Hi,
Neighbourhood can have a significant influence via the MRF (adjustable
via -H) and
so this could be causing what you see. The other thing might be the
bias field if
these voxels are in very different locations. Have you checked the
bias field output?
As for number of classes - it is still prominently displayed on the
GUI and
available as the first option on the command line (-n).
All the best,
Mark
On 24 Jan 2009, at 21:13, Oliver Trebbe wrote:
> Hi,
>
> i dont get, why a black and a white 'voxel' (low ad high intensity)
> are clustert with same probability...
> Maybe it is neighborhood connected but there is such 'high edge',
> that there cant be a connected area...
> After the newest Version of fast isnt it possbile to use it to
> segment other things instead of WM GM and CSF in a brain extracted
> Image?
> canīt i segment a Volume into 2 classes anymore?
>
>
> Greetings
>
> Oliver
>
>
> Mark Jenkinson schrieb:
>> Hi,
>>
>> Have you tried modifying the number of iterations (-I) the
>> spatial MRF (-H) or the bias field smoothing (-l) ?
>> These parameters can all have a strong effect (as can some
>> of the other parameters too, but I'd start with the ones above).
>>
>> It will help if you can identify what you think is going wrong.
>>
>> Does the bias field look OK? It shouldn't show visible anatomical
>> detail, and the restored image should have a nicer histogram.
>>
>> Are the results too smooth (decreasing -H might help) or maybe
>> not smooth enough (increase -H)?
>>
>> Also - have you done brain extraction with BET? This is essential
>> for getting good results.
>>
>> Finally, are you looking at the hard segmentations of the PVEs?
>> It is the PVEs that we recommend as the most useful outputs,
>> and these can look different than the hard segmentations around
>> the boundaries.
>>
>> All the best,
>> Mark
>>
>>
>>
>>
>>
>> On 24 Jan 2009, at 11:40, Oliver Trebbe wrote:
>>
>>> Hello ,
>>>
>>> after i tried my method with the new FAST version, but the results
>>> are bad!
>>> so i suggesting using the old FAST version. (dont know what
>>> exactly changed, but the stuff im segmenting is bad clustert, not
>>> like id wanted it or guessed it..)
>>>
>>> So far
>>>
>>> --
>>> Oliver Trebbe
>>> Department of Neurology
>>> University of Muenster
>>> 48129 Muenster, Germany
>>>
>>
>
>
> --
> Oliver Trebbe
> Department of Neurology
> University of Muenster
> 48129 Muenster, Germany
>
|