Hi,
We've investigated this now and have found out why this occurs.
It is due to the fact that the CSF distribution in some cases is being
estimated as having a very large variance, and hence this can
actually become larger than the other Gaussians (which have
much smaller variances) when the intensity is a long way away
from the means. This is a mathematical possibility in the Gaussian
mixture model that we did not know would happen in practice.
It occurs in the new and old versions of FAST. We will consider
trying to implement some constraints in the model to prevent this
behaviour in future, but for now I think your work-around is a
perfectly good idea so I would continue with that.
All the best,
Mark
On 6 Apr 2009, at 15:43, Wolfgang Weber-Fahr wrote:
> Hallo,
>
> I might have found a workaround for our problem. Just in case
> anybody else is experiencing the same problems. We just lowered
> the intensity of the 5% highest values in the 3D-datasets to the
> 95% cap. Now the segmentation looks good though I'm still
> not sure why it went wrong in the first place. If anybody has an
> idea......
>
> Cheers,
>
> Wolfgang
>
> Wolfgang Weber-Fahr schrieb:
>> Hi,
>>
>> I've uploaded a zip-file rat_segment.zip (4MB) with the reference
>> Number 650137.
>> It contains:
>> original.nii - The brain image before bias correction
>> spm_debias.nii - After spm debias
>> debias_seg.nii - The semgmented file from spm_debias.nii
>> containing my problem in the dentate gyrus
>> result.pdf - A one page spm check-reg output
>> documenting my problem.
>>
>> Thanks amd Cheers. I really appreciate any help or hint.
>>
>> Wolfgang
>>
>>
>> Mark Jenkinson wrote:
>>> Hi,
>>>
>>> Can you please upload your volumetric image to:
>>> http://www.fmrib.ox.ac.uk/cgi-bin/upload.cgi
>>> and send us the reference number?
>>> Hopefully we'll be able to sort it out easily.
>>>
>>> All the best,
>>> Mark
>>>
>>>
>>> On 29 Mar 2009, at 22:37, Wolfgang Weber-Fahr wrote:
>>>
>>>> Hallo,
>>>>
>>>> thanks a lot for the answer. We tried different parameters for
>>>> the FAST bias field correction, but the segmentation - especially
>>>> the seperation between grey matter and white matter - just was
>>>> not very good. When using the SPM bias field correction prior
>>>> the segmentation, the FAST segmentation looked quite promising,
>>>> only that some of the brightest spots (GM) are clasified as
>>>> CSF.... .I wonder about the algorithm which puts some of the
>>>> brightest voxels into the group with the lowest mean value. I
>>>> could attach or upload a little pdf (70kB) showing the problem.
>>>> Thanks for any help.
>>>>
>>>> Wolfgang
>>>>
>>>> Steve Smith schrieb:
>>>>> Hi - I would not generally recommend using bias field correction
>>>>> before running FAST; FAST explicitly models bias field as part
>>>>> of the processing.
>>>>>
>>>>> Does that improve things?
>>>>>
>>>>> Cheers.
>>>>>
>>>>>
>>>>> On 25 Mar 2009, at 17:41, Wolfgang Weber-Fahr wrote:
>>>>>
>>>>>> Hallo,
>>>>>>
>>>>>> I'm rather new to FSL and try to segment T1-weighted rat brains
>>>>>> using FAST.
>>>>>> (3D FISP TE/TR 4/8 ms 0.15*0.12*0.23 mm^3)
>>>>>>
>>>>>> After brain extraction (BET), some polishing and bias
>>>>>> correction (SPM) I used
>>>>>> fast -t 1 -n 3 -g --nopve ....
>>>>>> since the contrast is not really good I also tried "-t 3" (PD-
>>>>>> weighted).
>>>>>>
>>>>>> The results are pretty good considering the low contrast, but
>>>>>> in some parts of
>>>>>> the images - quite often in the hippocampus (dentate gyrus) -
>>>>>> where the image intensity is actually highest, areas of
>>>>>> pixels are strangely misclassified as CSF (lowest intensity).
>>>>>>
>>>>>> Can anybody give me a clue about what is going wrong?
>>>>>>
>>>>>> Thanks,
>>>>>>
>>>>>> Wolfgang
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>> --
>>>>>> ----------------------------------------------------------
>>>>>> Wolfgang Weber-Fahr, Dr.rer.nat.
>>>>>> Central Institute of Mental Health
>>>>>> Neuroimaging Department
>>>>>> J5
>>>>>> 68072 Mannheim
>>>>>> Germany
>>>>>>
>>>>>> email: [log in to unmask]
>>>>>> phone: ++49 621 1703 2961
>>>>>>
>>>>>
>>>>>
>>>>> ---------------------------------------------------------------------------
>>>>> Stephen M. Smith, Professor of Biomedical Engineering
>>>>> Associate Director, Oxford University FMRIB Centre
>>>>>
>>>>> FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK
>>>>> +44 (0) 1865 222726 (fax 222717)
>>>>> [log in to unmask] http://www.fmrib.ox.ac.uk/~steve
>>>>> ---------------------------------------------------------------------------
>>>>>
>>>>
>>>>
>>>> --
>>>> ----------------------------------------------------------
>>>> Wolfgang Weber-Fahr, Dr.rer.nat.
>>>> Central Institute of Mental Health
>>>> Neuroimaging Department
>>>> J5
>>>> 68072 Mannheim
>>>> Germany
>>>>
>>>> email: [log in to unmask]
>>>> phone: ++49 621 1703 2961
>>>>
>>>
>>
>>
>
>
> --
> ----------------------------------------------------------
> Wolfgang Weber-Fahr, Dr.rer.nat.
> Central Institute of Mental Health
> Neuroimaging Department
> J5
> 68072 Mannheim
> Germany
>
> email: [log in to unmask]
> phone: ++49 621 1703 2961
>
|