Re: [SPM] MarsBar ROI intensity values

Hi,

> I have written a code in Matlab to extract intensity values from fMRI data
> by applying MarsBar ROI on it. The output of this code is an array of
> intensity values.
>
> I used MarsBar to verify if these values are right. However MarsBar gives me
> only a 'single mean intensity value', which does not match with the mean of
> my intensity array.
>
> 1) I wanted to know if there is a way in which I could get the array of
> intensity values under the ROI from MarsBar.

MarsBaR does get the raw intensity values given the right inputs to
the routines via the GUI or batch - see for example:
http://marsbar.sourceforge.net/faq.html#raw_timecourse

MarsBaR does by default just calculate the standard mean - what are
you comparing the marsbar output to exactly?

Thanks,

Matthew

-------------------------------------------------------------------------
Using Tomcat but need to do more? Need to support web services, security?
Get stuff done quickly with pre-integrated technology to make your job easier.
Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642

percent signal change and number of voxels qn

-------------------------------------------------------------------------
Using Tomcat but need to do more? Need to support web services, security?
Get stuff done quickly with pre-integrated technology to make your job easier.
Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642

_______________________________________________
Marsbar-users mailing list
Marsbar-users <at> lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/marsbar-users

Re: percent signal change and number of voxels qn

Hi,

> I am trying to use MARSBAR to do some ROI analyses with our fMRI data. We
> have an event related study with 3 event types. We want to extract the
> percent signal change values and the % (or number) of voxels activated in a
> particular anatomical ROI. Our aim is to do some statistical analyses on
> these numbers to evaluate differences among the 3 event types. I am
> extracting the % of voxels activated using the code snippet from your FAQs.
> However I do not know what threshold would be an appropriate threshold to
> use? ( I believe the threshold is specified in t value?).

The threshold is rather arbitrary of course - you could use some
uncorrected threshold equivalent to p<0.001 or similar, or work out
the small volume correction threshold for that volume...

> Also if I want to compare the events (analogous to a FMRI contrast - say
> Event1 vs Event2), I would simply do a subtraction of the percent change
> values right?

Yes, you could do that - this will be the same as the normal contrast
if you have used only the HRF regressor, otherwise it will depend on
the exact shape of the response - for example you can imagine a longer
but lower response for event 2 that would mean that the max would be
greater for response 1, even though the area under the curve would be
greater for event 2 - but this is the normal problem of comparing
events made more obvious...

Best,

Matthew

-------------------------------------------------------------------------
Using Tomcat but need to do more? Need to support web services, security?
Get stuff done quickly with pre-integrated technology to make your job easier.
Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642

Re: percent signal change and number of voxels qn

-------------------------------------------------------------------------
Using Tomcat but need to do more? Need to support web services, security?
Get stuff done quickly with pre-integrated technology to make your job easier.
Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642

_______________________________________________
Marsbar-users mailing list
Marsbar-users <at> lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/marsbar-users

Re: percent signal change and number of voxels qn

Hi,

> I am getting very small percent signal change values for each of my event
> types. The maximum percent signal change output from marsbar was 0.7 and the
> lowest was -0.4
> Are these values too smal?? (do you think the change is not strong enough?)

No, they sound fairly ordinary - changes are low compared to block
design visual tasks of course, but that is because the events are
short and cognitive activations are typically smaller.

> From the archives, with an event related study, the event duration while
> extracting percent signal change is generally specified as 0. We also have
> an event related design with each event 3sec long. Using event durations of
> 3sec in one of the steps of extracting percent signal change resulted in
> higher percent signal change values. What is the right event duration to be
> specified?

If you specify 0, the interface (from SPM in fact) assumes that you
have something very similar to a 1 second long event.  If you specify
3, and the event really was 3 seconds long, this can give you a better
fit from the slightly longer event basis function, and could change
your signal change values.

> I can safely interpret negative percent signal change as deactivation right?

Er - well - you should look at the signal time course plots to
understand where the negative value came from - was it really an
inverted HRF.  In that case you are in the usual 'what does negative
signal change mean' bind in functional imaging.  It appears it can
mean reduced blood flow in some areas at least.

