I've taken a look at the images you sent, and come to the conclusion
that the effect is an artifact that arises from the way that the
globals are computed.
spm_global first computes the overall average intensity within the
entire FoV. After dividing it by eight, this provides an initial
threshold, which is used for separating brain from non-brain. The
second pass then computes the average of the image intensity that is
above this threshold. The effect that you see is not a result of head
motion, but is a result of the global computations amplifying some
statistical noise in the data.
What is happening in the case of your images is that the initial
threshold is chosen so that it is somewhere around the intensity of
the air in the background. A slight change in this threshold will
cause a different set of voxels to be included within the region that
is averaged in the second pass. See the attached histograms.pdf file,
in which the overall average is shown in green and the average/8 is
shown in red. Below, I've zoomed into the relevant part of the
histogram. You can see that if the red line moves either left or
right by a tiny amount, then there will be a different set of voxels
included in the second pass of the procedure. This happens because
image intensities are rounded (maybe up or down - I'm not sure) to the
closest integer.
Best regards,
-John
On 20 January 2012 14:17, Jan Gläscher <[log in to unmask]> wrote:
> Dear SPMers,
>
> does no-one have a comment or an explanation for the observed phenomenon
> below?
>
> Thanks a lot!
> Jan
>
>
>> Dear SPMers,
>>
>> after checking a few subjects in my latest data set, I found that the
>> global mean of an EPI time series (computed with spm_global) changes
>> significantly
>> after spatial preprocessing (i.e. even plain realignment). Attached is an
>> example: While slice timing changes the time series only very subtly, both
>> realignment and realign&unwarp change the pattern significantly (i.e. they
>> alleviate some of the spiking ripples, while introducing others at
>> different
>> time points.) Any comments or explanations for this are greatly
>> appreciated.
>>
>> Best wishes,
>> Jan
>
>
|