Hi,
You could try adding some upweighting to the ventricle
region (say values of 100) and see if that helped.
We are working on some different automatic registration
methods in-house to deal with precisely this problem
but they are not ready yet.
I suggest that you try some upweighting and more
downweighting of the signal loss areas and see if that
can fix the problem at the moment.
All the best,
Mark
On 27 May 2011, at 19:14, Zhongming Liu wrote:
> Dear Mark,
>
> I used sigloss to create inweight from fieldmap and te, but it did not
> help (as much as I hope).
> The misalignment is hard to judge by using "slice" or simply looking at
> the cortex. It becomes very obvious when looking at the ventricles. If I
> judged by how well the ventricles are aligned, I found not running flirt
> but just using scanner coordinates gave me best alignment (I have to hope
> the subject did not move).
> Any additional thought or Can you help look into it?
>
> Best,
>
> Zhongming
>
>> Dear Zhongming,
>>
>> The unwarping run with FEAT uses a weighting image calculated
>> from the fieldmap. The image is called EF_UD_fmap_sigloss.nii.gz
>> in the EPI space. You can add this to your flirt call by putting it in
>> using the -inweight option. If it does not do enough you could try
>> making a more stringent version by applying a threshold to have
>> more areas set to zero. This can all be done after a standard FEAT
>> registration is run. If you do it in the *.feat/reg/ directory then
>> once you have a good registration, just replace the
>> example_func2highres.mat
>> file and run "updatefeatreg" in this feat directory. That will then
>> combine this registration with all the others, regenerating appropriate
>> image for the Feat report and making everything consistent for
>> running higher level FEATs afterwards.
>>
>> All the best,
>> Mark
>>
>>
>>
>> On 26 May 2011, at 15:54, Zhongming Liu wrote:
>>
>>> All,
>>>
>>> Using flirt, I am having somewhat unsatisfying alignment between
>>> whole-brain EPI and T1w images. After alignment, It appears that the EPI
>>> volume is always "pulled down" by a few millimeters. It happens to
>>> almost
>>> all my data. I guess it is because of the signal dropout.
>>>
>>> Here is the command I typically use.
>>>
>>> flirt -in epi_dewarp_bet -ref T1w_bet -omat epi2t1.mat -out
>>> epi_dewarp_bet_aligned
>>>
>>> The EPI images have been corrected for distortion before running flirt.
>>>
>>> I heard of a solution to mask the images or weighted the voxels when
>>> computing the cost function. But not sure how to do it. Any thought or
>>> solution is gratefully appreciated?
>>>
>>> Zhongming Liu
>>> NIH
>>>
>>
>
|