Here are a few additional comments ...
The small field of view also means that the registration is likely to be less
than ideal. What works best is likely to be dependant upon your data, so I
would not like to prescribe any best approach for your data without empirical
evidence.
The limited field of view of your EPI data means that a good match can not be
obtained for the whole brain if the warps are estimated by matching an EPI
image to the EPI template. Because of this, the warps need to be
extrapolated - which is not especially sucessful.
Another contributing factor is that there is missing data in the resliced
fMRI. If spatial normalisation parameters are estimated by matching one of
these resliced images to the EPI template, then you will experience problems
because these regions have a value of zero (and so looks like the top and
bottom of the head have been sliced off). If you created a mean during the
reslicing, then using this to estimate the warps may do a better job. This
is because the mean should not have regions of missing data.
Best regards,
-John
|