Dear Gilberto,
We have FLAIR templates that maybe can help you to do Mark's first
suggestion. There are currently two versions:
- GG-FLAIR-462: This was built with images from 462 subjects. Mean age:
48.2y, std dev: 13.2y, range: 26-85y, 286 females.
- GG-FLAIR-181: This was built with a subset of 181 of the above
subjects. Only those with no or very few hyperintensities were selected.
Mean age: 39.9y, std dev: 9.3y, range: 26-76y, 102 females.
You can download these files from:
http://www.glahngroup.org/Members/anderson/flair-templates
These templates were constructed by first corregistering each subject's
FLAIR to their corresponding high-resolution T1-weighted image, then the
T1 was non-linearly aligned to the MNI space using SPM5's "unified
segmentation", then the warps were applied to the (bias-corrected) FLAIR
and sampled with resolution of 1mm. Some global intensity scaling was
applied to ensure that all subjects were in the same intensity range
before averaging. The process is described in this poster:
http://www.glahngroup.org/Members/anderson/publications/HBM2009_flair_poster.pdf
The images used for these templates are not all independent
observations, as the subjects are members of a number families that are
participating of a genetic study. Any feature observed on these images
are specific for the population that they have been drawn from and not
necessarily apply to other subjects or populations. For the purpose that
you described, however, I believe that these images should be good enough.
Notice, though, that we don't use these FLAIR templates as targets for
image registration, but the T1-weighted instead. FLAIR is very sensitive
to white matter lesions, which often appear as hyperintense areas. If
you have these hyperintensities in the brains of your subjects, they
have potential to cause undesired distortions and poor registrations.
Inspect each image after registration to make sure they are acceptable
before overlaying ROIs to do the hippocampal segmentation.
Hope this helps!
All the best,
Anderson
On 05/06/2011 07:29 AM, Mark Jenkinson wrote:
> Dear Gilberto,
>
> Oh, that's surprising - the vast majority of studies have a T1-weighted
> image, or one very like it (e.g. MPRAGE, etc.).
>
> If you really don't have one then there are two alternatives that you
> could try. One is an atlas-based approach where you do non-linear
> registration of your image to an atlas and then transfer an atlas
> segmentation to your image. You can use the atlases in MNI space
> for this, although you'll have to find a FLAIR template in MNI space,
> or create an effective one by modifying the intensities appropriately
> for the standard MNI image (though this might be difficult).
>
> The other approach would be to try and create a T1-style image
> from your FLAIR image. This would require a good segmentation
> of the CSF and for you to add together a weighted version of the
> FLAIR and the segmented CSF to look more like a T1 image. You
> can then run FIRST on this created image and it will hopefully work.
> As above, this might be difficult - I don't really know as I'm not very
> familiar with processing FLAIR images. It only needs to be a very
> good intensity match near the structures of interest in order for this
> to work OK, as long as you can get a reasonable initial affine
> registration too.
>
> Sorry that there isn't an easy solution, but this is the first time I've
> encountered this problem and so we don't have an implemented
> solution for it.
>
> Good luck and all the best,
> Mark
>
>
>
> On 6 May 2011, at 10:29, Gilberto Sousa Alves wrote:
>
>> Thanks for your nice reply, Mark.
>>
>> Unfortunately, I donīt have any T1 image. Therefore it seems that there is no alternative way to rate hippocampus volume with Flair images, isnīt it?
>>
>> Cheers,
>>
>> Gilberto
>>
|
