Hi everybody!
I have been using VBM8 in SPM12 for a few months now for volumetric analysis of the human Hippocampus, between two different groups. I'm completely aware that depending of the kind of modulation you apply to your data (affine vs non-linear) the different results you get. But in terms of correctness I didn't expect such a huge difference. Using the prior one (affine) and literately following the VBM8 manual I achieve very strange results being quite contradictory in comparison to the last named procedure. The major set-up difference between those two is the affine vs the non-linear selection and using the head size as covariate only for the affine method.
I have been thinking the possibility that this huge difference could be due to my contrast matrix being [-1 1 0 0 0] vs [-1 1 0 0] for the affine and non-linear method respectively. But it looks to be right for the two sample test as I have two groups and three covariates (age, head size (only for the affine) and HAMD.
I would appreciate if any of you could make this clear to me.
Thank you in advance!
Best regards; Cecilio.
Hi,
Could these differences be caused by the mere inclusion of an extra covariate (head size), which slightly alters the statistical model (e.g. losing one degree of freedom)? As the VBM manual states, the motivation to use non-linear only modulation is that it applies the brain size correction directly to the data, instead of in a statistical model. I would not have expected the difference between the two approaches to be huge, but I guess it depends on what you mean by a "huge" difference? It is interesting though, hope someone can clarify!
The default modulation in VBM8 is not considering the affine scaling of the spatial registration for modulating the data. This option was thought to get rid of the additional correction of intracranial volume (ICV) that is necessary if you use the standard SPM8 modulation. By ignoring the affine scaling in the modulation this is equivalent to a global scaling with the (inverse) scaling factor of the affine transformation. Please check the Freesurfer website for a more detailed explanation of the background of this idea:
https://surfer.nmr.mgh.harvard.edu/fswiki/eTIV
If you use the default modulation in VBM8 for non-linear effects only there is no need to additional correct for individual brain size (or ICV). Thus, it is important not to mix these steps. To further investigate the effect of the modulation and the impact of the affine scaling issue I recommend to use modulation with non-linear effects and try to add correct for individual brain size by:
1. global scaling with sum of GM and/or GM+WM
2. nuiscance parameter with sum of GM and/or GM+WM
The sum of GM (and maybe WM) should roughly approximate ICV and you can investigate more detailed where the large differences come from. However, I would not statistically compare all of these approaches because differences will be huge because they are so systematic (tiny variance in the groups).
Also check the SPM maillist for "modulation" to get more discussion on that topic.
Best regards,
Christian
Dear Cecilio,
As VBM8 Manual states: Affine+non-linear allows comparison of the absolute amount of tissue, while Non-linear only allows comparison of the absolute amount of tissue corrected for individual brain sizes. Do the two groups of subjects have significantly different average brain sizes? If so, the outcome measurement may differ 'hugely'.
Best,
Andy
- Badges
- Science topic
- Similar topics
- Biological Science
- Anatomy
- Tissue