| We have been very puzzled by some results of SPM statistical analysis, when we
| use co-registered MR images to place PET images into the SPM space. In
| particular, when AIR is used to co-register MR and PET, subsequent spm
analysis
| gives large, somewhat unrealistic/unexpected activations throughout the whole
| cortex. But when spm's co-registration module is used, smaller and more
| isolated activations are obtained--what we typically expect. What is more
| puzzling is that, by looking at the individual or even average images after
| co-registration and spatial normalization, no major difference between the
| AIR-registered and the spm-registered images is observed. We can see a
slightly
| better WM/GM contrast in the inter-subject average of AIR-registered images,
| which might imply a slightly better registration. Perhaps a grand mean
| normalization artifact? We would greatly appreciate feedback.
I'm not sure what is likely to be causing the different results. There
could be some kind of grand-mean effect, depending whether or not Roger
rescales images to their max/min before writing them. You can test this
theory by re-running the stats, but using proportional scaling so that
all the images are scaled to approximately the same intensity. A difference
could also arise because of our different treatments of certain fields in
the headers. SPM uses a scalefactor field, but I'm not sure if anything
like this is used with AIR.
Another possible source of the discrepancies could be due to interpolation
(sinc versus chirp-Z), but I think this is much less likely.
All the best,
-John
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|