> Another question we have is that we have two runs of the same task. While
> specifying the design in SPM2 we have specified them as two sessions. We
> have applied a sessionwise grand mean scaling, however no global
> normalization was applied. When combining the two sessions (either for SPM
> analyses of percent signal change calculations) is global normalization
> necessary?

No, for FMRI, global normalization is almost always a bad idea, and
the default session mean scaling is reasonable,

Best,

Matthew

-------------------------------------------------------------------------
Using Tomcat but need to do more? Need to support web services, security?
Get stuff done quickly with pre-integrated technology to make your job easier.
Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642

filter effects in Plot Data (full)

Dear MarsBaR users,

I am sorry if this question has been asked a million times before, but I 
had trouble finding something related in the archives ...

I want to extract some data from an ROI from a bunch of subjects for 
off-line analysis (with spm_ancova).

I simply used Extract Data (default) and everything works fine. However, 
when I plot the data in the ROI (Plot Data (full)), then -- depending on 
the selection -- I see filter effects of the pre-whitening filter (I think, 
it's SPM.xX.V). That is, the first and the last data points of each session 
have a much, much higher value than all others. Additionally, it seems that 
the overall signal ramps up in the first few time points and decreases in 
the last few time points. (see attachment of a 2-session experiment for an 
example: page 1: with design filters and whitening; page 2: with design 
filters but no whitening; page 3: no design filters).

This (page 1) looks actually very similar (or identical?) to the 
eigenvariate that spm_regions returns. I have seen this before, but always 
forgot about asking the experts about it ...

So, here are my questions:
1. Is this a visualization effect, because it's part of the "Plot Data" and 
not "Extract Data"?
2. What is really the underlying time course like in the "Extract Data 
(default)" and in the "Extract Data (full)"? Are the filters of the SPM 
design applied (e.g. HPF and AR(1))?
3. Would you recommend using time courses in off-line analyses with or 
without the filters applied?

Maybe I should say a little bit more about the last question: I want to use 
the SPM design specification machinery to generate different design 
matrices (based on different model parameters from a learning model) and 
then fit this design to a time course from an ROI. I want to do this 
iteratively in the context of non-linear optimization. Thus, one further 
question is:

4. If you recommend the filtered time course (the one with the large values 
at the beginning and end of each session), doesn't this have a severe 
impact on my model fit, because these two data points will create a large 
residual? The filtered and pre-whitened SPM design matrix (SPM.xX.xKXs.X) 
does not exhibit these very high values at the beginning and end of each 
session).

Thanks a lot for your insights.
Jan

--

-- 
Jan Gläscher, Ph.D.         Div. Humanities & Social Sciences
+1 (626) 395-4976 (office)  Caltech, Broad Center, M/C 114-96
+1 (626) 395-2000 (fax)     1200 California Blvd
glascher <at> hss.caltech.edu    Pasadena, CA 91125

-------------------------------------------------------------------------
Using Tomcat but need to do more? Need to support web services, security?
Get stuff done quickly with pre-integrated technology to make your job easier.
Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642

_______________________________________________
Marsbar-users mailing list
Marsbar-users <at> lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/marsbar-users

Re: filter effects in Plot Data (full)

Hi,

> I simply used Extract Data (default) and everything works fine. However,
> when I plot the data in the ROI (Plot Data (full)), then -- depending on
> the selection -- I see filter effects of the pre-whitening filter (I think,
> it's SPM.xX.V).

Actually it's SPM.xX.W, but yes, you are right...

> That is, the first and the last data points of each session
> have a much, much higher value than all others.

> So, here are my questions:
> 1. Is this a visualization effect, because it's part of the "Plot Data" and
> not "Extract Data"?

Yes - probably - try plotting the extracted data in matlab outside
marsbar, I think you will probably see that the effect is gone...

> 2. What is really the underlying time course like in the "Extract Data
> (default)" and in the "Extract Data (full)"? Are the filters of the SPM
> design applied (e.g. HPF and AR(1))?

No - the SPM design filters are not used for the extract data options.

> 3. Would you recommend using time courses in off-line analyses with or
> without the filters applied?

Don't forget that if you filter the data, you also have to filter
whatever design matrix you are using for the extracted data.  If you
do that, it might be easier to do all filtering outside of SPM /
marsbar.  Or you could filter both in marsbar, and use the filtered
design and data outside marsbar.

