On 21 June 2012 21:14, Christopher Bell <[log in to unmask]> wrote:
>
> Rolf,
>
> Thanks that is very useful. Could you expand on why the Neuroimage method
> can underestimate? Is it because the blur isn't big enough to cover all the
> CSF
> in cases of atrophy or just that the dura contracts in some cases?
The former. You will sometimes get brains that are so atrophied that
the blur doesn't cover the CSF-filled space between the brain surface
and the true intracranial/extracranial boundary. In such cases you
can, of course, tweak the sigma of the blur, at the cost of requiring
visual checking and giving up on full automation. If you increase the
sigma for all subjects, you will get overestimations of ICV on normal
ones.
>
> Also, I agree the trickiest piece is where to define the intra/extracranial
> boundry.
> Any suggestions on that? I would like to exclude cerebellum ideally, as some
> of our data does not have complete cerebellum.
There are two ways to approach this question. If you need an
anatomical definition of the intracranial space, it becomes a matter
of using or defining a segmentation protocol, to be used by an expert
for manual outlining (e.g. Eritaia et al., MRM, 2000). But the second
part of your question suggests that what you need is a normalization
factor that corrects for head size, to be used within just one study
or perhaps a small series of studies.
If your images do not consistently show the entire cerebellum, this
will seriously limit the accuracy with which you will be able to
estimate ICV, to the point that I would say there is no point. To
estimate a normalization factor appropriate for this situation, you
could use an MNI brain mask and reverse-map it to your targets
(Keihaninejad et al., Neuroimage, 2010), reduce those masks from the
inferior end to a defined thickness and take the volume of that.
Compared to a real ICV estimation, that would be fairly coarse,
though, so you'd have to report it as a limitation arising from the
lack of a complete field of view across the intracranial space.
You may also find Free et al. (AJNR, 1995) useful, a comparison of
different normalization factors.
Hope that helps
Rolf
>
> Chris
>
>
>
>
> On Thu, Jun 21, 2012 at 1:13 PM, Rolf A. Heckemann
> <[log in to unmask]> wrote:
>>
>> On 21 June 2012 17:40, Christopher Bell <[log in to unmask]>
>> wrote:
>> > What do people think about using FAST segmentations to get an estimate
>> > of
>> > ICV? As long as you
>> > get good segmentations wouldn't atrophy just result in an increase in
>> > CSF
>> > and a corresponding decrease in WM and/or GM?
>>
>> The problem is brain extraction. You could run FAST on an unextracted
>> head image -- this would take a long time, of course, and atypical
>> intensities outside the head could distort or invalidate the result.
>> You'd still have the problem of where to draw the boundary between
>> intra- and extracranial.
>>
>> In principle, you are right, though. If you have a brain label from
>> somewhere (GM + WM), this can be extended into a reasonable estimate
>> of ICV: apply a Gaussian blur with a sigma of 6mm, threshold at 27%,
>> use FAST to label CSF within the resulting "dilated" mask, add that
>> CSF to the brain label, fill any holes. This achieves a fairly
>> consistent ICV estimate across normals and Alzheimer subjects
>> (Heckemann et al., Neuroimage 2011, Figure 5). It can still
>> underestimate ICV if there is very severe atrophy.
>>
>> You would probably achieve the best possible result if you had a
>> library of manual ICV delineations on a set of MRIs that are
>> reasonably similar to your target image. Then you use a multi-atlas
>> label propagation and fusion approach to transform them into a target
>> ICV label. This is (in a nutshell) how "pincram" works, which is
>> currently ranked #2 behind the gold-standard labels on the
>> Segmentation Validation Engine (http://sve.loni.ucla.edu/archive/).
>>
>> Regards
>>
>> Rolf
>>
>>
>> >
>> > Chris Bell
>> >
>> > On Thu, Jun 21, 2012 at 1:13 AM, Mark Jenkinson <[log in to unmask]>
>> > wrote:
>> >>
>> >> Dear Gabor,
>> >>
>> >> Given the accuracy in calculating the ICV using SIENAX I think it is
>> >> fine
>> >> to use this value and scale it to give you ICV.
>> >> If you have a very good brain extraction then FAST can give you a good
>> >> TBV
>> >> (which is obviously not exactly the same as ICV, but for most purposes
>> >> can
>> >> be substituted fine). It is very sensitive to the quality of the brain
>> >> extraction though.
>> >>
>> >> All the best,
>> >> Mark
>> >>
>> >>
>> >> On 21 Jun 2012, at 06:38, Gabor Perlaki wrote:
>> >>
>> >> > I need the TIVs because I want to report them. I know that correction
>> >> > with the TIV and correction with the scaling factor are identical,
>> >> > however
>> >> > if I'd like to report TIVs I need them. I think the TBV from FAST is
>> >> > not the
>> >> > best for this purpose, because using TBV as correction for head size
>> >> > may
>> >> > miss a global brain atrophy. Does anyone know how to calculate the
>> >> > ICVs from
>> >> > scaling factors? Can I use the volume of 1847712 mm^3 as the ICV of
>> >> > MNI152
>> >> > for this purpose, or do I need to recalculate it somehow? Or, any
>> >> > other
>> >> > method?
>> >> >
>> >> > Thanks,
>> >> > Gabor
>> >> >
>> >> >
>> >> >
>> >> >> You need to control for each individual subject's total brain
>> >> >> volume.
>> >> >> It wouldnt make sense to regress out the volume of MNI template,
>> >> >> since the
>> >> >> value of the covariate is >identical across subjects and
>> >> >> contributes
>> >> >> equally to the dependent measure. FAST should output a TBV.
>> >> >> Good luck.
>> >> >> Christine
>> >> >
>> >
>> >
>>
>>
>>
>> --
>> Rolf A Heckemann, MD PhD
>> Médecin chercheur, Fondation Neurodis
>> CERMEP - Imagerie du Vivant
>> Hôpital Neurologique Pierre Wertheimer
>> 59 Boulevard Pinel
>> 69003 Lyon
>> France
>>
>> 2nd affiliation: Honorary Fellow, Imperial College London
>
>
--
Rolf A Heckemann, MD PhD
Médecin chercheur, Fondation Neurodis
CERMEP - Imagerie du Vivant
Hôpital Neurologique Pierre Wertheimer
59 Boulevard Pinel
69003 Lyon
France
2nd affiliation: Honorary Fellow, Imperial College London
|