Hi,
Actually I'm not too surprised by this. We do not turn priors on
by default because we don't think they help most of the time.
If we did think they were better in general, then we'd have them
as the default. Sometimes they can help, when the images have
issues with SNR or bias field, but if the images are good quality
without artefacts then it is normally better without using priors.
We also advise using the partial volume estimate outputs (pves)
as these normally give the best segmentations.
You should also try using FIRST, as this is what we recommend
for segmenting subcortical structures.
All the best,
Mark
On 13 Aug 2009, at 10:22, Yannis Paloyelis wrote:
> Dear FSL users /Mark,
>
> Some further comments on segmentation with FAST: having followed the
> steps
> as outlined in previous post, I get segmented GM (in subject's native
> space): [1] using priors throughout (-a -P), as well as [2] using
> priors
> only to initialize parameter estimation (-a), or [3] without using
> priors at
> all.
>
> Looking at the binary images (_seg), I get best results for basal
> ganglia
> when I do not use priors [3], and they become worse as I move to [2]
> and [1]
> (in the latter virtually no putamen!) - wasn't it supposed to be the
> other
> way round?
>
> Indeed, when I look at the probability maps (_pro), basal ganglia
> become
> clearer as I move from [3] to [1: priors used throughout]; in [1] the
> probabilistic BG segmentation is brilliant.
>
> Any comments would be greatly appreciated :)
>
> Would anybody know what the weird _gm_stdspace.nii.gz images I get
> when
> using priors are? On two planes only part of the brain shows.
>
> many thanks!
>
> All the best,
> Yannis
>
> --
> Yannis Paloyelis, M.Sc.
> PhD Student
> Room C3.18 (P080)
> MRC Social Genetic and
> Developmental Psychiatry Centre
> Institute of Psychiatry,
> King's College London
> De Crespigny Park
> SE5 8AF London
> Tel. +44-20-7848-5403
> Fax +44-20-7848-0866
>
|