Dear Alexis,
personally I don't see a strong reason, why unmodulated data should be additionally corrected for TIV or global GM. For the sake of completeness I have also considered this case on my modulation site:
http://dbm.neuro.uni-jena.de/vbm/segmentation/modulation
However, I have difficulties to interpret the meaning of this correction, because the brains are already corrected for different head size/TIV due to (affine) spatial normalízation. This is even more difficult to interpret if unmodulated segmentations are corrected by parameters derived from modulated segmentations.
The atlas scaling factor of spatial normalization (which is the inverse determinant of the affine transformation matrix) is a indeed good approximator for TIV:
http://surfer.nmr.mgh.harvard.edu/fswiki/eTIV
This is even valid for older people with dementia:
http://dx.doi.org/10.1016/j.neuroimage.2004.06.018
This was my motivation to not consider the affine transformation term for modulation if you want to correct for different head size or TIV. The results are the "m0" modulated images, where only the modulation for non-linear terms was applied. Most of the variance explained by age and gender will be removed from the data if you use either unmodulated data without additional correction or data modulated with the non-linear term only.
Keep in mind, that accuracy and spatial resolution of newer segmentation approaches (as used in spm5 or spm8) is largely improved. Most of the structural information is now encoded in the deformations and gray matter differences are diminished. Hence, I would rather use modulated images, where both effects (gray matter differences and deformations) are considered to get a gray matter volume, which is corrected for the changes due to non-linear spatial normalization. The extreme case is if you normalize your data using the Dartel approach. If you then check the unmodulated images you will notice the large similarity between all images (and will almost nothing detect after statistical testing).
Regards,
Christian
____________________________________________________________________________
Christian Gaser, Ph.D.
Assistant Professor of Computational Neuroscience
Department of Psychiatry
Friedrich-Schiller-University of Jena
Jahnstrasse 3, D-07743 Jena, Germany
Tel: ++49-3641-934752 Fax: ++49-3641-934755
e-mail: [log in to unmask]
http://dbm.neuro.uni-jena.de
On Fri, 14 May 2010 14:40:02 +0100, Cullen, Alexis <[log in to unmask]> wrote:
>
>Dear SPM experts,
>
>I am wondering if someone can help me with my query regarding controlling for total GM and WM in VBM analysis of unmodulated data.
>
>In brief, I am using the VBM5 toolbox and plan to examine both volume and concentration differences. To assess volume, I am using the modulated non-linear (m0) outputs from the toolbox, which, according to the Christian Gaser's website, do not need to be corrected for differences in brain size due to the absence of affine normalisation.
>
>My query relates to what to control for if you are analysing unmodulated images. I understand from previous threads that it is not absolutely necessary to control for total GM for unmodulated data, as the normalisation process should take account of different brain sizes, however, as registration is imperfect you may still want to control for total GM. I also see from the Christian Gaser's website that whether you decide to control or not depends on your question. Several VBM studies that analysed unmodulated data appear to have controlled for total GM/TIV , but others have not. I had intended to control for total GM and total WM and total in my analysis of GM and WM respectively.
>
>Although, I have not found anything specifically saying this, I assume that in an analysis of unmodulated data you would control for volume (i.e. modulated) not concentration? If this is the case, I am still wondering whether I calculate volumes based on my m0* (modulated non-linear only) files or whether I use mwc* (modulated affine and non-linear) files. Using spm_volumes I have generated volumes based on both my m0* and mwc* files, there is not a particularly strong correlation between the GM volumes (only marginally significant) but the WM volumes are strongly correlated, these findings encouraged me to examine the whether you get different results depending on which you use as a covariate. I have started to analyse some of my data using 3 approaches (i) unmodulated no covariate, (ii) unmodulated using total GM/WM volume based on mwc* images as a covariate, and (iii) unmodulated using total GM/WM volume based on m0* images as a covariate. There is some overlap with the results and interestingly the results for option 3 (controlling for volume based on m0) differ the most from the other two options which yield quite similar findings.
>
>I would be most grateful to hear your thoughts on (a) whether I am right in thinking that you contol for volume not concentration, and (b) if volume, whether I should be using m0* or mwc* files to generate volumes so that I can control for total GM/WM in my unmodulated analyses.
>
>Many thanks in advance, Alexis
>
>Alexis Cullen
>Research Worker/PhD Student
>R&R and CHADS projects
>Department of Forensic Mental Health Science (PO23)
>Institute of Psychiatry, De Crespigny Park
>Denmark Hill, SE5 8AF
>[log in to unmask]
>020 7848 5678
>
>
>
>Alexis Cullen
>Research Worker/PhD Student
>R&R and CHADS projects
>Department of Forensic Mental Health Science (PO23)
>Institute of Psychiatry, De Crespigny Park
>Denmark Hill, SE5 8AF
>[log in to unmask]
>020 7848 5678
>
>
>
|