The DARTEL toolbox has much more flexibility to warp structures than the old
approach. Because of the extra flexibility, the Jacobians should be more
variable (but still > 0). Therefore, if the volume of a structure (eg a
sulcus) doubles, then it is not so surprising. Values greater than one are a
result of the grey matter maps (with values between 0 and 1) being multiplied
by Jacobian determinants that are greater than one.
In registration, there is a tradeoff between matching the images and keeping
the warps smooth. I'm still not exactly certain what the optimal balance is,
but it is possible that slightly more regularisation (more emphasis on
smoothness) could be used for the final iteration.
Best regards,
-John
> Thanks for the previous code, it did exactly what I wanted in fewer steps.
>
> However, I've started looking at individual participants and noticed that
> there is a much larger range of "probabilities" than with the traditional
> VBM. One participant has a maximum "probability" of about 2. Is this
> feasible? Would one expect a higher range? And finally, how does one
> interpret values over 1? Should this effect of being greater than 1 wash
> out across subjects?
>
> Thanks.
|