Dear Zaman,
> I was keen to know how to compute the average percentage change of
> signal intensity (% signal change) in an fMRI motor cortex study.
>
> I recently came across a paper with used the formula:
>
> [{signal intensity during task-signal intensity during rest}/signal
> intensity during rest] * 100
>
> I was wondering if there was a more sophisticated approach in SPM.
This formula sounds fine to me. There are possibly more sophisticated
approaches around. I would like to use this opportunity to describe how
the %signal change is determined in SPM99 and what are the reasons for
our approach.
In SPM99, a global mean intensity (GMI) is computed for each image. The
procedure used by default defines the GMI as
m = mean(Y(Y > mean(Y)/8));
i.e. take the mean of an image Y, divide this value by 8, use it as a
threshold for the image and take the mean of all voxels above this
threshold. For some reasons, this computation has been found to give you
an estimate of the mean of all intracortical voxels in functional
images.
For fMRI, SPM (by default) scales each image within a session with
100/(mean of the GMIs of this session). For a single session, this does
not have any effect on the parameter estimation or inference, etc. It is
just a global scaling of the whole session data. For multiple sessions,
this grand mean session scaling has the advantegeous effect that each
session is scaled to the same grand mean of 100. Note that this is done
by default, i.e. if you click 'none' when asked for proportional
scaling. You can instead apply proportional scaling, which scales each
image by 100/(mean of its GMI).
In SPM99, the usual approach to determine the % signal change of a given
effect is to plot its fitted response. Because the global mean was
scaled to 100, the height of an effect is the % signal change with
respect to the global mean intensity of the scaled images.
There was some question recently on the helpline, whether this should be
the preferred estimate of % signal change. Instead of computing the %
signal change with respect to the estimated GMI, it was proposed to take
the % signal change to the mean of the voxel, where a given activation
was found. The main argument for this was that the mean in a voxel can
vary a lot over the brain. Therefore in one voxel a % change with
respect to its voxel mean can mean something different than the same
change, but in a voxel with a different mean.
The reason why this approach is not adopted in SPM99, is the following:
Given partial volume effects as they are found with the typically used
functional resolution (2 - 4 mm) and some spatial smoothing as applied
in imaging studies, it is rather difficult to determine the origin of an
underlying activation. Therefore an activation stands always for some
area in the brain, where this area is mainly determined by the
underlying anatomy, the measurement process, in particular partial
volume effects, and the smoothing kernel. One can think of all kinds of
spatial configurations, where these effects can render the link between
measured signal change and underlying voxel mean questionable. A simple
case would be e.g. two activations at voxels A and B, where A and B both
are in grey matter, but the neighbouring voxels are of different image
intensities. Given partial volume effects and some smoothing, the mean
intensities of A and B will be different at the point of making
inferences, although the signal change should be related to the same
grey matter value at A and B.
Stefan
--
Stefan Kiebel
Functional Imaging Laboratory
Wellcome Dept. of Cognitive Neurology
12 Queen Square
WC1N 3BG London, UK
Tel.: +44-(0)20-7833-7478
FAX : -7813-1420
email: [log in to unmask]
|