Thanks Michael,
But I tried to manually reorient images where there were big jumps
between two images before running realignment and this did not help. It
seems as though realignment ignores the changes that I make. Running
the data through my preprocess batch (which is below) twice seemed to
help line up the images though, but this means reslicing the data twice.
A
On 03/07/2014 17:04, Michael Erb wrote:
> As long as the images are not resliced, they are still in the same
> voxel space and therefore the same transformation is necessary (which
> is estimated by realign). Therefore you will get always the same
> realignment parameters as long as the realignment works well. But if
> the starting condition for the estimation is wrong (big jump between
> two images), the estimation may run into a wrong local minimum. By
> changing the start conditions (manually reorienting the images), this
> can be improved.
>
> Hope this helps,
>
> Michael
>
> Am 03.07.2014 17:41, schrieb Anjanette P Harris:
>> Marko - thanks for your comments/ help with this.
>>
>> Manually reorienting the images does change the images - I can check reg
>> and see that they are roughly aligned, but then running realignment
>> seems to ignore the changes and they are not aligned anymore.
>>
>> One thing you said that worries me - Can I clarify, when I coreg my mean
>> functional to my structural, I do not select the other functionals as
>> "other" images. My preprocess batch for my rat fMRI data is :
>>
>> - realignment:est and reslice (mean only);
>> - cogreg: est (ref image the mean functional, and source image
>> structural, other images blank - because the mean image is in alignement
>> with - the other functionals and the structural is jiggled to match
>> this);
>> - segment coreg images ie the structural;
>> - normalise: write (segment params to realigned images, ie the
>> functionals);
>> - smooth;
>> - normalise params written to structural image.
>>
>> Is this sounding sensible? it is the above processing that I have to run
>> twice to get some data to line up...
>> thanks again,
>> A
>>
>>
>>
>> On 03/07/2014 12:29, Marko Wilke wrote:
>>> Hello,
>>>
>>> looks like I started this, so...
>>>
>>>> I am having exactly the same problem as you! I am scanning rats and
>>>> some
>>>> of them managed to rotate around slowly during the scan - when I
>>>> run my
>>>> preprocessing (realign, reslice only mean image, coreg mean to
>>>> structural and then normalise and segment to template) I find that the
>>>> functional scans are still rotated round by the end of the scan.
>>>
>>> I am still under the impression that, if you use spm to display your
>>> images, you should see a difference in the orientation after
>>> realignment (potentially play around with voxel and world space, but
>>> it should be visible). It may not be that obvious but one way to find
>>> out whether anything happened at all to the images is to check the
>>> timestamp of their last modification (that is one reason why I still
>>> like img/hdr combination, you can tell much easier :)
>>>
>>> Also, what I meant is when doing "coreg mean to structural", you have
>>> to specify all other EPI images from this run as "other" images. This
>>> is where I have repeatedly failed to do the obvious myself, hence my
>>> warning to check that this actually happened. Again, time of header
>>> modification is one way to be sure.
>>>
>>>> you, I tried to use reorient to put the functional scans roughly in
>>>> alignment , but when I ran realign etc again the motion curves were
>>>> identical as if I had not reoriented the images and they were all
>>>> still
>>>> rotated by the end.
>>>
>>> Not sure this is correct but theoretically, if you rotate all images
>>> in one run in an identical way, the difference between them, and what
>>> it needs to register them on top of each other, will not change. This
>>> would be one explanation, another could be that the realignment
>>> procedure uses the original orientation which may still be encoded in
>>> the header somewhere.
>>>
>>>> One suggestion I had was to use cogregister to put the functional in
>>>> alignment - but this did not work.
>>>
>>> You mean you ran coreg, estimate on all images individually, with
>>> image 1 as the reference? That should work, albeit it seems a bit
>>> cumbersome and may not be as good as simple realignment as it is
>>> mainly meant for between-modality matching.
>>>
>>>> The only thing that worked was to run my preprocessing twice. But is
>>>> this a good idea? I have no idea, I only know that now my scans
>>>> line up
>>>> nicely. Could someone tell me why running realign etc twice is not a
>>>> good idea?
>>>
>>> If you do not reslice, my only concern would be that it should not be
>>> necessary, and that I would rather like to know where the underlying
>>> problem is.
>>>
>>> Not sure this helps but I think I am going to leave it at that.
>>>
>>> Cheers,
>>> Marko
>>
>>
>
--
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.
|