what's the difference between image and cluster image when export?

------------------------------------------------------------------------------

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

Re: what's the difference between image and cluster image when export?

Hi,

> I just installed and use this to extract the ROI.
> When export, it asks me about where to export ROI (e.g., to image/ clustered
> image/ number labeling image. )
> I tried image and clustered image,but I don't see the difference in the
> MRIcro between those two although the file size of clustered image is
> bigger.

The difference is, that if you have more than one cluster, the
clusters will have different numbers is their voxels, e.g. ROI1 will
have 1s in its voxels, ROI2 will have 2s in its voxels, and so on...

Best,

Matthew

------------------------------------------------------------------------------

Percent signal change and FIR vs. beta values

------------------------------------------------------------------------------
Check out the new SourceForge.net Marketplace.
It is the best place to buy or sell services for
just about anything Open Source.
http://p.sf.net/sfu/Xq1LFB

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

Re: Percent signal change and FIR vs. beta values

Hi,

> I extracted percent signal change, beta values and FIRs for the condition in
> my design. However these measures differed dramatically. The values from the
> FIR time-courses matched the values from the percent-signal change data,
> however, the beta weights were completely different. I tried to do so for 3
> independent studies and every time I got the same problem. So I went on to
> compare the extracted data with the SPM-maps and the beta weights
> corresponded well with the activations shown in SPM5. In contrast, percent
> signal change and the FIRs did not match what SPM5 showed me: e.g. I defined
> one ROI by use of an activation cluster A>B. The beta weights showed me
> therefore A>B. However the percent signal change and the FIR shoed an
> opposite behavior. How is this possible? Shouldn't the beta values, the
> percent signal change and the height of the FIR be related?

I don't understand this either.  The percent signal change, by
default, uses a function of the beta weights and the mean signal.  So,
if the beta is positive, the percent signal change should also be
positive.

You mention A > B.  Do you mean, that you extracted a percent signal
change for A, and a percent signal change for B, and found that signal
change for B was greater than signal change for A?  Were A and B in
the same session?

The FIR is more complicated, because it introduces a more flexible
model, that can either pick up unexpected time course behavior, OR
pick up unmodeled stuff from other effects OR pick up noise.  Any of
these can cause odd effects that are difficult to predict.

Best,

Matthew

------------------------------------------------------------------------------
Check out the new SourceForge.net Marketplace.
It is the best place to buy or sell services for
just about anything Open Source.
http://p.sf.net/sfu/Xq1LFB

Re: Percent signal change and FIR vs. beta values

Hi,

> I don't understand this either.  The percent signal change, by
> default, uses a function of the beta weights and the mean signal.  So,
> if the beta is positive, the percent signal change should also be
> positive.
>
> You mention A > B.  Do you mean, that you extracted a percent signal
> change for A, and a percent signal change for B, and found that signal
> change for B was greater than signal change for A?  Were A and B in
> the same session?
>
> The FIR is more complicated, because it introduces a more flexible
> model, that can either pick up unexpected time course behavior, OR
> pick up unmodeled stuff from other effects OR pick up noise.  Any of
> these can cause odd effects that are difficult to predict.
>
> Best,
>
> Matthew

the percent signal change and the beta values (averaged over my e.g. 6
session) are not correlated at all.
This means I can have positive beta (averaged over my 6 session by myself)
and a negative % signal change and
vice versa. All my conditions were equally distributed and occurred in every
single session. 
However, to explain the A>B problem in more detail: I defined a ROI in an
attention switching paradigm where I found activations 
of the intraparietal sulcus only for the switch attention condition (T~10.0
!!!) but no activation for the hold attention condition in SPM5. 
I defined a spherical ROI of 5 mm at the maximum. I extracted the beta
values for all my conditions, averaged them over session and afterwards over
subjects. I did the same for % signal change and for the FIR (without
averaging over sessions because MarsBar does by itself). When I looked on
the 
Results I saw something like 
Betas:              Switch: 0.85; Hold 0.52
% signal change:    Switch: 0.06; Hold 0.10
FIR at 6 seconds:   Switch: 0.05; Hold 0.11

So the beta values and the percent signal change have an opposite behavior,
which unfortunately in both cases leads to a significant difference between 
Both conditions. I trust the results from the betas, because they reflect
what I see when looking into the SPM results. However I don't really see,
why 
Both measures aren't correlated... 

Greetz

Chris

------------------------------------------------------------------------------
Check out the new SourceForge.net Marketplace.
It is the best place to buy or sell services for
just about anything Open Source.
http://p.sf.net/sfu/Xq1LFB

Multiple comparison correction for several ROIs

------------------------------------------------------------------------------
This SF.net email is sponsored by:
SourcForge Community
SourceForge wants to tell your story.
http://p.sf.net/sfu/sf-spreadtheword

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

Re: Multiple comparison correction for several ROIs

Hi,

> Do you think if we need multiple comparison correction for many ROIs in
> group analysis? If so, what way would be appropriate, bonferroni or FDR?

