Dear Natalia,
On Sun, 23 Oct 2016 21:26:32 +0200, Natalia Kowalczyk <[log in to unmask]> wrote:
>Dear SPM users
>
>I wan to compare two groups (experts vs control) (VBM- gray matter volume).
>I have a question about 2 level model:
>
>I have choosen this options:
>
>
>Design -> two sample t-test
>
>Independence -> Yes
>
>Variance -> Unequal
>
>Grand mean scaling -> No
>
>ANCOVA -> No
>
>Covariates -> age (Interactions -> none; Centering-> Overall mean)
>
>*Masking -> threshold -> 0.006
>
>**Global Calculation -> Global values -> TIV (total intracranial volume)
>like 1.5 or 1.7 for each participant
>
>Global normalization -> overall grand mean scaling -> Yes
>
>Grand mean scaled value -> 50 (default option)
>
>Normalisation -> Proportional
>
>
>*threshold for masking -original is 0.2 but global normalization will also
>affect the absolute threshold for the masking because my images will be now
>scaled to a global value of 50. So if the mean TIV is 1700 in my group all
>images are globally scaled to a value of 50. Thus, the overall scaling is
>50/1700 = 1/34
>To get the (old) absolute threshold of 0.2 now I use 0.2/34 = 0.006 – *am I
>correct?* *Is this a recommended/common procedure?*
This is correct if
- you use global scaling of TIV instead of using TIV as nuisance variable
- and if your mean TIV is 1700. If your mean TIV is different you have to adapt the equation above accordingly
I would recommend to use TIV as nuisance parameter for most cases. Only of TIV is correlating to much with parameters of interest the global scaling is the more appropriate approach.
>
>** TIV should be in format 1.5 not 1500 (ml) – *am I correct?*
No, TIV should be given in ml. Otherwise you have to again adapt the equation above.
Best,
Christian
>
>
>*I am not sure if my model is correct? **I would be very interested in any
>insights from you.*
>
>
>Prior to statistical analysis we took following pre-processing steps: [1]
>Checking for anatomical abnormalities and scanner artifacts for each
>participants, [2] setting the image origin to the anterior
>commissure;[3] Segmentation
>of tissues classes; [4] Normalization using DARTEL; [5] Modulation of
>different tissue segments; [6] Smoothing. Statistical Parametric Mapping
>(SPM8) was used for data processing and statistical analyses.
>
>
>Because head size vary, so this normalisation (proportional scaling) should
>reduce the residual variance from the model. Using proportional scaling to
>total grey matter will show differences where a region contains a
>disproportionately large or small region of total grey matter. I think that
>proportional scaling is better approach that ANCOVA - I have found some
>description:
>
>*''Global are a compromise (fudge) that is needed to introduce some kind of
>multi-variateness into the analysis. People generally want to see a map of
>significant blobs, so a mass-univariate approach (SPM) is usually adopted.
>In reality, shape (and hence relative volumes) is really multivariate. The
>volume of one structure is related to the volume of another. Bigger brains
>may (on average) have bigger hippocampi. Brains with globally thicker grey
>matter are more likely to have grey matter that is thicker at a particular
>sulcus. Also, brains where the segmentation (for some reason) overestimates
>the amount of grey matter, are also likely to appear to have more grey at a
>particular region. Unfortunately, this multi-variateness is not a simple
>linear relationship. For example, bigger brains are likely to have
>proportionally more white matter than grey. For brains that are globally
>different, it becomes very difficult to interpret exactly what the
>differences mean in a mass univariate way. Basically, it is not clear what
>to use as globals for any particular experiment. This will depend on what
>theory you have about your subjects.*
>
>*Shapes are multivariate. The volume of one structure is likely to be
>related to that of another. For example, smaller brains may have smaller
>putamen. If you include "globals" in the model by proportional scaling,
>then this could be accounted for. Similarly, the size of a putamen may, or
>may not, correlate with the size of surrounding structures. A modulated
>analysis attempts to correct the volumes for regional expansion/shrinkage
>during spatial normalisation.*
>
>*Proportional scaling converts volumes into values that are a proportion of
>total brain/GM volume (for me TIV). For example, you may be able to say
>that a region around some point contains 3% of the total GM volume. If you
>are interested in differences in such measures, then a proportional scaling
>model may be preferable. Alternatively, if you want to localise regions
>where the trends in GM volume differ from the total GM volume, then an
>ANCOVA model may be preferable. **''*
>
>
>
>
>
>Thanks a lot for your help in advance!
>
>All the best!
>
|