Hi,
If you register your images to standard space using our
non-linear registration (FNIRT) and then calculate the
inverse warp (INVWARP) then you can transfer any of
the standard space masks (e.g. one of the atlases - such
as one extracted from the HarvardOxford atlas) and apply
this to your FAST results to separate ventricular CSF from
sulcal CSF. This does depend on the registration working
well, but for the ventricles it usually does unless the images
are of very diseased patients.
Make sure that you crop your images before running either
FNIRT or INVWARP, as that has a huge impact on the
timing. You can crop using fslroi and make sure that you
remove any space from around the head in every direction
(just use FSLView to figure out the coordinates that bound
the head well).
To create a ventricle mask, you might try the following:
fslroi $FSLDIR/data/atlases/HarvardOxford/HarvardOxford-sub-
prob-2mm.nii.gz LVentricle 2 1
fslroi $FSLDIR/data/atlases/HarvardOxford/HarvardOxford-sub-
prob-2mm.nii.gz RVentricle 13 1
fslmaths LVentricle.nii.gz -add RVentricle.nii.gz -thr 0.1 -bin -
dilF VentricleMask
which creates the image VentricleMask.nii.gz which you can the
transform to your original image space using applywarp with
the output of invwarp and then mask the FAST output with the
result.
The thing to be careful about is whether it cleanly differentiates
between ventricle and non-ventricle CSF in the posterior horns,
as it gets tricky in this area. So be careful with this. If it is
including
too much then you can tighten the above mask by removing the
"-dilF" from the fslmaths line and/or increasing the threshold from
0.1 to a higher value.
Hopefully this will work though and give you a good segmentation.
Try it on a few images and carefully inspect the output to see if it
is acceptable or not.
All the best,
Mark
On 3 Nov 2009, at 11:03, Emil Malucelli wrote:
> Hi
> Thanks for your replay. If i well understand you suggest to segment
> 3DT1 using FAST but i don't understand how i can isolate the
> ventricles using a standard space mask.
>
> Emil
>
> ----- Original Message ----- From: "Mark Jenkinson" <[log in to unmask]
> >
> To: <[log in to unmask]>
> Sent: Tuesday, October 27, 2009 8:31 PM
> Subject: Re: [FSL] First and Ventricles
>
>
> Hi,
>
> Unfortunately FIRST does not perform very well on the ventricles
> and explicitly excludes the posterior portion. If you want an
> accurate
> segmentation of the CSF then it is better to use FAST. You can use
> a standard space mask to isolate the ventricles from the sulcal CSF.
>
> All the best,
> Mark
>
>
>
>
> On 27 Oct 2009, at 16:34, Emil Malucelli wrote:
>
>> Hi
>> I'm trying to segment my 3D vol T1 image using FIRST 1.2 in a group
>> of control subjects. I obtain good results for all the structures
>> except for the ventricles. I segment all the stuctures usin
>> run_first_all.
>> Does anyone have a suggestion?
>> Thanks
>>
>>
>>
>> Emil Malucelli
>> Dipartimento di Medicina Interna, dell'Invecchiamento e Malattie
>> Nefrologiche
>> Università di Bologna
>> Via Massarenti, 9
>> 40138 Bologna
>> Italy
>> Tel: +39 051 4296704
>> Fax: +39 051 303 962
>> e-mail: [log in to unmask]
>> http://www.cermiv.unibo.it/
>
|