Dear Michiru Makuuchi,
> My idea is masking or removing white matter, CSF, or any part
> that is not expected to show BOLD signal change from EPI data
> before statistical computation. This will reduce the number of
> statistical comparisons
> and confer more sensitivity.
I think the basic idea (remove 'uninteresting' regions from the
statistical analysis) to decrease the volume searched is a good one.
Compare for 'small volume correction'.
As a start, if you remove CSF from your functional images, you will most
likely remove most of your BOLD effect (measured at 1.5 T). Another
point is the smoothing filter you use prior to your masking. If you
smooth too much, it may well be that activation foci end up outside your
anatomically informed mask because of the convoluted structure of the
brain.
If you want to use the correction for multiple comparisons based on
Gaussian Random Fields, you have to take into account that the p-value
does not only depend on your search volume, but also on its surface. In
other words, while you decrease your search volume by exact segmentation
of some convoluted structure you also increase its surface. In the end,
reducing the volume decreases the estimated p-value, but increasing the
surface increases it. I haven't really looked into this, but my first
impression (doing something along your lines) was that using a smoothed
mask reduces the search volume while keeping the surface term in check.
> used to calculate "a global mean" that do not reflect coritical
> activity induced by
> experimental tasks and seem more appropriate as an index of a base
> line drift of the signal.
Jesper Andersson might want to comment on this one...
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]
|