Dear Antanas,
I'll answer your off-list message on the list because I think the
questions are of general interest. First I'll summarize the details
you gave me in generic terms. You have two populations of subjects,
lets call them patients and controls. You are showing them a series of
visual stimuli, each presented for several hundred ms. Just two
stimuli in each trial are really important. There are two conditions A
and B. You expect that there won't be a difference between these
conditions for controls, but there will be for patients and you have
some specific ideas for two models of possible connectivity changes.
This sounds two me like a classical DCM-ERP situation. You don't
actually have a 4s continuous stimulus but two discrete events of
interest that elicit ERP peaks so it's perfectly OK.
There are two approaches you can take - modelling the grand averages
and modelling the subjects individually. You'd probably want to try
both. The advantage of modelling the grand average is that you will
have very clear ERPs and you will be able to specify differences
between controls and patients directly in the DCM. The advantage of
modelling individually is that you might be able to see differences
between patients (although 6 sounds like too few for attempting
classification) and you can do classical statistics on parameter
estimates. I suggest that you start with the grand average.
You should merge all the individual ERPs in one file, rename the
conditions if necessary to something like 'Control A', 'Control B',
'Patient A', 'Patient B' and then average. So you'll have a dataset
with 4 trials, 4 conditions. Make sure that the order makes sense to
you. You can use 'Other/Sort conditions' for that. I suggest that you
cut your trials in such a way that they end before the third
irrelevant stimulus because you don't want to include it in your
model.
Then you load this file in DCM and in 'between trial effects' part
where it says 1 by default you should write 1 2 3 4 (that's where it's
important to know the order). I assume that the order is as above.
Then you can specify 3 eexperimental effects: effect of group (control
vs. patient) effect of condition and the interaction between the two.
Your hypothesis is actually about interaction because you expect that
there will be a difference between conditions only for one of the
groups. Look at http://www.socialresearchmethods.net/kb/expfact.php to
learn more about factorial designs.
So in the text box on the left you can write the names for those
effects and in the central big box you should write the contrast
coefficients. It should look something like the following:
| 1 2 3 4
-----------------------------------------------
group | -1 -1 1 1
condition | -1 1 -1 1
interaction | -1 1 1 -1
Then you go to the 'electromagnetic model' par and specify your
sources of interest. You should have all the sources in all the
models. If for some models there are sources that are not involved,
just don't connect them.
Now, something specific for you. You have two stimuli of interest. It
sounds like you are only modelling the response to the second one, but
perhaps it'd be better to model both because your baseline doesn't
look very flat. So one option would be two have the baseline before
the presentation of the first stimulus and then in the 'onset[s] (ms)'
part you can specify two numbers, lets say 100 500 (these should be
the times the stimulus reaches V1 rather than when it's presented).
Then you'll have two inputs in the connectivity matrix and you can
connect them both to V1. Alternatively you can use high-pass filter
with higher cutoff to make the baseline flat, but you should make sure
you still have some differences between conditions in the ERPs.
Then in the neuronal model as I understand your basic model is fixed
and the differences between the hypotheses are in connections affected
by the experimental effects. I think it'd make sense to put all the 3
experiemental effects on the same connections.
For the null model you just don't have experimental effects, just the
basic model.
You can then use BMS to compare those models.
In the single subject case you'll have only one experimental effect,
that of condition. You should still merge the two conditions per
subject in one file. Then you can compare models within subjects and
also get parameter estimates (probably the condition effect on
connections is the most interesting one) and do a t-test betwen
patients and controls.
I hope this helps. I'd be happy to answer any further questions you have.
Best,
Vladimir
On Tue, Aug 18, 2009 at 12:35 PM, Spokas
Antanas<[log in to unmask]> wrote:
> Dear Vladimir,
>
> Thank you for your response, but my stimulus is visual, that lasts 4 sec,
> does it not change the modelling itself, from giving a short burst of
> impulse (as in auditory for instance) to primary area, or lasting
> stimulation? Thank you,
>
> Antanas
>
> ________________________________
> From: Vladimir Litvak <[log in to unmask]>
> To: Spokas Antanas <[log in to unmask]>
> Cc: [log in to unmask]
> Sent: Tuesday, 18 August, 2009 12:30:11
> Subject: Re: [SPM] DCM changing neural model priors
>
> Dear Antanas,
>
> The parameters in the 'Review priors' window are for advanced
> applications and you wouldn't normally go there for a standard ERP
> study. The things you are referring to are just for plotting the
> transfer functions and they do not affect your actual DCM inversion
> and do not need to be saved. I'd recommend you to keep the default
> settings and only change the parameters in the main DCM window.
>
> Best,
>
> Vladimir
>
> On Tue, Aug 18, 2009 at 12:05 PM, Spokas
> Antanas<[log in to unmask]> wrote:
>> Dear SPM,
>>
>> I would like to change some of the priors in neural model of DCM
>> (peristimulus time to 398ms, recording frequency to 500Hz) as they are set
>> to those from mismatch negativity study, but when I save and then reopen
>> my
>> DCM again and load priors it again shows default settings? Please, advice.
>> Appreciate.
>>
>> Regards,
>> Antanas
>>
>
>
|