Hello Philipp,
>Dear SPM-users,
>
>here comes an admittedly very basic question on a VBM analysis (SPM2, two
>groups, GM and WM maps, three covariates. We are mainly interested in volume
>differences and use modulated GM (WM) after optimized preprocessing).
>
>Discussions came up, which model to use and how to encode the group variable.
>(1) Multiple regression with constant: should I estimate two betas, i. e. use
>one regressor 1 (..) 0 (...) and the other inversely? Is there an advantage
>(apart from one degree of freedom) if effect coding is used, i. e. -1 and 1
>in one regressor and estimate one coefficient?
>or alternatively
>(2) ANCOVA?
>What role do different variances play in the two models?
>
>
Well, if you want to 'remove' the effect of several covariables, ANCOVA
models are well suited. However, if you use a multiple regression with
two columns for your groups, then you will have exactly the same
results. Looking at the GLM, the way to code an ANCOVA, is not different
to the way to code an ANOVA (or t-test for you); you just add a
factor/column.
>Two more questions relate to the estmation of total GM and WM: can we simply
>use the total signal in a modulated (cleaned) GM (WM) map?
>
>
?? I do not know.. What people usually do, is to use the GM/WM volumes.
Now, if you refer to the value of the modulated images, I guess it's
possible (this value*nb voxels = volume; no? - and all brains are
normalized/same nb of voxels...)
>For total intracranial volume (TIV): is the sum of the totals of the three
>(GM, WM, CSF) (cleaned) modulated maps a sufficiently precise value to
>control for interindividual head size / TIV?
>
>
Given the previous e-mail of John Ashburner I don't think so.
July 27:
"I wouldn't trust the CSF volumes produced by SPM2. The intention was not
really to produce CSF segments, but some people wanted them, so they were
retained. You may need to do some thresholding in order to get a more
accurate guess of the volume (look at the data with an inverse grey scale),
but for many datasets, the CSF and skull intensities are very close, so
accurate segmentation is not possible."
Hope it helps,
Best
- cyril
|