> I am currently working on a voxel-based morphometry.
> When I estimate a model (multiple regression with three covariates) using an absolute threshold of 0.2 the estimation fails with the error message
>
> "No inmask voxels - empty analysis!"
>
> appearing in a new window. In the command window there no error is indicated.
>
> I already figured out that it must have to do with the proportional normalisation function since using an ANCOVA for global normalisation (just gave it a try though not appropriate in this case) did not end up in an error and the results made sense (most of CSF and the majority of WM was not analysed).
That's a good approach to figuring out how things work.
>
> As I am using total intracranial volumes in ml as global values for the "global calculation" I further noticed John Ashburner's comment in his 2010 VBM tutorial "that with Proportional scaling correction, the values are divided by the globals prior to generating the mask" ( http://www.fil.ion.ucl.ac.uk/~john/misc/VBMclass10.pdf ). For my model this would mean, that the voxel-values are devided for example by 1500(ml, total intracranial volume), which of course even for a voxel-value of 1 would result in a value of 1/1500. If this is the reason of all voxels being excluded from the analysis then would you suggest some kind of transformation of the globals around 1?
This seems like a very plausible explanation.
>
> Moreover, i am wondering, why the default value of absolute threshold is "100" since all values need to be >/=0 and </=1.
This value of 100 was established in the early days of PET, although I
don't know where it comes from exactly. I would guess people were
scaling their data so that the average intensity was 100 - or
something like that.
Best regards,
-John
|