of 1. The lines
intersect at X=0.
>
> Best Regards
> Ce
>
Hope you find this useful.
|