Dear Gwen and Jasper,
that's good news! (I didn't know what had happened that all my brain
volumes were so similar :-) Thank you for you answers.
So, I'll compute the determinant of the affine.mat matrix and multiply
the volumes I computed from the jacobian by that value. What does the
nitpicky solution involve? (Jasper, it would be great to have the full
jacobian right from fnirt!)
On Wed, Aug 5, 2009 at 8:45 PM, Jesper Andersson<[log in to unmask]> wrote:
> Dear Roberto and Gwen,
> it is indeed the case that the map you get with --jout only considers the
> non-linear part. So if you want to know the full geometric scaling you would
> need to incorporate the affine part yourself. It is a reasonable first
> approximation to just multiply the jacobians together, though if one wants
> to be very nitpicky it is slightly more involved.
> Internally in fnirt the "full" jacobian is calculated and used (this is what
> we constrain to be positive). If there is some interest in it I can always
> include an option to write it out in the next release.
> Good Luck Jesper
>> Jesper would need to confirm it, but the Jacobian obtained with FNIRT does
>> *not* take into account the affine (initial) part of the registration. So
>> you would need to get the scaling factor(s) using avscale on the
>> corresponding flirt matrix.
>> --- En date de : Mer 5.8.09, roberto toro <[log in to unmask]> a écrit :
>>> De: roberto toro <[log in to unmask]>
>>> Objet: [FSL] Jacobian in FNIRT
>>> À: [log in to unmask]
>>> Date: Mercredi 5 Août 2009, 16h56
>>> I have an ROI defined in MNI152 space, and I would like to
>>> measure its
>>> volume in native space using the jacobian volume produced
>>> by FNIRT
>>> with the --jout option. But does the jacobian include the
>>> scaling present in the affine matrix or I have to add it
>>> (multiply it
>>> in fact...)?
>>> If for example, the affine matrix changes the orientation
>>> but also the
>>> size of the input brain, do I have to add this scaling to
>>> the values
>>> of the jacobian at each voxel?
>>> thanks in advance!