Meg,

 

I just read your initial posting, and I remember getting a similar error as yours at one point, namely betainc function being outside of its range. I don't remember exact details anymore, but I think it was an unusual combination of parameters that lead to an error. Anyway, I suggest you to try the following. I hope you can resolve the problem.

 

1. Check the FWHM at the bottom of your SPM results. If it's very small (e.g., FWHM<0.2 voxels), that typically means your analysis contains a large number of non-brain voxels due to smoothing associated with VBM data processing.

 

1.5. If you haven't done so, set a threshold (e.g., GM>15%) during the model-specification process to avoid an error explained in (1) above.

 

2. When SPM prompts you "FWE", "FDR", and "none", choose "none" and try 0.001 (or 0.01, 0.005 etc). If you choose "FWE" p<0.05, then SPM uses an FWE-corrected threshold at the VOXEL-LEVEL, which is often creates spurious results on cluster p-values. The corrected cluster p-values are FWE corrected, even if you choose "none", or an uncorrected threshold. Cluster p-values are adaptive, meaning that they are adjusted for the threshold chosen by the user.

 

3. Make sure you use an appropriate cluster-extent correction tool. Unfortunately you cannot use cluster p-values SPM produces directly if you are analyzing a VBM data set. There are two options out there to calculate cluster p-values correctly for VBM analyses: VBM5 toolbox by Christian Gaser, and the NS extension by Hayasaka et al (http://fmri.wfubmc.edu/cms/NS-General). If your lab does a lot of VBM processing, then either of these might be already available.

 

Good luck!

-Satoru

 

Satoru Hayasaka PhD ----------
Assistant Professor, Public Health Sciences & Radiology
Wake Forest University School of Medicine
(ph) +1-336-716-8504 / (fax) +1-336-716-0798
(email) shayasak _at_ wfubmc _dot_ edu

---

From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]] On Behalf Of Megan Walsh
Sent: Sunday, June 01, 2008 11:08 PM

To: [log in to unmask]
Subject: Re: [SPM] SPM 5 VBM: FWE/FDR errors

 

Thanks for your response.  I will definitely look into this. 

To clarify what I was doing.  I wanted to use FWE correction for multiple comparisons.  When this didn't work (with the error previously sent) I attempted a cluster level threshold.  I just want an FWE of 0.05. Is that clear?  So am I doing the right thing?

Thanks,

Meg

On Sun, Jun 1, 2008 at 5:34 PM, d gitelman <[log in to unmask]> wrote:

Megan

Are you trying to do cluster-wise thresholding in VBM using a probability
threshold? This is available in Christian Gaser's VBM toolbox, but is not
available in SPM, which only allows the specification of a number of voxels
for thresholding clusters. You will have to get the toolbox in order to do
the probability type of cluster thresholding
http://dbm.neuro.uni-jena.de/vbm/

Remember that once a toolbox is placed in the toolbox directory it is always
loaded into the spm path. This can give unexpected results with toolboxes
such as the VBM toolbox which uses some functions with the same names as
those in SPM.

Darren

> -----Original Message-----
> From: SPM (Statistical Parametric Mapping)
> [mailto:[log in to unmask]] On Behalf Of Thomas Nichols
> Sent: Sunday, June 01, 2008 4:28 PM
> To: [log in to unmask]
> Subject: Re: [SPM] SPM 5 VBM: FWE/FDR errors
>
> Dear Megan,
>
> This sounds very strange, as FWE and FDR calculations are
> quite different.
>
> Can you clarify exactly what you enter ('0.05' ?) and in
> response to exactly which question?
>
> Also not that you say 'cluster = 100', but that's just an
> arbitrary number (all clusters smaller than 100 are omitted)
> and not a specific FWE-level threshold.  FWE cluster
> inferences are read off of the table for any uncorrected
> threshold you give SPM.
>
> -Tom
>
>
>
> On Fri, May 30, 2008 at 2:56 PM, Megan Walsh <[log in to unmask]> wrote:
>
>
>       Hey fellow SPM-ers-
>
>       I am currently doing a multiple regression analysis on
> some VBM data.  I understand that cluster-wise thresholding
> is certainly not ideal for multiple comparison corrections.
> To correct for multiple comparisons I wanted to specify my
> FWE as p<0.05.  When I specify this I get the following error message:
>
>       ??? Error using ==> betainc at 42
>       X must be in the interval [0,1].
>       Error in ==> spm_Tcdf at 106
>       F(Q) = xQxPos
> -(xQxPos*2-1).*0.5.*betainc(v(Qv)./(v(Qv)+x(Qx).^2),v(Qv)/2,1/2);
>       Error in ==> spm_P_RF>spm_ECdensity at 157
>        EC(1,:) = 1 - spm_Tcdf(t,v);
>       Error in ==> spm_P_RF at 59
>       EC      = spm_ECdensity(STAT,Z,df);
>       Error in ==> spm_uc_RF at 47
>           p       = spm_P_RF(1,0,u,df,STAT,R,n);
>       Error in ==> spm_uc at 39
>       u = spm_uc_RF(a,df,STAT,R,n);
>       Error in ==> spm_getSPM at 602
>        u  = spm_uc(u,df,STAT,R,n,S);
>       Error in ==> spm_results_ui at 274
>        [SPM,xSPM] = spm_getSPM;
>       ??? Error while evaluating uicontrol Callback
>
>       The same idea goes for the FDR correction.  However, if
> I cluster threshold it works fine (cluster=100).
>       Can anyone help me figure out how to either a) fix the
> problem, or b) an alternative manner in which I can correct
> for multiple comparisons?
>
>       Thanks,
>       Meg
>