Yes, if you don't have a strong hypothesis about which of the regions
you are expecting signal in, multiple comparison correction is the
standard.   The choice of Bonferroni or FDR depends on whether you
accept the tendency of FDR to give you a greater number of false
positives when more regions are activated.

> But
> if we do the correction, then the more ROIs we test, the less possibility we
> can get significant results.

That's true - but that is to control the possibility of false positives.

Best,

Matthew

------------------------------------------------------------------------------
This SF.net email is sponsored by:
SourcForge Community
SourceForge wants to tell your story.
http://p.sf.net/sfu/sf-spreadtheword

Re: Percent signal change and FIR vs. beta values

Hi,

> However, to explain the A>B problem in more detail: I defined a ROI in an
> attention switching paradigm where I found activations
> of the intraparietal sulcus only for the switch attention condition (T~10.0
> !!!) but no activation for the hold attention condition in SPM5.
> I defined a spherical ROI of 5 mm at the maximum. I extracted the beta
> values for all my conditions, averaged them over session and afterwards over
> subjects. I did the same for % signal change and for the FIR (without
> averaging over sessions because MarsBar does by itself). When I looked on
> the
> Results I saw something like
> Betas:              Switch: 0.85; Hold 0.52
> % signal change:    Switch: 0.06; Hold 0.10
> FIR at 6 seconds:   Switch: 0.05; Hold 0.11

Hmm.   Just to check, you are using the '% signal change' option in
the marsbar menu, you haven't to your knowledge changed the difference
function for the event, and you are not using the 'FIR event time
course' for the the '% signal change' option above?

I would expect that, if you extract the betas within sessions, and the
percent signal change within sessions, for these to be well-correlated
(before averaging across sessions).  Is that the case?

Best,

Matthew

------------------------------------------------------------------------------
This SF.net email is sponsored by:
SourcForge Community
SourceForge wants to tell your story.
http://p.sf.net/sfu/sf-spreadtheword

Re: Percent signal change and FIR vs. beta values

> Hmm.   Just to check, you are using the '% signal change' option in
> the marsbar menu, you haven't to your knowledge changed the difference
> function for the event, and you are not using the 'FIR event time
> course' for the the '% signal change' option above?
> 
> I would expect that, if you extract the betas within sessions, and the
> percent signal change within sessions, for these to be well-correlated
> (before averaging across sessions).  Is that the case?

I don't use the marsbar menu. I scripted everything. For the extraction of
the betas and the % signal change I use:

    D  = mardo(spm_name);
    R  = maroi(roi_file);
    Y  = get_marsy(R, D, 'mean');
    E  = estimate(D, Y);
    E  = set_contrasts(E, xCon);
    ets = event_types_named(E);
    n_event_types = length(ets);	
    run_length = size(block_rows(E),2);

% here I state if movement params are used in the design 

    if movement_reg == 1
    reg_num = 6;
    else
    reg_num = 0;
    end    

    beta_values_raw = betas(E);

    beta_to_beta = n_event_types + reg_num;    

% here I average the betas over the sessions

        for z=1:n_event_types
            a=[beta_values_raw(z)];    
            for y=1:run_length-1
                a=[a, beta_values_raw(z+beta_to_beta*y)];
            end
            b=mean(a);
            beta_mean{z}={['con', num2str(z),' - mean beta-value'];b};
        end

% here I extract % signal change, which is not extracted per session but
averaged over all sessions (as far as I know)

    for e_t = 1:n_event_types
        dur = 0;
        pct_ev(e_t) = event_signal(E, ets(e_t).e_spec, dur);
    end

% here I get the FIR per event

    bin_size = tr(E);
    fir_length = 24;
    bin_no = fir_length / bin_size;
    opts = struct('single', 1, 'percent', 1);

    for e_t = 1:n_event_types
        fir_tc(:, e_t) = event_fitted_fir(E, ets(e_t).e_spec, bin_size, ...
        bin_no, opts);
    end

So as you see here, I don't know if the betas and the % signal change are
correlated within a session, because the script line

%%%%%%%% pct_ev(e_t) = event_signal(E, ets(e_t).e_spec, dur) %%%%%%%

gives me the percent signal change per condition for the entire experiment
and not per session (as far as I understand).

And to my knowledge I didn't changed the difference function for the event
(or did I???), 
and as far as I can determine I am not using the 'FIR event time course' for
the '% signal change' (or do I???)

Sorry for all this.......

Greetz

Chris

Dr. med. Christian Stoppel
Klinik für Neurologie, 
Universitätsklinikum Magdeburg A.ö.R 
Leipziger Str. 44, 39120 Magdeburg
Tel.: +49391-6117537
Fax.: +49391-6117531 

------------------------------------------------------------------------------
This SF.net email is sponsored by:
SourcForge Community
SourceForge wants to tell your story.
http://p.sf.net/sfu/sf-spreadtheword

obtaining label for 3D coordinate in MNI space

------------------------------------------------------------------------------
This SF.net email is sponsored by:
SourcForge Community
SourceForge wants to tell your story.
http://p.sf.net/sfu/sf-spreadtheword

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