Hi,
I'd try mutual information and see if that is any better.
If they are more or less aligned at the beginning then
that's fine. You can check this by creating a copy of
the ASL data in the same space as the T1, but without
any transformation applied, by doing:
flirt -in ASL -ref T1 -applyxfm -out ASL_resamp
and then overlay ASL_resamp and T1 in FSLView.
All the best,
Mark
On 4 May 2009, at 20:22, John Kuster wrote:
> Hi Mark!
>
> Thanks for your questions! I am not sure of the cost function I'm
> using,
> I think it is default? I am starting nearly aligned as far as I can
> tell, this is just because we take the ASL and T1 scans very close
> together, and as such hope that the images should be fairly closely
> aligned to start with, barring movement. However, does this mean
> that in
> FSL the images will stay closely aligned? I cannot load the ASL and
> mprage together because of the dimension mismatch.
>
> Is there a specific cost function that you would recommend? In
> 'advanced options' it says 'correlation ratio'.
>
> Thank you!
>
> John Kuster
>
>
>> Hi,
>>
>> What cost function are you using?
>> Have you tried different ones?
>> Also, do you start nearly aligned or in
>> quite different orientations?
>>
>> All the best,
>> Mark
>>
>>
>> On 1 May 2009, at 16:50, John Kuster wrote:
>>
>>> Hello Hedok,
>>>
>>> Our slices are about 5mm thick, 3.5x3.5 voxes, for 16 slices,
>>> partial
>>> field, with no gap, but I have to check on the gap, it may be 1mm
>>> because of constraints.
>>>
>>> Flirt basically worked well for most of our subjects, but failed
>>> for a
>>> few. Do you think the resolution differences combined with the
>>> partial
>>> field may be putting flirt on the threshold of failing? I have
>>> tried
>>> cleaning up the ASL scans by cropping the background using bet with
>>> f=.2. This helped for a few, but not for the worst ones. Then, I
>>> tried masking the original image and the result was similar with the
>>> same few failing.
>>>
>>> I am basically at a loss as to how I can get flirt to work a little
>>> better.
>>>
>>> thanks for your input!
>>>
>>> Jake
>>>
>>>
>>>> Dear John,
>>>>
>>>> Just curious about your ASL(arterial spin labeling) image
>>>> parameters.
>>>>
>>>> Do you acquire them with very thick slice(~7mm) with gap ~1mm?.
>>>> Is this partial or full brain coverage?
>>>>
>>>> I don't have much experience in ASL, but I remember it was quite
>>>> difficult to even coregister anatomical with perfusion weighted
>>>> image.
>>>> Let me know how it goes.
>>>>
>>>> Hedok
>>>>
>>>> Jesper Andersson wrote:
>>>>> Dear John,
>>>>>
>>>>> just to add to Matt's previous comment.
>>>>>
>>>>>> I tried to run the following command,
>>>>>>
>>>>>> Command:
>>>>>> fnirt --ref=swap_CTRL02_T1.nii.gz --in=swap_CTRL02_ASL.nii.gz
>>>>>> --aff=CTRL02_ASL2T1_auto1.mat --cout=CTRL02_ASL2T1_auto2.mat
>>>>>
>>>>> From the names of your --ref and your --in file I am guessing you
>>>>> are
>>>>> trying to register a subjects ASL scan to the same subjects
>>>>> T1-weighted structural scan. Is that correct?
>>>>>
>>>>> If so there are two issues:
>>>>> 1. Since it is the same subject you really only need 6 dof, i.e.
>>>>> all
>>>>> you need to do is to make sure the images are in the same
>>>>> "position".
>>>>> 2. The contrast in an ASL image is very different from that of a
>>>>> T1-weighted image, something that fnirt would struggle with.
>>>>>
>>>>> Both these issues suggest that what you really need is flirt, and
>>>>> that
>>>>> will run just fine on a 32bit machine.
>>>>>
>>>>> Good luck Jesper
>>>>
>>>>
>>>>
>>>
>>
>>
>>
>
|