Hi Colm,
I see... Your images are indeed incredibly noisy...
Looking briefly at your 4D image (the unsmoothed one would have been better though), volumes 13, 31, 54, 57, 80-83 amongst others are indeed of poor quality.
Few points (not mutually exclusive) that you might want to consider :
1) Using the -n option would definitely help a bit (though not solving the noisy white matter issue), so I would consider re-running the analysis with it.
2) You can try removing the above volumes from the dataset (painful and drastic but safe)
3) You can also try applying a median filter to all the native GM images to see if this helps removing the noise and the voxels identified wrongly as GM.
4) You can fiddle a bit with the GM mask (currently created from the GM mean thresholded at 0.01) by thresholding it higher to see if it excludes the problematic WM areas.
5) Finally, I'll see with MJ (Mark Jenkinson) if there's anyway to improve the output of FAST but I would not count too much on this as it's still doing a good job at identifying the GM correctly in the cortex and deep grey matter despite the noise.
Cheers,
Gwenaelle
--- En date de : Mer 21.10.09, Colm Connolly <[log in to unmask]> a écrit :
> De: Colm Connolly <[log in to unmask]>
> Objet: Re: [FSL] Re : [FSL] fsl vbm differences in white matter around ventricles
> À: [log in to unmask]
> Date: Mercredi 21 Octobre 2009, 15h37
> Hi Gwenaelle,
>
> Thanks for replying.
>
> On 21 Oct 2009, at 14:52, Gwenaëlle DOUAUD wrote:
>
> > Hi Colm,
> >
> > looking at your image, I think it's fine.
> > There must be some hypointensities in the white matter
> of your patients that explains the resuts in the white
> matter.
> > The results in the ventricles is quite likely the
> consequence of two things: first the interface
> ventricle/white matter can be sometimes identified as grey
> matter by the automated tissue segmentation and second,
> there might be some slight residual misalignment of the
> ventricles (presumably quite big in your patients?) which
> leads to this result. Looking at your native GM images in
> the patients would give you a clue about this.
>
>
> I have noticed that some of the MPRAGE images are noisy and
> that the segmented images reflect this: some white is
> classified as grey. My understanding is that FAST should be
> able to deal with some noise in the data but maybe there is
> more noise than the default parameters in the fsl scripts
> can handle. Is there a way to tune it to make it more robust
> in the face of noise in the T1s?
>
> Take for example this person:
> https://tcin.tchpc.tcd.ie/~colmconn/short.66128.anat_struc.nii.gz
> and the corresponding GM map https://tcin.tchpc.tcd.ie/~colmconn/short.66128.anat_struc_GM.nii.gz
>
> some of the white matter is classified as GM. Do you think
> is there a way of dealing with this?
>
> ( https://tcin.tchpc.tcd.ie/~colmconn/GM_mod_merg_s2.nii.gz
> is also available. Volumes 80-83 (among others) show regions
> which to my mind should be WM )
>
> One more thing, with respect to residual misalignment, I
> have run the vbm as follows:
> fslvbm_1_bet -N -B -f 0.3
> fslvbm_2_template -a
> fslvbm_3_proc
>
> might it be advisable to run the analysis with
> fslvbm_2_template -n? Might the nonlinear tuning mitigate
> misalignment assuming it's there?
>
>
> > There are still some significant and very well defined
> results in the precentral gyrus bilaterally for instance...
> >
>
> Sure, but there remains the problem of trying to explain
> the large blobs located around the ventricles, it could be
> problematic to get by reviewers, otherwise there are some
> really neat results in the dataset.
>
> Regards,
> --
> Dr Colm G. Connolly
> Institute of Neuroscience
> The Lloyd Building
> University of Dublin
> Trinity College, Dublin 2, Éire
> Fax: +353-1-671-3183
>
> Please note that electronic mail to, from or within the
> Trinity College Dublin, may be the
> subject of a request under the Freedom of Information Act.
>
>
>
>
|