I have two questions regarding the weighting images within the
normalization of PET images in SPM5:
1) I have a PET scan which is somewhat hard to normalize
accurately, ie. the result is really distorted (I already tried
to use coregistration first or increase regularization, etc -
without success). So I thought to use a weighting image
containing 1s within the brain and 0s outside (similar to the
"skull stripping" approach in MRI). Interestingly, this
weighting image doesn't really improve the normalization
process. However, it does work when I first multiply the
weighting image with the actual scan and then normalize the scan
(which is then basically the brain only). I thought the two
approaches would be the same - could someone explain the
difference to me, please?
2) Is it possible/feasible to use a weighting image to focus on
specific parts of the brain, eg. brainstem? For instance, I have
a problem with (another) PET scan where the cortex is nicely
registered to the template but the brainstem is not. Radioligand
uptake is high for both, but of course brainstem is smaller than
cortex, thus, I guess the cost function will be mostly based on
the cortex - so should a weighting image help or is there a
better approach?
Spatial normalisation (via Normalise) is based on minimising the sum of
squares difference between the data (see earlier email). I'm not sure
what kind of radioligands you are using, but it may be the case that the
uptake of tracer (ie the pattern of signal intensities in your images)
is not well represented by the pattern of signal intensities in the PET
template (which were based on blood flow studies).
Skull-stripping of both image and template data can help, as it provides
user-input that helps define the outline of the brain. This is most
useful in situations where tracer uptake outside the brain is especially
different from uptake within the brain.
The difference between weighting ans skull stripping one of the images
is that the sum of squared difference is computed either by summing
either:
weight*(template - image)^2
or
(template - weight*image)^2
Best regards,
-John
--
John Ashburner <[log in to unmask]>
|