Hi,

> I have recently posted a question about img2talcoord (thanks for fixing
> that bug) and this is, sort of, a follow up on that.
> Let's say that I have an image of a brain that I want to use as a
> standard image for registration. And let's say that this image is
> 91x109x91 with resolution of -2x2x2. Furthermore, the origin (that
> corresponds to AC) is at voxel (46,64,37) and that is specified in the
> header. So basically, the dataset is the same at the mni brain (except
> for the datatype; header files are almost identical) and the brain
> itself is very similar to the mni brain, but smaller (it is based on T1
> images). If I used that brain as reference brain for registration, will
> the "talairach" cluster coordinates that Feat outputs be in the space
> that is defined by the parameters above (resolution, origin...)?

Yes.  The coordinates are determined by header information only, so
if it is 91x109x91 with -2x2x2 mm voxels and origin at (46,64,37) then
that's all you need.  The registration of the images obviously use the
image intensities and so that will be different, which is of course, what
you want it to do.  So, redefining a standard volume like this is fine.

> If the above is correct, and if I say further that this brain fits the
> Talairach atlas (it fits the orientation and extent of the Talairach
> atlas), then the cluster coordinates would be in talairach space (as
> opposed to mni space, when mni standard brain is used). Right?

Yes, if you believe your image is a good representation of the
Talairach brain.

> Well, I tried a little test. I have run the same analysis twice, the
> difference being only in the reference brain used. If I use the above
> assumption, the coordinates calculated with mni brain as reference
> (mni) and then corrected (using correction described by Matthew Brett)
> to fit talairach atlas (mni->tal), will be approximately the same as
> those calculated using the Talairach brain (tal). Of course, under the
> assumption that Brett's transformation maps mni to the "same" brain as
> the one I described above.

Yes, I agree.

> My assumption above has proved wrong. The difference between corrected
> mni coordinates and tal coordinates is much bigger than expected: x
> values are the closest - the biggest difference is under 2mm; y
> coordinates differ by as much as 10mm and z even 17mm (and that's with
> a sample of 38 points; these are the same points in the original space
> for both analysis, I doublechacked). That is much too big of a
> difference if mni and tal brains differ up to 10mm. I have used slicer
> to see how well the brains match (in their original space, no
> transformation between the two), and they are not that different.. (I
> have that picture and will try to attach it). Here are a few points:
>
> mni coordinates:
> 6.95            -92.4   19.1
> -9.38           -85.7   2.32
> 6.96            -91.3   8.14
> 16.8            -93.7   29.8
>
> mni -> tal corrected:
> 6.8805  -88.638 22.03
> -9.2862 -82.918 6.2865
> 6.8904  -88.076 11.906
> 16.632  -89.405 31.924
>
> tal coordinates:
> 6.6             -96             5.3
> -9.59   -86.8   -10.1
> 6.62    -93.3   -5.23
> 16.4    -99             15.5
>
> I even tried calculating the transformation between mni and tal brains
> - I wanted to see whether applying this transformation to mni
> coordinates would give me numbers close to the tal coordinates but no.
> When I applied that transformation to mni coordinates I got numbers
> close to the mni -> tal corrected values (Brett) but still far from tal
> coordinates.
>
> Any thoughts on this would be helpful... thanks!

Sounds like the registrations are not very similar.
I am surprised by the large differences in z.

Some questions that I would like to have answers for are:
Did you look at the difference in struct -> mni and struct -> tal
transforms?
How different was your new standard ("tal") image to the existing one?
What was the mni -> tal transformation?
Do they look very different with "slices tal mni"  ?
Also, what was your input image, and did it or the "tal" image have the
brain extracted?

However, even with these answers I may not be able to help remotely.
Could you arrange to send me your data (on a website or ftp - not posted
the list) and then I can see if the images are correct and track down the
difference.

Cheers,
    Mark