Hi Erlend,
Your best bet may be to see the "intensity normalization" section in
$FSLDIR/tcl/featlib.tcl, and then just trace back the input params based
on the specific FEAT params that you're using. Note also that intensity
normalization is applied BEFORE perfusion subtraction and temporal
filtering. So, if you're doing either of those (and you're likely doing
temporal filtering of some sort), then you can't expect
filtered_func_data to have a 4D median of 10000. (Also, as you'll see
in featlib.tcl it is the *median*, not mean that is used to calculate
the scale factor).
cheers,
-MH
On Wed, 2012-04-11 at 11:03 +0100, Erlend Hodneland wrote:
> Hi,
>
> I have a question regarding the grand mean normalization in FEAT and how it is performed. I have searched through the archives and I refer for instance to post 026180 where the topic of intensity normalization is discussed:
>
> "I have noticed that there are sometimes systematic differences between different BOLD
> runs (same subject) in the average intensity value of the mean_func image (averaged
> spatially across the entire brain), and would like to understand why. I thought the average
> intensity value was "grand mean scaled" by FSL software to 10,000; the values are near
> 10,000, but why are they not exactly 10,000? The grand mean scaling is only for non-
> background voxels, but I assume mean_func only includes non-background voxels? Does the
> standard deviation of the signal in the raw data affect the calculation of the scaled values?"
>
> In post 007632 I find
>
> "Feat uses ip to do the filtering. At a minimum, ip performs a
> grand mean scaling to 10000 (i.e. the entire 4D data is scaled such
> that the global 4D mean is consistent; this is necessary for later
> higher level stats)"
>
> I want to apply the grand mean scaling in FEAT which I guess is happening in this line (taken from the log)
>
> /usr/share/fsl/bin/fslmaths prefiltered_func_data_smooth -mul 9.89261464905 prefiltered_func_data_intnorm
>
> My question is; how is the value 9.89261464905 computed? I tried to follow the post above and scale the non-background voxels to 10000 but then I get another scaling factor. Also, when I scale the overall mean to 10000 I get another factor than the one that appears in the log.
>
> Best,
> Erlend Hodneland
|