I suspect the first_flirt registration failed, at which point the
models will not work.
You should always check to make sure registrations worked.
Cheers,
Brian
P.S. I can have a look at a sample image if you'd like.
On 25 Sep 2007, at 13:42, Najmeh Khalili M. wrote:
> Sorry I had to be more clear.
>
> Yes, I did first_flirt both on the raw (as proposed by your
> website) and on the standardized image (as I tried to avoid the
> noise problem).
>
> However the difference between the HC volumes obtained from
> the standardized image was 1.8 times larger than the volume
> obtained from the native image.
>
> I think cropping the neck would fix the failing cases.
> If you like I can send you a sample data.
>
> Cheers
> Naj
>
>
>
> On Mon, 24 Sep 2007, Brian Patenaude wrote:
>
>> Hi,
>>
>> To clarify, "The average 1.8 scaling factor is not explained by the
>> transformation parameters.". Are you saying that the
>> transformation matrix
>> calculated from first_flirt differs in the scale component to that
>> applied
>> to get the standardized image by a factor of 1.8.
>>
>> If so, did you look at the output of first_flirt? Did it clearly
>> fail?
>> If so, the failure may be due to the neck or possibly bias field?
>> If the
>> neck is the problem, it may be corrected by masking the neck out.
>> Did you
>> try and apply first_flirt to the standardized image?
>>
>> Cheers,
>>
>> Brian
>>
>>
>>
>>> Hi Brian,
>>>
>>> Yes, I mean the amount of noise in the original data prior to
>>> any noise reduction.
>>>
>>> The other question is:
>>> result of FIRST on standardized data > result of FIRST on
>>> native.
>>>
>>> The average 1.8 scaling factor is not explained by the
>>> transformation parameters.
>>>
>>> Any ideas?
>>>
>>> Naj
>>>
>>>
>>>
>>>
>>> On Mon, 24 Sep 2007, Brian Patenaude wrote:
>>>
>>>> Hi,
>>>>
>>>> The only disadvantage is that you've interpolated the data
>>>> during the
>>>> transofrmation to MNI152 space. By senstive to the noise in the raw
>>>> image, are you meaning prior to non-uniformity correction?
>>>>
>>>> When applying FIRST to the pre-processed data, you should still use
>>>> first_flirt to re-align the image since it uses the sub-cortical
>>>> mask
>>>> (the segmentation will be in the original MNI152 space).
>>>>
>>>> The volumes are do take into account the boundary correction for
>>>> the
>>>> set threshold.
>>>>
>>>> Cheers,
>>>>
>>>> Brian
>>>>
>>>>
>>>>> Hi,
>>>>>
>>>>> I am interested in comparing the results of FIRST with manual
>>>>> segmentation.
>>>>>
>>>>> Is there an impediment to doing FIRST on data that has
>>>>> undergone: non-uniformity correction, neck cropping and 12
>>>>> param-affine registeration to MNI152 space? (The reason: 1)
>>>>> FISRT is sensitive to noise in the raw image; 2) our manual
>>>>> segmentation is done in standard space.)
>>>>>
>>>>> Also, would you be considering the boundary in calculation of
>>>>> the structure's volume?
>>>>>
>>>>> Many thanks
>>>>> Naj
>>>>
|