In response to my question, Tom Nichols has provided valuable information
that should be shared on the email list.
If you are getting NaNs for the cluster analysis,
1) Check that the mask.img file that is generated when estimating the
model is the same as your explicit mask.
2) Check the RPV.img to make sure that the RPV.img contains finite values
for voxels included in your mask and then NaN everywhere else.
3) Calculate the FWHM.img using imcalc and the equation RPV.img^(-1/3) and
examine it again for finite values inside the mask. The values should be
roughly homogeneous; if they're wildly different it says you have
'nonstationarity', spatially-varying smoothness.
Nonstationarity is like homogeneous variance... if you look hard enough,
the assumption is probably never true. E.g. patients are probably always
more variable than controls; a more difficult task probably has a more
variable response than an easy one. And fMRI data, if you had enough data
to see it, doesn't have uniform smoothness.
For voxel-wise inference there's no problem... it doesn't depend on a
stationarity assumption. For cluster-size inference there is a problem...
and if you know you have nonstationarity (e.g. as you do in VBM) you
should use a non-stationarity-adjusted cluster test, as in
http://fmri.wfubmc.edu/cms/NS-General (or use a nonparametric procedure).
Most likely your nonstationarity was coming from including the ventricles,
which are different beasts than GM & WM. The easiest thing will simply be
to ensure that you're not analyzing any CSF voxels (or at least, exclude
the core of the ventricles).
Other useful tips were received: Use nearest neighbor interpolation and
unsigned integer 8bit when creating the mask using imCalc.
Overall solution: Lower the default threshold (originally set at 0.8) and
include the explicit mask combining gray and white matter.
Hope this helps other users.
> Dear SPMers,
>
>
> We would like to disable the analysis threshold to conduct stats in
> regions affected by susceptibility artifact. We have followed John's Gems
> to set sPM.xM.TH to -Inf and set an explicit mask by modifying SPM.xM.VM.
> We created the explicit mask by 1) adding the gray and white matter
> 2)smoothing the image (8mm) 3)masking all points. We selected double as
> our output type but otherwise kept the defaults. The explicit image is
> written out in NII format.
>
> When generating the stats table. The following errors are generated:
>
> Warning: Returning NaN for out of range arguments
>> In spm_Pcdf at 88
> In spm_P_RF at 106
> In spm_P at 48
> In spm_list at 489
>
> In addition, the cluster statistics are all reported with NaN.
>
> I am working in SPM5, but believe this problem would apply to any version.
>
> Any suggestions.
>
> Thanks,
> Jennifer
>
>
>
>
>
The information transmitted in this electronic communication is intended only for the person or entity to whom it is addressed and may contain confidential and/or privileged material. Any review, retransmission, dissemination or other use of or taking of any action in reliance upon this information by persons or entities other than the intended recipient is prohibited. If you received this information in error, please contact the Compliance HelpLine at 800-856-1983 and properly dispose of this information.
|