Best,

Matthew

-------------------------------------------------------------------------
Using Tomcat but need to do more? Need to support web services, security?
Get stuff done quickly with pre-integrated technology to make your job easier.
Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642

Re: filter effects in Plot Data (full)

Dear Matthew,

thank you so much for you immediate and helpful response. May I ask a 
couple of quick follow-up questions ...

Matthew Brett wrote:
>> 1. Is this a visualization effect, because it's part of the "Plot 
>> Data" and
>> not "Extract Data"?
> 
> Yes - probably - try plotting the extracted data in matlab outside
> marsbar, I think you will probably see that the effect is gone...

You are absolutely right, but this makes me wonder about a related issue: I 
believe that most people you are doing PPI analyses use spm_regions to 
extract their data ... and spm_regions returns the first eigenvariate which 
displays those edge effects of the whitening filter. However, the design 
matrix does not have the edge effects, so it seems that there is a obvious 
mismatch between the data (eigenvariate) and the model, thus creating large 
residuals and a bad model fit which might influence the statistics. I 
wonder if you can comment on this ...

Maybe I should also post this question on the SPM mailing list and see what 
the folks at the FIL have to say about this ...

> 
>> 2. What is really the underlying time course like in the "Extract Data
>> (default)" and in the "Extract Data (full)"? Are the filters of the SPM
>> design applied (e.g. HPF and AR(1))?
> 
> No - the SPM design filters are not used for the extract data options.

So, does MarsBaR access the images via the SPM.xY.VY handle or rather 
directly? The reason I am asking is, that the global scaling factor is only 
written to SPM.xY.VY(i).pinfo(1,:), but not to the images themselves. Thus, 
when you DO NOT you access the images through SPM.xY.VY then you don't get 
the global scaling ...

see http://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind0701&L=SPM&P=R41923

Thanks a lot,
Jan

--

-- 
Jan Gläscher, Ph.D.         Div. Humanities & Social Sciences
+1 (626) 395-4976 (office)  Caltech, Broad Center, M/C 114-96
+1 (626) 395-2000 (fax)     1200 California Blvd
glascher <at> hss.caltech.edu    Pasadena, CA 91125

-------------------------------------------------------------------------
Using Tomcat but need to do more? Need to support web services, security?
Get stuff done quickly with pre-integrated technology to make your job easier.
Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642

Re: filter effects in Plot Data (full)

Hi,

> > Yes - probably - try plotting the extracted data in matlab outside
> > marsbar, I think you will probably see that the effect is gone...
>
> You are absolutely right, but this makes me wonder about a related issue: I
> believe that most people you are doing PPI analyses use spm_regions to
> extract their data ... and spm_regions returns the first eigenvariate which
> displays those edge effects of the whitening filter. However, the design
> matrix does not have the edge effects, so it seems that there is a obvious
> mismatch between the data (eigenvariate) and the model, thus creating large
> residuals and a bad model fit which might influence the statistics. I
> wonder if you can comment on this ...

I've never done PPI or other connectivity stuff because it scared me.
My memory of this is that you just put the filtered time course back
into the design matrix with some interaction applied, so in this case
the filtered time course will be filtered again, with the problems you
describe.  Maybe there is some mathematical magic that makes this OK,
but if so I don't know what that is.

> Maybe I should also post this question on the SPM mailing list and see what
> the folks at the FIL have to say about this ...

Certainly seems a reasonable question.

> So, does MarsBaR access the images via the SPM.xY.VY handle or rather
> directly? The reason I am asking is, that the global scaling factor is only
> written to SPM.xY.VY(i).pinfo(1,:), but not to the images themselves. Thus,
> when you DO NOT you access the images through SPM.xY.VY then you don't get
> the global scaling ...

The extract data default is to use the scaling from VY; the full
options extract data offers you this option.

Best,

Matthew

-------------------------------------------------------------------------
Using Tomcat but need to do more? Need to support web services, security?
Get stuff done quickly with pre-integrated technology to make your job easier.
Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642

displaying rois on a glass brain?

-------------------------------------------------------------------------
Using Tomcat but need to do more? Need to support web services, security?
Get stuff done quickly with pre-integrated technology to make your job easier.
Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642

_______________________________________________
Marsbar-users mailing list
Marsbar-users <at> lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/marsbar-users