In papers looking at (Jacobian-1)*GM, the aim is to identify
volumetric differences in GM. Consider waping image A to match image
B. Segmenting image B, and smoothing the GM, will give the regional
GM volumes of image B. If we assume that the alignment of image A to
B is exact (such that segmenting the warped version of A gives the
same results as segmenting B), then Jacobian scaling of the GM and
smoothing should give the regional GM volume of image A. The
difference in regional GM between A and B is therefore Jacobian*GM -
GM.
After the recent-ish papers about inverse consistency (or lack of)
we'd probably do things slightly differently, treating the data in a
more symmetric way.
Best regards,
-John
On 22 November 2012 15:50, Eugenio Abela <[log in to unmask]> wrote:
> Dear SPM experts,
>
> while reviewing the tensor-based morphometry literature, I've come across different approaches to preprocess the Jacobian determinants and grey matter (GM) images before statsitical analysis.
>
> Some authors calculate (Jacobian-1)*GM, others calculate log(Jacobian) and proceed to stats without GM multiplication I've no difficulty understandig the log, but haven't been able to derive the rationale for the first formula. Also, wouldn't GM multiplication also be desirable in the log-case?
>
> Thanks for any help in these basic questions
>
> Regards
>
> Eugenio
|