Hi,
Fine asymmetry, as Joe said, is generally preserved under affine transforms.
However, any gross asymmetries can be reduced by either 9 or 12 dof.
The reason is that rotations + independent scalings can act together to
generate a skew. It does depend on what order you apply the scalings
with respect to the rotations and what orientations your images are in.
In Flirt the affine matrix is give by (rotation * skew * scale).
This means that the input coordinate is scaled first, then skewed, then
rotated (multiplication goes right to left for the matrices).
So, if your original image is beautifully aligned with the axes then
the scaling won't affect left/right asymmetries. If the order was the
opposite (with scaling last) then this would preserve symmetries if
the reference image was aligned (which is true for the standard brain).
However, I'm afraid that's not what is done.
So, the bottom line is that either 9 or 12 dof will reduce gross asymmetry.
And Joe is right in that it is much better to measure things like this on
the original images as then you know nothing has effected it.
All the best,
Mark
Clare Mackay wrote:
>Hi,
>
>My two-penneth is that you might want to consider changing the number
>DOF in the registration. Reducing from 12 to 9 misses out the 'skew'
>step, which I think is the only thing in an affine that can really
>interfere with asymmetry.
>
>Cheers,
>Clare
>
>
>
>>>>[log in to unmask] 07/22/04 3:40 PM >>>
>>>>
>>>>
>Joe,
>
>Thanks for the answer, you confirmed my fears.
>
>Because I am mesuring Heschl's and planum temporale/parietal, it would
>seem better for me to do these measurements on the raw image data after
>correct for pitch, roll, and yaw. Do I have this correct?
>
>If so, then what are rebuttals to claims of lack of generalizability
>because of problems with inter-individual comparisons, and a lack of
>common stereotaxic space?
>
>Jason
>
>________________________________
>
>From: FSL - FMRIB's Software Library on behalf of Joe Devlin
>Sent: Thu 7/22/2004 10:30 AM
>To: [log in to unmask]
>Subject: Re: [FSL] registration question
>
>
>
>Hi Jason,
>
>No, structural asymmetries present in the original images are maintained
>in
>the transformed brain, assuming that they don't dominate the image. For
>instance, left and right Heschl's gyrus differ on the rostral-caudal
>axis
>in normal brains and this difference is maintained when transforming the
>brain into MNI space because the linear transformation can not alter
>relative positions within the image. If the asymmetry was enormous, on
>the
>other hand, then it would be reduced by trying to match a normal
>template. For instance, large lateralised strokes and hydrocephalus
>typically wreak havoc with registration.
>
>Hope this helps.
>Joe
>
>
|