The images should be in approximate alignment before you begin. The closeness
depends on FOV and image quality, but about 3cm and 15 degrees is usually
enough. Use Check Reg before you begin to check out the initial alignment.
If this doesn't work well, then I don't have so many other good ideas. Maybe
skull stripping and bias correcting the MRI could help. If you use SPM5,
then you can bias correct via the segment button, and mask out non-brain
stuff via ImCalc (with the c1, c2 and c3 images).
Best regards,
-John
On Tuesday 13 June 2006 15:34, Remi Vieux wrote:
> Dear SPMers,
>
> We are trying to coregister SPECT images to MRI images with SPM2 and
> SPM99. For each patient, we have a spect and a MRI image, both in
> Analyze file format.
> With SPM99 coregistration works pretty accurately, but with SPM2, it is
> not correct at all. I'd think it could happen if it was not for most of
> the images of our set, so we are probably doing something wrong!
>
> The images we are using are SPECT and MRI images that are already in the
> same orientation. Coregistration with SPM2 ends up with the SPECT image
> beeing "upside-down", while coregistration should have translate the
> image and rotate by only a few degrees.
>
> Looking at this result we are wondering if we "feed" the algorithm with
> good images: does SPECT and MRI have to be in the same orientation
> before we coregister, or not, or it does not matter?
>
> Is there no "maximum rotation value" in the algorithm, for example if
> the best match gives a rotation of more than 45 degree along x, y or z
> then it's probably wrong anyway, or is it possible to set it up through
> a parameter value?
>
> This is all we could think of, if you have any suggestions!
>
> Best Regards,
> Remi.
|