Hi,
From the question, it is worth having in mind that the registration
process involves two parts: the first is to make a series of guesses
about what transformations might be needed to improve the registration
- these take place in 3D. The second, is to assess each of the guesses
for whether it resulted in a better or worse registration. This is done
in many cases with a 2D histogram and some form of metric (mutual
information, least squared difference, etc) that gives an idea of how
well the histograms "match". If we find a transformation that leads to
a good match as described by Torben, we probably have a good
registration. Note tough, that the images are of different types, the
matching becomes more complicated (we may need for example that air
(dark in T2) and CSF (bright in T2) both match with areas that are dark
in T1). A good registration here won't result just in points near the
line of identity in the 2d histogram, but will likely have sharper
clusters off the line of identity than a bad registration. The
different metrics change the assumptions as to what in the 2D histogram
constitutes a good "match".
Paul S
> The intensities of the two images (which are both 3D) is what goes into
> the 2D histogram. An area will be bright in the 2D histogram if many
> voxels have similar voxelintensities in both images. If the images are of
> same modality, the bright areas will be around the line of identity when
> the images are in register.
>
> Best
> Torben
>
> Sendt fra min iPhone
>
>> Den 26/05/2014 kl. 16.54 skrev Eva Talve <[log in to unmask]>:
>>
>> Hello,
>>
>> Could anyone explain me how in coregister of a T1 and fMRI the
>> 3-dimensionality of the images is taken into account, as the joint
>> histogram is 2D.
>>
>> Thanks,
>> Eva Talve
>
|